module @module {
util.global private @__auto.token_embd.weight = #stream.parameter.named<"model"::"token_embd.weight"> : tensor<128256x4096xf16>
util.global private @__auto.blk.0.attn_norm.weight = #stream.parameter.named<"model"::"blk.0.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.0.attn_q.weight = #stream.parameter.named<"model"::"blk.0.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.0.attn_k.weight = #stream.parameter.named<"model"::"blk.0.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.0.attn_v.weight = #stream.parameter.named<"model"::"blk.0.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.0.attn_output.weight = #stream.parameter.named<"model"::"blk.0.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.0.ffn_norm.weight = #stream.parameter.named<"model"::"blk.0.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.0.ffn_gate.weight = #stream.parameter.named<"model"::"blk.0.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.0.ffn_up.weight = #stream.parameter.named<"model"::"blk.0.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.0.ffn_down.weight = #stream.parameter.named<"model"::"blk.0.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.1.attn_norm.weight = #stream.parameter.named<"model"::"blk.1.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.1.attn_q.weight = #stream.parameter.named<"model"::"blk.1.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.1.attn_k.weight = #stream.parameter.named<"model"::"blk.1.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.1.attn_v.weight = #stream.parameter.named<"model"::"blk.1.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.1.attn_output.weight = #stream.parameter.named<"model"::"blk.1.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.1.ffn_norm.weight = #stream.parameter.named<"model"::"blk.1.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.1.ffn_gate.weight = #stream.parameter.named<"model"::"blk.1.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.1.ffn_up.weight = #stream.parameter.named<"model"::"blk.1.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.1.ffn_down.weight = #stream.parameter.named<"model"::"blk.1.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.2.attn_norm.weight = #stream.parameter.named<"model"::"blk.2.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.2.attn_q.weight = #stream.parameter.named<"model"::"blk.2.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.2.attn_k.weight = #stream.parameter.named<"model"::"blk.2.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.2.attn_v.weight = #stream.parameter.named<"model"::"blk.2.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.2.attn_output.weight = #stream.parameter.named<"model"::"blk.2.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.2.ffn_norm.weight = #stream.parameter.named<"model"::"blk.2.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.2.ffn_gate.weight = #stream.parameter.named<"model"::"blk.2.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.2.ffn_up.weight = #stream.parameter.named<"model"::"blk.2.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.2.ffn_down.weight = #stream.parameter.named<"model"::"blk.2.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.3.attn_norm.weight = #stream.parameter.named<"model"::"blk.3.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.3.attn_q.weight = #stream.parameter.named<"model"::"blk.3.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.3.attn_k.weight = #stream.parameter.named<"model"::"blk.3.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.3.attn_v.weight = #stream.parameter.named<"model"::"blk.3.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.3.attn_output.weight = #stream.parameter.named<"model"::"blk.3.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.3.ffn_norm.weight = #stream.parameter.named<"model"::"blk.3.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.3.ffn_gate.weight = #stream.parameter.named<"model"::"blk.3.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.3.ffn_up.weight = #stream.parameter.named<"model"::"blk.3.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.3.ffn_down.weight = #stream.parameter.named<"model"::"blk.3.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.4.attn_norm.weight = #stream.parameter.named<"model"::"blk.4.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.4.attn_q.weight = #stream.parameter.named<"model"::"blk.4.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.4.attn_k.weight = #stream.parameter.named<"model"::"blk.4.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.4.attn_v.weight = #stream.parameter.named<"model"::"blk.4.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.4.attn_output.weight = #stream.parameter.named<"model"::"blk.4.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.4.ffn_norm.weight = #stream.parameter.named<"model"::"blk.4.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.4.ffn_gate.weight = #stream.parameter.named<"model"::"blk.4.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.4.ffn_up.weight = #stream.parameter.named<"model"::"blk.4.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.4.ffn_down.weight = #stream.parameter.named<"model"::"blk.4.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.5.attn_norm.weight = #stream.parameter.named<"model"::"blk.5.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.5.attn_q.weight = #stream.parameter.named<"model"::"blk.5.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.5.attn_k.weight = #stream.parameter.named<"model"::"blk.5.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.5.attn_v.weight = #stream.parameter.named<"model"::"blk.5.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.5.attn_output.weight = #stream.parameter.named<"model"::"blk.5.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.5.ffn_norm.weight = #stream.parameter.named<"model"::"blk.5.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.5.ffn_gate.weight = #stream.parameter.named<"model"::"blk.5.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.5.ffn_up.weight = #stream.parameter.named<"model"::"blk.5.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.5.ffn_down.weight = #stream.parameter.named<"model"::"blk.5.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.6.attn_norm.weight = #stream.parameter.named<"model"::"blk.6.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.6.attn_q.weight = #stream.parameter.named<"model"::"blk.6.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.6.attn_k.weight = #stream.parameter.named<"model"::"blk.6.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.6.attn_v.weight = #stream.parameter.named<"model"::"blk.6.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.6.attn_output.weight = #stream.parameter.named<"model"::"blk.6.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.6.ffn_norm.weight = #stream.parameter.named<"model"::"blk.6.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.6.ffn_gate.weight = #stream.parameter.named<"model"::"blk.6.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.6.ffn_up.weight = #stream.parameter.named<"model"::"blk.6.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.6.ffn_down.weight = #stream.parameter.named<"model"::"blk.6.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.7.attn_norm.weight = #stream.parameter.named<"model"::"blk.7.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.7.attn_q.weight = #stream.parameter.named<"model"::"blk.7.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.7.attn_k.weight = #stream.parameter.named<"model"::"blk.7.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.7.attn_v.weight = #stream.parameter.named<"model"::"blk.7.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.7.attn_output.weight = #stream.parameter.named<"model"::"blk.7.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.7.ffn_norm.weight = #stream.parameter.named<"model"::"blk.7.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.7.ffn_gate.weight = #stream.parameter.named<"model"::"blk.7.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.7.ffn_up.weight = #stream.parameter.named<"model"::"blk.7.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.7.ffn_down.weight = #stream.parameter.named<"model"::"blk.7.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.8.attn_norm.weight = #stream.parameter.named<"model"::"blk.8.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.8.attn_q.weight = #stream.parameter.named<"model"::"blk.8.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.8.attn_k.weight = #stream.parameter.named<"model"::"blk.8.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.8.attn_v.weight = #stream.parameter.named<"model"::"blk.8.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.8.attn_output.weight = #stream.parameter.named<"model"::"blk.8.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.8.ffn_norm.weight = #stream.parameter.named<"model"::"blk.8.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.8.ffn_gate.weight = #stream.parameter.named<"model"::"blk.8.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.8.ffn_up.weight = #stream.parameter.named<"model"::"blk.8.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.8.ffn_down.weight = #stream.parameter.named<"model"::"blk.8.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.9.attn_norm.weight = #stream.parameter.named<"model"::"blk.9.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.9.attn_q.weight = #stream.parameter.named<"model"::"blk.9.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.9.attn_k.weight = #stream.parameter.named<"model"::"blk.9.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.9.attn_v.weight = #stream.parameter.named<"model"::"blk.9.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.9.attn_output.weight = #stream.parameter.named<"model"::"blk.9.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.9.ffn_norm.weight = #stream.parameter.named<"model"::"blk.9.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.9.ffn_gate.weight = #stream.parameter.named<"model"::"blk.9.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.9.ffn_up.weight = #stream.parameter.named<"model"::"blk.9.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.9.ffn_down.weight = #stream.parameter.named<"model"::"blk.9.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.10.attn_norm.weight = #stream.parameter.named<"model"::"blk.10.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.10.attn_q.weight = #stream.parameter.named<"model"::"blk.10.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.10.attn_k.weight = #stream.parameter.named<"model"::"blk.10.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.10.attn_v.weight = #stream.parameter.named<"model"::"blk.10.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.10.attn_output.weight = #stream.parameter.named<"model"::"blk.10.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.10.ffn_norm.weight = #stream.parameter.named<"model"::"blk.10.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.10.ffn_gate.weight = #stream.parameter.named<"model"::"blk.10.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.10.ffn_up.weight = #stream.parameter.named<"model"::"blk.10.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.10.ffn_down.weight = #stream.parameter.named<"model"::"blk.10.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.11.attn_norm.weight = #stream.parameter.named<"model"::"blk.11.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.11.attn_q.weight = #stream.parameter.named<"model"::"blk.11.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.11.attn_k.weight = #stream.parameter.named<"model"::"blk.11.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.11.attn_v.weight = #stream.parameter.named<"model"::"blk.11.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.11.attn_output.weight = #stream.parameter.named<"model"::"blk.11.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.11.ffn_norm.weight = #stream.parameter.named<"model"::"blk.11.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.11.ffn_gate.weight = #stream.parameter.named<"model"::"blk.11.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.11.ffn_up.weight = #stream.parameter.named<"model"::"blk.11.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.11.ffn_down.weight = #stream.parameter.named<"model"::"blk.11.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.12.attn_norm.weight = #stream.parameter.named<"model"::"blk.12.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.12.attn_q.weight = #stream.parameter.named<"model"::"blk.12.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.12.attn_k.weight = #stream.parameter.named<"model"::"blk.12.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.12.attn_v.weight = #stream.parameter.named<"model"::"blk.12.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.12.attn_output.weight = #stream.parameter.named<"model"::"blk.12.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.12.ffn_norm.weight = #stream.parameter.named<"model"::"blk.12.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.12.ffn_gate.weight = #stream.parameter.named<"model"::"blk.12.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.12.ffn_up.weight = #stream.parameter.named<"model"::"blk.12.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.12.ffn_down.weight = #stream.parameter.named<"model"::"blk.12.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.13.attn_norm.weight = #stream.parameter.named<"model"::"blk.13.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.13.attn_q.weight = #stream.parameter.named<"model"::"blk.13.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.13.attn_k.weight = #stream.parameter.named<"model"::"blk.13.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.13.attn_v.weight = #stream.parameter.named<"model"::"blk.13.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.13.attn_output.weight = #stream.parameter.named<"model"::"blk.13.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.13.ffn_norm.weight = #stream.parameter.named<"model"::"blk.13.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.13.ffn_gate.weight = #stream.parameter.named<"model"::"blk.13.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.13.ffn_up.weight = #stream.parameter.named<"model"::"blk.13.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.13.ffn_down.weight = #stream.parameter.named<"model"::"blk.13.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.14.attn_norm.weight = #stream.parameter.named<"model"::"blk.14.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.14.attn_q.weight = #stream.parameter.named<"model"::"blk.14.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.14.attn_k.weight = #stream.parameter.named<"model"::"blk.14.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.14.attn_v.weight = #stream.parameter.named<"model"::"blk.14.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.14.attn_output.weight = #stream.parameter.named<"model"::"blk.14.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.14.ffn_norm.weight = #stream.parameter.named<"model"::"blk.14.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.14.ffn_gate.weight = #stream.parameter.named<"model"::"blk.14.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.14.ffn_up.weight = #stream.parameter.named<"model"::"blk.14.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.14.ffn_down.weight = #stream.parameter.named<"model"::"blk.14.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.15.attn_norm.weight = #stream.parameter.named<"model"::"blk.15.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.15.attn_q.weight = #stream.parameter.named<"model"::"blk.15.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.15.attn_k.weight = #stream.parameter.named<"model"::"blk.15.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.15.attn_v.weight = #stream.parameter.named<"model"::"blk.15.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.15.attn_output.weight = #stream.parameter.named<"model"::"blk.15.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.15.ffn_norm.weight = #stream.parameter.named<"model"::"blk.15.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.15.ffn_gate.weight = #stream.parameter.named<"model"::"blk.15.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.15.ffn_up.weight = #stream.parameter.named<"model"::"blk.15.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.15.ffn_down.weight = #stream.parameter.named<"model"::"blk.15.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.16.attn_norm.weight = #stream.parameter.named<"model"::"blk.16.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.16.attn_q.weight = #stream.parameter.named<"model"::"blk.16.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.16.attn_k.weight = #stream.parameter.named<"model"::"blk.16.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.16.attn_v.weight = #stream.parameter.named<"model"::"blk.16.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.16.attn_output.weight = #stream.parameter.named<"model"::"blk.16.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.16.ffn_norm.weight = #stream.parameter.named<"model"::"blk.16.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.16.ffn_gate.weight = #stream.parameter.named<"model"::"blk.16.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.16.ffn_up.weight = #stream.parameter.named<"model"::"blk.16.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.16.ffn_down.weight = #stream.parameter.named<"model"::"blk.16.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.17.attn_norm.weight = #stream.parameter.named<"model"::"blk.17.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.17.attn_q.weight = #stream.parameter.named<"model"::"blk.17.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.17.attn_k.weight = #stream.parameter.named<"model"::"blk.17.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.17.attn_v.weight = #stream.parameter.named<"model"::"blk.17.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.17.attn_output.weight = #stream.parameter.named<"model"::"blk.17.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.17.ffn_norm.weight = #stream.parameter.named<"model"::"blk.17.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.17.ffn_gate.weight = #stream.parameter.named<"model"::"blk.17.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.17.ffn_up.weight = #stream.parameter.named<"model"::"blk.17.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.17.ffn_down.weight = #stream.parameter.named<"model"::"blk.17.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.18.attn_norm.weight = #stream.parameter.named<"model"::"blk.18.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.18.attn_q.weight = #stream.parameter.named<"model"::"blk.18.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.18.attn_k.weight = #stream.parameter.named<"model"::"blk.18.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.18.attn_v.weight = #stream.parameter.named<"model"::"blk.18.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.18.attn_output.weight = #stream.parameter.named<"model"::"blk.18.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.18.ffn_norm.weight = #stream.parameter.named<"model"::"blk.18.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.18.ffn_gate.weight = #stream.parameter.named<"model"::"blk.18.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.18.ffn_up.weight = #stream.parameter.named<"model"::"blk.18.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.18.ffn_down.weight = #stream.parameter.named<"model"::"blk.18.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.19.attn_norm.weight = #stream.parameter.named<"model"::"blk.19.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.19.attn_q.weight = #stream.parameter.named<"model"::"blk.19.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.19.attn_k.weight = #stream.parameter.named<"model"::"blk.19.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.19.attn_v.weight = #stream.parameter.named<"model"::"blk.19.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.19.attn_output.weight = #stream.parameter.named<"model"::"blk.19.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.19.ffn_norm.weight = #stream.parameter.named<"model"::"blk.19.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.19.ffn_gate.weight = #stream.parameter.named<"model"::"blk.19.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.19.ffn_up.weight = #stream.parameter.named<"model"::"blk.19.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.19.ffn_down.weight = #stream.parameter.named<"model"::"blk.19.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.20.attn_norm.weight = #stream.parameter.named<"model"::"blk.20.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.20.attn_q.weight = #stream.parameter.named<"model"::"blk.20.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.20.attn_k.weight = #stream.parameter.named<"model"::"blk.20.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.20.attn_v.weight = #stream.parameter.named<"model"::"blk.20.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.20.attn_output.weight = #stream.parameter.named<"model"::"blk.20.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.20.ffn_norm.weight = #stream.parameter.named<"model"::"blk.20.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.20.ffn_gate.weight = #stream.parameter.named<"model"::"blk.20.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.20.ffn_up.weight = #stream.parameter.named<"model"::"blk.20.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.20.ffn_down.weight = #stream.parameter.named<"model"::"blk.20.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.21.attn_norm.weight = #stream.parameter.named<"model"::"blk.21.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.21.attn_q.weight = #stream.parameter.named<"model"::"blk.21.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.21.attn_k.weight = #stream.parameter.named<"model"::"blk.21.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.21.attn_v.weight = #stream.parameter.named<"model"::"blk.21.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.21.attn_output.weight = #stream.parameter.named<"model"::"blk.21.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.21.ffn_norm.weight = #stream.parameter.named<"model"::"blk.21.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.21.ffn_gate.weight = #stream.parameter.named<"model"::"blk.21.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.21.ffn_up.weight = #stream.parameter.named<"model"::"blk.21.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.21.ffn_down.weight = #stream.parameter.named<"model"::"blk.21.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.22.attn_norm.weight = #stream.parameter.named<"model"::"blk.22.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.22.attn_q.weight = #stream.parameter.named<"model"::"blk.22.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.22.attn_k.weight = #stream.parameter.named<"model"::"blk.22.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.22.attn_v.weight = #stream.parameter.named<"model"::"blk.22.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.22.attn_output.weight = #stream.parameter.named<"model"::"blk.22.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.22.ffn_norm.weight = #stream.parameter.named<"model"::"blk.22.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.22.ffn_gate.weight = #stream.parameter.named<"model"::"blk.22.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.22.ffn_up.weight = #stream.parameter.named<"model"::"blk.22.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.22.ffn_down.weight = #stream.parameter.named<"model"::"blk.22.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.23.attn_norm.weight = #stream.parameter.named<"model"::"blk.23.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.23.attn_q.weight = #stream.parameter.named<"model"::"blk.23.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.23.attn_k.weight = #stream.parameter.named<"model"::"blk.23.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.23.attn_v.weight = #stream.parameter.named<"model"::"blk.23.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.23.attn_output.weight = #stream.parameter.named<"model"::"blk.23.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.23.ffn_norm.weight = #stream.parameter.named<"model"::"blk.23.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.23.ffn_gate.weight = #stream.parameter.named<"model"::"blk.23.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.23.ffn_up.weight = #stream.parameter.named<"model"::"blk.23.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.23.ffn_down.weight = #stream.parameter.named<"model"::"blk.23.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.24.attn_norm.weight = #stream.parameter.named<"model"::"blk.24.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.24.attn_q.weight = #stream.parameter.named<"model"::"blk.24.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.24.attn_k.weight = #stream.parameter.named<"model"::"blk.24.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.24.attn_v.weight = #stream.parameter.named<"model"::"blk.24.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.24.attn_output.weight = #stream.parameter.named<"model"::"blk.24.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.24.ffn_norm.weight = #stream.parameter.named<"model"::"blk.24.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.24.ffn_gate.weight = #stream.parameter.named<"model"::"blk.24.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.24.ffn_up.weight = #stream.parameter.named<"model"::"blk.24.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.24.ffn_down.weight = #stream.parameter.named<"model"::"blk.24.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.25.attn_norm.weight = #stream.parameter.named<"model"::"blk.25.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.25.attn_q.weight = #stream.parameter.named<"model"::"blk.25.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.25.attn_k.weight = #stream.parameter.named<"model"::"blk.25.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.25.attn_v.weight = #stream.parameter.named<"model"::"blk.25.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.25.attn_output.weight = #stream.parameter.named<"model"::"blk.25.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.25.ffn_norm.weight = #stream.parameter.named<"model"::"blk.25.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.25.ffn_gate.weight = #stream.parameter.named<"model"::"blk.25.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.25.ffn_up.weight = #stream.parameter.named<"model"::"blk.25.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.25.ffn_down.weight = #stream.parameter.named<"model"::"blk.25.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.26.attn_norm.weight = #stream.parameter.named<"model"::"blk.26.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.26.attn_q.weight = #stream.parameter.named<"model"::"blk.26.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.26.attn_k.weight = #stream.parameter.named<"model"::"blk.26.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.26.attn_v.weight = #stream.parameter.named<"model"::"blk.26.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.26.attn_output.weight = #stream.parameter.named<"model"::"blk.26.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.26.ffn_norm.weight = #stream.parameter.named<"model"::"blk.26.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.26.ffn_gate.weight = #stream.parameter.named<"model"::"blk.26.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.26.ffn_up.weight = #stream.parameter.named<"model"::"blk.26.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.26.ffn_down.weight = #stream.parameter.named<"model"::"blk.26.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.27.attn_norm.weight = #stream.parameter.named<"model"::"blk.27.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.27.attn_q.weight = #stream.parameter.named<"model"::"blk.27.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.27.attn_k.weight = #stream.parameter.named<"model"::"blk.27.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.27.attn_v.weight = #stream.parameter.named<"model"::"blk.27.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.27.attn_output.weight = #stream.parameter.named<"model"::"blk.27.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.27.ffn_norm.weight = #stream.parameter.named<"model"::"blk.27.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.27.ffn_gate.weight = #stream.parameter.named<"model"::"blk.27.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.27.ffn_up.weight = #stream.parameter.named<"model"::"blk.27.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.27.ffn_down.weight = #stream.parameter.named<"model"::"blk.27.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.28.attn_norm.weight = #stream.parameter.named<"model"::"blk.28.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.28.attn_q.weight = #stream.parameter.named<"model"::"blk.28.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.28.attn_k.weight = #stream.parameter.named<"model"::"blk.28.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.28.attn_v.weight = #stream.parameter.named<"model"::"blk.28.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.28.attn_output.weight = #stream.parameter.named<"model"::"blk.28.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.28.ffn_norm.weight = #stream.parameter.named<"model"::"blk.28.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.28.ffn_gate.weight = #stream.parameter.named<"model"::"blk.28.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.28.ffn_up.weight = #stream.parameter.named<"model"::"blk.28.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.28.ffn_down.weight = #stream.parameter.named<"model"::"blk.28.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.29.attn_norm.weight = #stream.parameter.named<"model"::"blk.29.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.29.attn_q.weight = #stream.parameter.named<"model"::"blk.29.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.29.attn_k.weight = #stream.parameter.named<"model"::"blk.29.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.29.attn_v.weight = #stream.parameter.named<"model"::"blk.29.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.29.attn_output.weight = #stream.parameter.named<"model"::"blk.29.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.29.ffn_norm.weight = #stream.parameter.named<"model"::"blk.29.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.29.ffn_gate.weight = #stream.parameter.named<"model"::"blk.29.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.29.ffn_up.weight = #stream.parameter.named<"model"::"blk.29.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.29.ffn_down.weight = #stream.parameter.named<"model"::"blk.29.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.30.attn_norm.weight = #stream.parameter.named<"model"::"blk.30.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.30.attn_q.weight = #stream.parameter.named<"model"::"blk.30.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.30.attn_k.weight = #stream.parameter.named<"model"::"blk.30.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.30.attn_v.weight = #stream.parameter.named<"model"::"blk.30.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.30.attn_output.weight = #stream.parameter.named<"model"::"blk.30.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.30.ffn_norm.weight = #stream.parameter.named<"model"::"blk.30.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.30.ffn_gate.weight = #stream.parameter.named<"model"::"blk.30.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.30.ffn_up.weight = #stream.parameter.named<"model"::"blk.30.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.30.ffn_down.weight = #stream.parameter.named<"model"::"blk.30.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.blk.31.attn_norm.weight = #stream.parameter.named<"model"::"blk.31.attn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.31.attn_q.weight = #stream.parameter.named<"model"::"blk.31.attn_q.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.31.attn_k.weight = #stream.parameter.named<"model"::"blk.31.attn_k.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.31.attn_v.weight = #stream.parameter.named<"model"::"blk.31.attn_v.weight"> : tensor<1024x4096xf16>
util.global private @__auto.blk.31.attn_output.weight = #stream.parameter.named<"model"::"blk.31.attn_output.weight"> : tensor<4096x4096xf16>
util.global private @__auto.blk.31.ffn_norm.weight = #stream.parameter.named<"model"::"blk.31.ffn_norm.weight"> : tensor<4096xf32>
util.global private @__auto.blk.31.ffn_gate.weight = #stream.parameter.named<"model"::"blk.31.ffn_gate.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.31.ffn_up.weight = #stream.parameter.named<"model"::"blk.31.ffn_up.weight"> : tensor<14336x4096xf16>
util.global private @__auto.blk.31.ffn_down.weight = #stream.parameter.named<"model"::"blk.31.ffn_down.weight"> : tensor<4096x14336xf16>
util.global private @__auto.output_norm.weight = #stream.parameter.named<"model"::"output_norm.weight"> : tensor<4096xf32>
util.global private @__auto.output.weight = #stream.parameter.named<"model"::"output.weight"> : tensor<128256x4096xf16>
func.func @prefill_bs4(%arg0: !torch.vtensor<[4,?],si64>, %arg1: !torch.vtensor<[4],si64>, %arg2: !torch.vtensor<[4,?],si64>, %arg3: !torch.tensor<[?,2097152],f16>) -> !torch.vtensor<[4,?,128256],f16> attributes {torch.assume_strict_symbolic_shapes} {
%__auto.token_embd.weight = util.global.load @__auto.token_embd.weight : tensor<128256x4096xf16>
%0 = torch_c.from_builtin_tensor %__auto.token_embd.weight : tensor<128256x4096xf16> -> !torch.vtensor<[128256,4096],f16>
%__auto.blk.0.attn_norm.weight = util.global.load @__auto.blk.0.attn_norm.weight : tensor<4096xf32>
%1 = torch_c.from_builtin_tensor %__auto.blk.0.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.0.attn_q.weight = util.global.load @__auto.blk.0.attn_q.weight : tensor<4096x4096xf16>
%2 = torch_c.from_builtin_tensor %__auto.blk.0.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.0.attn_k.weight = util.global.load @__auto.blk.0.attn_k.weight : tensor<1024x4096xf16>
%3 = torch_c.from_builtin_tensor %__auto.blk.0.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.0.attn_v.weight = util.global.load @__auto.blk.0.attn_v.weight : tensor<1024x4096xf16>
%4 = torch_c.from_builtin_tensor %__auto.blk.0.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.0.attn_output.weight = util.global.load @__auto.blk.0.attn_output.weight : tensor<4096x4096xf16>
%5 = torch_c.from_builtin_tensor %__auto.blk.0.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.0.ffn_norm.weight = util.global.load @__auto.blk.0.ffn_norm.weight : tensor<4096xf32>
%6 = torch_c.from_builtin_tensor %__auto.blk.0.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.0.ffn_gate.weight = util.global.load @__auto.blk.0.ffn_gate.weight : tensor<14336x4096xf16>
%7 = torch_c.from_builtin_tensor %__auto.blk.0.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.0.ffn_up.weight = util.global.load @__auto.blk.0.ffn_up.weight : tensor<14336x4096xf16>
%8 = torch_c.from_builtin_tensor %__auto.blk.0.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.0.ffn_down.weight = util.global.load @__auto.blk.0.ffn_down.weight : tensor<4096x14336xf16>
%9 = torch_c.from_builtin_tensor %__auto.blk.0.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.1.attn_norm.weight = util.global.load @__auto.blk.1.attn_norm.weight : tensor<4096xf32>
%10 = torch_c.from_builtin_tensor %__auto.blk.1.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.1.attn_q.weight = util.global.load @__auto.blk.1.attn_q.weight : tensor<4096x4096xf16>
%11 = torch_c.from_builtin_tensor %__auto.blk.1.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.1.attn_k.weight = util.global.load @__auto.blk.1.attn_k.weight : tensor<1024x4096xf16>
%12 = torch_c.from_builtin_tensor %__auto.blk.1.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.1.attn_v.weight = util.global.load @__auto.blk.1.attn_v.weight : tensor<1024x4096xf16>
%13 = torch_c.from_builtin_tensor %__auto.blk.1.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.1.attn_output.weight = util.global.load @__auto.blk.1.attn_output.weight : tensor<4096x4096xf16>
%14 = torch_c.from_builtin_tensor %__auto.blk.1.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.1.ffn_norm.weight = util.global.load @__auto.blk.1.ffn_norm.weight : tensor<4096xf32>
%15 = torch_c.from_builtin_tensor %__auto.blk.1.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.1.ffn_gate.weight = util.global.load @__auto.blk.1.ffn_gate.weight : tensor<14336x4096xf16>
%16 = torch_c.from_builtin_tensor %__auto.blk.1.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.1.ffn_up.weight = util.global.load @__auto.blk.1.ffn_up.weight : tensor<14336x4096xf16>
%17 = torch_c.from_builtin_tensor %__auto.blk.1.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.1.ffn_down.weight = util.global.load @__auto.blk.1.ffn_down.weight : tensor<4096x14336xf16>
%18 = torch_c.from_builtin_tensor %__auto.blk.1.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.2.attn_norm.weight = util.global.load @__auto.blk.2.attn_norm.weight : tensor<4096xf32>
%19 = torch_c.from_builtin_tensor %__auto.blk.2.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.2.attn_q.weight = util.global.load @__auto.blk.2.attn_q.weight : tensor<4096x4096xf16>
%20 = torch_c.from_builtin_tensor %__auto.blk.2.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.2.attn_k.weight = util.global.load @__auto.blk.2.attn_k.weight : tensor<1024x4096xf16>
%21 = torch_c.from_builtin_tensor %__auto.blk.2.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.2.attn_v.weight = util.global.load @__auto.blk.2.attn_v.weight : tensor<1024x4096xf16>
%22 = torch_c.from_builtin_tensor %__auto.blk.2.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.2.attn_output.weight = util.global.load @__auto.blk.2.attn_output.weight : tensor<4096x4096xf16>
%23 = torch_c.from_builtin_tensor %__auto.blk.2.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.2.ffn_norm.weight = util.global.load @__auto.blk.2.ffn_norm.weight : tensor<4096xf32>
%24 = torch_c.from_builtin_tensor %__auto.blk.2.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.2.ffn_gate.weight = util.global.load @__auto.blk.2.ffn_gate.weight : tensor<14336x4096xf16>
%25 = torch_c.from_builtin_tensor %__auto.blk.2.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.2.ffn_up.weight = util.global.load @__auto.blk.2.ffn_up.weight : tensor<14336x4096xf16>
%26 = torch_c.from_builtin_tensor %__auto.blk.2.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.2.ffn_down.weight = util.global.load @__auto.blk.2.ffn_down.weight : tensor<4096x14336xf16>
%27 = torch_c.from_builtin_tensor %__auto.blk.2.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.3.attn_norm.weight = util.global.load @__auto.blk.3.attn_norm.weight : tensor<4096xf32>
%28 = torch_c.from_builtin_tensor %__auto.blk.3.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.3.attn_q.weight = util.global.load @__auto.blk.3.attn_q.weight : tensor<4096x4096xf16>
%29 = torch_c.from_builtin_tensor %__auto.blk.3.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.3.attn_k.weight = util.global.load @__auto.blk.3.attn_k.weight : tensor<1024x4096xf16>
%30 = torch_c.from_builtin_tensor %__auto.blk.3.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.3.attn_v.weight = util.global.load @__auto.blk.3.attn_v.weight : tensor<1024x4096xf16>
%31 = torch_c.from_builtin_tensor %__auto.blk.3.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.3.attn_output.weight = util.global.load @__auto.blk.3.attn_output.weight : tensor<4096x4096xf16>
%32 = torch_c.from_builtin_tensor %__auto.blk.3.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.3.ffn_norm.weight = util.global.load @__auto.blk.3.ffn_norm.weight : tensor<4096xf32>
%33 = torch_c.from_builtin_tensor %__auto.blk.3.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.3.ffn_gate.weight = util.global.load @__auto.blk.3.ffn_gate.weight : tensor<14336x4096xf16>
%34 = torch_c.from_builtin_tensor %__auto.blk.3.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.3.ffn_up.weight = util.global.load @__auto.blk.3.ffn_up.weight : tensor<14336x4096xf16>
%35 = torch_c.from_builtin_tensor %__auto.blk.3.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.3.ffn_down.weight = util.global.load @__auto.blk.3.ffn_down.weight : tensor<4096x14336xf16>
%36 = torch_c.from_builtin_tensor %__auto.blk.3.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.4.attn_norm.weight = util.global.load @__auto.blk.4.attn_norm.weight : tensor<4096xf32>
%37 = torch_c.from_builtin_tensor %__auto.blk.4.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.4.attn_q.weight = util.global.load @__auto.blk.4.attn_q.weight : tensor<4096x4096xf16>
%38 = torch_c.from_builtin_tensor %__auto.blk.4.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.4.attn_k.weight = util.global.load @__auto.blk.4.attn_k.weight : tensor<1024x4096xf16>
%39 = torch_c.from_builtin_tensor %__auto.blk.4.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.4.attn_v.weight = util.global.load @__auto.blk.4.attn_v.weight : tensor<1024x4096xf16>
%40 = torch_c.from_builtin_tensor %__auto.blk.4.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.4.attn_output.weight = util.global.load @__auto.blk.4.attn_output.weight : tensor<4096x4096xf16>
%41 = torch_c.from_builtin_tensor %__auto.blk.4.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.4.ffn_norm.weight = util.global.load @__auto.blk.4.ffn_norm.weight : tensor<4096xf32>
%42 = torch_c.from_builtin_tensor %__auto.blk.4.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.4.ffn_gate.weight = util.global.load @__auto.blk.4.ffn_gate.weight : tensor<14336x4096xf16>
%43 = torch_c.from_builtin_tensor %__auto.blk.4.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.4.ffn_up.weight = util.global.load @__auto.blk.4.ffn_up.weight : tensor<14336x4096xf16>
%44 = torch_c.from_builtin_tensor %__auto.blk.4.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.4.ffn_down.weight = util.global.load @__auto.blk.4.ffn_down.weight : tensor<4096x14336xf16>
%45 = torch_c.from_builtin_tensor %__auto.blk.4.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.5.attn_norm.weight = util.global.load @__auto.blk.5.attn_norm.weight : tensor<4096xf32>
%46 = torch_c.from_builtin_tensor %__auto.blk.5.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.5.attn_q.weight = util.global.load @__auto.blk.5.attn_q.weight : tensor<4096x4096xf16>
%47 = torch_c.from_builtin_tensor %__auto.blk.5.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.5.attn_k.weight = util.global.load @__auto.blk.5.attn_k.weight : tensor<1024x4096xf16>
%48 = torch_c.from_builtin_tensor %__auto.blk.5.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.5.attn_v.weight = util.global.load @__auto.blk.5.attn_v.weight : tensor<1024x4096xf16>
%49 = torch_c.from_builtin_tensor %__auto.blk.5.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.5.attn_output.weight = util.global.load @__auto.blk.5.attn_output.weight : tensor<4096x4096xf16>
%50 = torch_c.from_builtin_tensor %__auto.blk.5.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.5.ffn_norm.weight = util.global.load @__auto.blk.5.ffn_norm.weight : tensor<4096xf32>
%51 = torch_c.from_builtin_tensor %__auto.blk.5.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.5.ffn_gate.weight = util.global.load @__auto.blk.5.ffn_gate.weight : tensor<14336x4096xf16>
%52 = torch_c.from_builtin_tensor %__auto.blk.5.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.5.ffn_up.weight = util.global.load @__auto.blk.5.ffn_up.weight : tensor<14336x4096xf16>
%53 = torch_c.from_builtin_tensor %__auto.blk.5.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.5.ffn_down.weight = util.global.load @__auto.blk.5.ffn_down.weight : tensor<4096x14336xf16>
%54 = torch_c.from_builtin_tensor %__auto.blk.5.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.6.attn_norm.weight = util.global.load @__auto.blk.6.attn_norm.weight : tensor<4096xf32>
%55 = torch_c.from_builtin_tensor %__auto.blk.6.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.6.attn_q.weight = util.global.load @__auto.blk.6.attn_q.weight : tensor<4096x4096xf16>
%56 = torch_c.from_builtin_tensor %__auto.blk.6.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.6.attn_k.weight = util.global.load @__auto.blk.6.attn_k.weight : tensor<1024x4096xf16>
%57 = torch_c.from_builtin_tensor %__auto.blk.6.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.6.attn_v.weight = util.global.load @__auto.blk.6.attn_v.weight : tensor<1024x4096xf16>
%58 = torch_c.from_builtin_tensor %__auto.blk.6.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.6.attn_output.weight = util.global.load @__auto.blk.6.attn_output.weight : tensor<4096x4096xf16>
%59 = torch_c.from_builtin_tensor %__auto.blk.6.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.6.ffn_norm.weight = util.global.load @__auto.blk.6.ffn_norm.weight : tensor<4096xf32>
%60 = torch_c.from_builtin_tensor %__auto.blk.6.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.6.ffn_gate.weight = util.global.load @__auto.blk.6.ffn_gate.weight : tensor<14336x4096xf16>
%61 = torch_c.from_builtin_tensor %__auto.blk.6.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.6.ffn_up.weight = util.global.load @__auto.blk.6.ffn_up.weight : tensor<14336x4096xf16>
%62 = torch_c.from_builtin_tensor %__auto.blk.6.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.6.ffn_down.weight = util.global.load @__auto.blk.6.ffn_down.weight : tensor<4096x14336xf16>
%63 = torch_c.from_builtin_tensor %__auto.blk.6.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.7.attn_norm.weight = util.global.load @__auto.blk.7.attn_norm.weight : tensor<4096xf32>
%64 = torch_c.from_builtin_tensor %__auto.blk.7.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.7.attn_q.weight = util.global.load @__auto.blk.7.attn_q.weight : tensor<4096x4096xf16>
%65 = torch_c.from_builtin_tensor %__auto.blk.7.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.7.attn_k.weight = util.global.load @__auto.blk.7.attn_k.weight : tensor<1024x4096xf16>
%66 = torch_c.from_builtin_tensor %__auto.blk.7.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.7.attn_v.weight = util.global.load @__auto.blk.7.attn_v.weight : tensor<1024x4096xf16>
%67 = torch_c.from_builtin_tensor %__auto.blk.7.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.7.attn_output.weight = util.global.load @__auto.blk.7.attn_output.weight : tensor<4096x4096xf16>
%68 = torch_c.from_builtin_tensor %__auto.blk.7.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.7.ffn_norm.weight = util.global.load @__auto.blk.7.ffn_norm.weight : tensor<4096xf32>
%69 = torch_c.from_builtin_tensor %__auto.blk.7.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.7.ffn_gate.weight = util.global.load @__auto.blk.7.ffn_gate.weight : tensor<14336x4096xf16>
%70 = torch_c.from_builtin_tensor %__auto.blk.7.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.7.ffn_up.weight = util.global.load @__auto.blk.7.ffn_up.weight : tensor<14336x4096xf16>
%71 = torch_c.from_builtin_tensor %__auto.blk.7.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.7.ffn_down.weight = util.global.load @__auto.blk.7.ffn_down.weight : tensor<4096x14336xf16>
%72 = torch_c.from_builtin_tensor %__auto.blk.7.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.8.attn_norm.weight = util.global.load @__auto.blk.8.attn_norm.weight : tensor<4096xf32>
%73 = torch_c.from_builtin_tensor %__auto.blk.8.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.8.attn_q.weight = util.global.load @__auto.blk.8.attn_q.weight : tensor<4096x4096xf16>
%74 = torch_c.from_builtin_tensor %__auto.blk.8.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.8.attn_k.weight = util.global.load @__auto.blk.8.attn_k.weight : tensor<1024x4096xf16>
%75 = torch_c.from_builtin_tensor %__auto.blk.8.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.8.attn_v.weight = util.global.load @__auto.blk.8.attn_v.weight : tensor<1024x4096xf16>
%76 = torch_c.from_builtin_tensor %__auto.blk.8.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.8.attn_output.weight = util.global.load @__auto.blk.8.attn_output.weight : tensor<4096x4096xf16>
%77 = torch_c.from_builtin_tensor %__auto.blk.8.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.8.ffn_norm.weight = util.global.load @__auto.blk.8.ffn_norm.weight : tensor<4096xf32>
%78 = torch_c.from_builtin_tensor %__auto.blk.8.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.8.ffn_gate.weight = util.global.load @__auto.blk.8.ffn_gate.weight : tensor<14336x4096xf16>
%79 = torch_c.from_builtin_tensor %__auto.blk.8.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.8.ffn_up.weight = util.global.load @__auto.blk.8.ffn_up.weight : tensor<14336x4096xf16>
%80 = torch_c.from_builtin_tensor %__auto.blk.8.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.8.ffn_down.weight = util.global.load @__auto.blk.8.ffn_down.weight : tensor<4096x14336xf16>
%81 = torch_c.from_builtin_tensor %__auto.blk.8.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.9.attn_norm.weight = util.global.load @__auto.blk.9.attn_norm.weight : tensor<4096xf32>
%82 = torch_c.from_builtin_tensor %__auto.blk.9.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.9.attn_q.weight = util.global.load @__auto.blk.9.attn_q.weight : tensor<4096x4096xf16>
%83 = torch_c.from_builtin_tensor %__auto.blk.9.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.9.attn_k.weight = util.global.load @__auto.blk.9.attn_k.weight : tensor<1024x4096xf16>
%84 = torch_c.from_builtin_tensor %__auto.blk.9.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.9.attn_v.weight = util.global.load @__auto.blk.9.attn_v.weight : tensor<1024x4096xf16>
%85 = torch_c.from_builtin_tensor %__auto.blk.9.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.9.attn_output.weight = util.global.load @__auto.blk.9.attn_output.weight : tensor<4096x4096xf16>
%86 = torch_c.from_builtin_tensor %__auto.blk.9.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.9.ffn_norm.weight = util.global.load @__auto.blk.9.ffn_norm.weight : tensor<4096xf32>
%87 = torch_c.from_builtin_tensor %__auto.blk.9.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.9.ffn_gate.weight = util.global.load @__auto.blk.9.ffn_gate.weight : tensor<14336x4096xf16>
%88 = torch_c.from_builtin_tensor %__auto.blk.9.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.9.ffn_up.weight = util.global.load @__auto.blk.9.ffn_up.weight : tensor<14336x4096xf16>
%89 = torch_c.from_builtin_tensor %__auto.blk.9.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.9.ffn_down.weight = util.global.load @__auto.blk.9.ffn_down.weight : tensor<4096x14336xf16>
%90 = torch_c.from_builtin_tensor %__auto.blk.9.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.10.attn_norm.weight = util.global.load @__auto.blk.10.attn_norm.weight : tensor<4096xf32>
%91 = torch_c.from_builtin_tensor %__auto.blk.10.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.10.attn_q.weight = util.global.load @__auto.blk.10.attn_q.weight : tensor<4096x4096xf16>
%92 = torch_c.from_builtin_tensor %__auto.blk.10.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.10.attn_k.weight = util.global.load @__auto.blk.10.attn_k.weight : tensor<1024x4096xf16>
%93 = torch_c.from_builtin_tensor %__auto.blk.10.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.10.attn_v.weight = util.global.load @__auto.blk.10.attn_v.weight : tensor<1024x4096xf16>
%94 = torch_c.from_builtin_tensor %__auto.blk.10.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.10.attn_output.weight = util.global.load @__auto.blk.10.attn_output.weight : tensor<4096x4096xf16>
%95 = torch_c.from_builtin_tensor %__auto.blk.10.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.10.ffn_norm.weight = util.global.load @__auto.blk.10.ffn_norm.weight : tensor<4096xf32>
%96 = torch_c.from_builtin_tensor %__auto.blk.10.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.10.ffn_gate.weight = util.global.load @__auto.blk.10.ffn_gate.weight : tensor<14336x4096xf16>
%97 = torch_c.from_builtin_tensor %__auto.blk.10.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.10.ffn_up.weight = util.global.load @__auto.blk.10.ffn_up.weight : tensor<14336x4096xf16>
%98 = torch_c.from_builtin_tensor %__auto.blk.10.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.10.ffn_down.weight = util.global.load @__auto.blk.10.ffn_down.weight : tensor<4096x14336xf16>
%99 = torch_c.from_builtin_tensor %__auto.blk.10.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.11.attn_norm.weight = util.global.load @__auto.blk.11.attn_norm.weight : tensor<4096xf32>
%100 = torch_c.from_builtin_tensor %__auto.blk.11.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.11.attn_q.weight = util.global.load @__auto.blk.11.attn_q.weight : tensor<4096x4096xf16>
%101 = torch_c.from_builtin_tensor %__auto.blk.11.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.11.attn_k.weight = util.global.load @__auto.blk.11.attn_k.weight : tensor<1024x4096xf16>
%102 = torch_c.from_builtin_tensor %__auto.blk.11.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.11.attn_v.weight = util.global.load @__auto.blk.11.attn_v.weight : tensor<1024x4096xf16>
%103 = torch_c.from_builtin_tensor %__auto.blk.11.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.11.attn_output.weight = util.global.load @__auto.blk.11.attn_output.weight : tensor<4096x4096xf16>
%104 = torch_c.from_builtin_tensor %__auto.blk.11.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.11.ffn_norm.weight = util.global.load @__auto.blk.11.ffn_norm.weight : tensor<4096xf32>
%105 = torch_c.from_builtin_tensor %__auto.blk.11.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.11.ffn_gate.weight = util.global.load @__auto.blk.11.ffn_gate.weight : tensor<14336x4096xf16>
%106 = torch_c.from_builtin_tensor %__auto.blk.11.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.11.ffn_up.weight = util.global.load @__auto.blk.11.ffn_up.weight : tensor<14336x4096xf16>
%107 = torch_c.from_builtin_tensor %__auto.blk.11.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.11.ffn_down.weight = util.global.load @__auto.blk.11.ffn_down.weight : tensor<4096x14336xf16>
%108 = torch_c.from_builtin_tensor %__auto.blk.11.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.12.attn_norm.weight = util.global.load @__auto.blk.12.attn_norm.weight : tensor<4096xf32>
%109 = torch_c.from_builtin_tensor %__auto.blk.12.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.12.attn_q.weight = util.global.load @__auto.blk.12.attn_q.weight : tensor<4096x4096xf16>
%110 = torch_c.from_builtin_tensor %__auto.blk.12.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.12.attn_k.weight = util.global.load @__auto.blk.12.attn_k.weight : tensor<1024x4096xf16>
%111 = torch_c.from_builtin_tensor %__auto.blk.12.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.12.attn_v.weight = util.global.load @__auto.blk.12.attn_v.weight : tensor<1024x4096xf16>
%112 = torch_c.from_builtin_tensor %__auto.blk.12.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.12.attn_output.weight = util.global.load @__auto.blk.12.attn_output.weight : tensor<4096x4096xf16>
%113 = torch_c.from_builtin_tensor %__auto.blk.12.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.12.ffn_norm.weight = util.global.load @__auto.blk.12.ffn_norm.weight : tensor<4096xf32>
%114 = torch_c.from_builtin_tensor %__auto.blk.12.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.12.ffn_gate.weight = util.global.load @__auto.blk.12.ffn_gate.weight : tensor<14336x4096xf16>
%115 = torch_c.from_builtin_tensor %__auto.blk.12.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.12.ffn_up.weight = util.global.load @__auto.blk.12.ffn_up.weight : tensor<14336x4096xf16>
%116 = torch_c.from_builtin_tensor %__auto.blk.12.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.12.ffn_down.weight = util.global.load @__auto.blk.12.ffn_down.weight : tensor<4096x14336xf16>
%117 = torch_c.from_builtin_tensor %__auto.blk.12.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.13.attn_norm.weight = util.global.load @__auto.blk.13.attn_norm.weight : tensor<4096xf32>
%118 = torch_c.from_builtin_tensor %__auto.blk.13.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.13.attn_q.weight = util.global.load @__auto.blk.13.attn_q.weight : tensor<4096x4096xf16>
%119 = torch_c.from_builtin_tensor %__auto.blk.13.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.13.attn_k.weight = util.global.load @__auto.blk.13.attn_k.weight : tensor<1024x4096xf16>
%120 = torch_c.from_builtin_tensor %__auto.blk.13.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.13.attn_v.weight = util.global.load @__auto.blk.13.attn_v.weight : tensor<1024x4096xf16>
%121 = torch_c.from_builtin_tensor %__auto.blk.13.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.13.attn_output.weight = util.global.load @__auto.blk.13.attn_output.weight : tensor<4096x4096xf16>
%122 = torch_c.from_builtin_tensor %__auto.blk.13.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.13.ffn_norm.weight = util.global.load @__auto.blk.13.ffn_norm.weight : tensor<4096xf32>
%123 = torch_c.from_builtin_tensor %__auto.blk.13.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.13.ffn_gate.weight = util.global.load @__auto.blk.13.ffn_gate.weight : tensor<14336x4096xf16>
%124 = torch_c.from_builtin_tensor %__auto.blk.13.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.13.ffn_up.weight = util.global.load @__auto.blk.13.ffn_up.weight : tensor<14336x4096xf16>
%125 = torch_c.from_builtin_tensor %__auto.blk.13.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.13.ffn_down.weight = util.global.load @__auto.blk.13.ffn_down.weight : tensor<4096x14336xf16>
%126 = torch_c.from_builtin_tensor %__auto.blk.13.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.14.attn_norm.weight = util.global.load @__auto.blk.14.attn_norm.weight : tensor<4096xf32>
%127 = torch_c.from_builtin_tensor %__auto.blk.14.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.14.attn_q.weight = util.global.load @__auto.blk.14.attn_q.weight : tensor<4096x4096xf16>
%128 = torch_c.from_builtin_tensor %__auto.blk.14.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.14.attn_k.weight = util.global.load @__auto.blk.14.attn_k.weight : tensor<1024x4096xf16>
%129 = torch_c.from_builtin_tensor %__auto.blk.14.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.14.attn_v.weight = util.global.load @__auto.blk.14.attn_v.weight : tensor<1024x4096xf16>
%130 = torch_c.from_builtin_tensor %__auto.blk.14.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.14.attn_output.weight = util.global.load @__auto.blk.14.attn_output.weight : tensor<4096x4096xf16>
%131 = torch_c.from_builtin_tensor %__auto.blk.14.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.14.ffn_norm.weight = util.global.load @__auto.blk.14.ffn_norm.weight : tensor<4096xf32>
%132 = torch_c.from_builtin_tensor %__auto.blk.14.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.14.ffn_gate.weight = util.global.load @__auto.blk.14.ffn_gate.weight : tensor<14336x4096xf16>
%133 = torch_c.from_builtin_tensor %__auto.blk.14.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.14.ffn_up.weight = util.global.load @__auto.blk.14.ffn_up.weight : tensor<14336x4096xf16>
%134 = torch_c.from_builtin_tensor %__auto.blk.14.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.14.ffn_down.weight = util.global.load @__auto.blk.14.ffn_down.weight : tensor<4096x14336xf16>
%135 = torch_c.from_builtin_tensor %__auto.blk.14.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.15.attn_norm.weight = util.global.load @__auto.blk.15.attn_norm.weight : tensor<4096xf32>
%136 = torch_c.from_builtin_tensor %__auto.blk.15.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.15.attn_q.weight = util.global.load @__auto.blk.15.attn_q.weight : tensor<4096x4096xf16>
%137 = torch_c.from_builtin_tensor %__auto.blk.15.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.15.attn_k.weight = util.global.load @__auto.blk.15.attn_k.weight : tensor<1024x4096xf16>
%138 = torch_c.from_builtin_tensor %__auto.blk.15.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.15.attn_v.weight = util.global.load @__auto.blk.15.attn_v.weight : tensor<1024x4096xf16>
%139 = torch_c.from_builtin_tensor %__auto.blk.15.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.15.attn_output.weight = util.global.load @__auto.blk.15.attn_output.weight : tensor<4096x4096xf16>
%140 = torch_c.from_builtin_tensor %__auto.blk.15.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.15.ffn_norm.weight = util.global.load @__auto.blk.15.ffn_norm.weight : tensor<4096xf32>
%141 = torch_c.from_builtin_tensor %__auto.blk.15.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.15.ffn_gate.weight = util.global.load @__auto.blk.15.ffn_gate.weight : tensor<14336x4096xf16>
%142 = torch_c.from_builtin_tensor %__auto.blk.15.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.15.ffn_up.weight = util.global.load @__auto.blk.15.ffn_up.weight : tensor<14336x4096xf16>
%143 = torch_c.from_builtin_tensor %__auto.blk.15.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.15.ffn_down.weight = util.global.load @__auto.blk.15.ffn_down.weight : tensor<4096x14336xf16>
%144 = torch_c.from_builtin_tensor %__auto.blk.15.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.16.attn_norm.weight = util.global.load @__auto.blk.16.attn_norm.weight : tensor<4096xf32>
%145 = torch_c.from_builtin_tensor %__auto.blk.16.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.16.attn_q.weight = util.global.load @__auto.blk.16.attn_q.weight : tensor<4096x4096xf16>
%146 = torch_c.from_builtin_tensor %__auto.blk.16.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.16.attn_k.weight = util.global.load @__auto.blk.16.attn_k.weight : tensor<1024x4096xf16>
%147 = torch_c.from_builtin_tensor %__auto.blk.16.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.16.attn_v.weight = util.global.load @__auto.blk.16.attn_v.weight : tensor<1024x4096xf16>
%148 = torch_c.from_builtin_tensor %__auto.blk.16.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.16.attn_output.weight = util.global.load @__auto.blk.16.attn_output.weight : tensor<4096x4096xf16>
%149 = torch_c.from_builtin_tensor %__auto.blk.16.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.16.ffn_norm.weight = util.global.load @__auto.blk.16.ffn_norm.weight : tensor<4096xf32>
%150 = torch_c.from_builtin_tensor %__auto.blk.16.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.16.ffn_gate.weight = util.global.load @__auto.blk.16.ffn_gate.weight : tensor<14336x4096xf16>
%151 = torch_c.from_builtin_tensor %__auto.blk.16.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.16.ffn_up.weight = util.global.load @__auto.blk.16.ffn_up.weight : tensor<14336x4096xf16>
%152 = torch_c.from_builtin_tensor %__auto.blk.16.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.16.ffn_down.weight = util.global.load @__auto.blk.16.ffn_down.weight : tensor<4096x14336xf16>
%153 = torch_c.from_builtin_tensor %__auto.blk.16.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.17.attn_norm.weight = util.global.load @__auto.blk.17.attn_norm.weight : tensor<4096xf32>
%154 = torch_c.from_builtin_tensor %__auto.blk.17.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.17.attn_q.weight = util.global.load @__auto.blk.17.attn_q.weight : tensor<4096x4096xf16>
%155 = torch_c.from_builtin_tensor %__auto.blk.17.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.17.attn_k.weight = util.global.load @__auto.blk.17.attn_k.weight : tensor<1024x4096xf16>
%156 = torch_c.from_builtin_tensor %__auto.blk.17.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.17.attn_v.weight = util.global.load @__auto.blk.17.attn_v.weight : tensor<1024x4096xf16>
%157 = torch_c.from_builtin_tensor %__auto.blk.17.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.17.attn_output.weight = util.global.load @__auto.blk.17.attn_output.weight : tensor<4096x4096xf16>
%158 = torch_c.from_builtin_tensor %__auto.blk.17.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.17.ffn_norm.weight = util.global.load @__auto.blk.17.ffn_norm.weight : tensor<4096xf32>
%159 = torch_c.from_builtin_tensor %__auto.blk.17.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.17.ffn_gate.weight = util.global.load @__auto.blk.17.ffn_gate.weight : tensor<14336x4096xf16>
%160 = torch_c.from_builtin_tensor %__auto.blk.17.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.17.ffn_up.weight = util.global.load @__auto.blk.17.ffn_up.weight : tensor<14336x4096xf16>
%161 = torch_c.from_builtin_tensor %__auto.blk.17.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.17.ffn_down.weight = util.global.load @__auto.blk.17.ffn_down.weight : tensor<4096x14336xf16>
%162 = torch_c.from_builtin_tensor %__auto.blk.17.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.18.attn_norm.weight = util.global.load @__auto.blk.18.attn_norm.weight : tensor<4096xf32>
%163 = torch_c.from_builtin_tensor %__auto.blk.18.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.18.attn_q.weight = util.global.load @__auto.blk.18.attn_q.weight : tensor<4096x4096xf16>
%164 = torch_c.from_builtin_tensor %__auto.blk.18.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.18.attn_k.weight = util.global.load @__auto.blk.18.attn_k.weight : tensor<1024x4096xf16>
%165 = torch_c.from_builtin_tensor %__auto.blk.18.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.18.attn_v.weight = util.global.load @__auto.blk.18.attn_v.weight : tensor<1024x4096xf16>
%166 = torch_c.from_builtin_tensor %__auto.blk.18.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.18.attn_output.weight = util.global.load @__auto.blk.18.attn_output.weight : tensor<4096x4096xf16>
%167 = torch_c.from_builtin_tensor %__auto.blk.18.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.18.ffn_norm.weight = util.global.load @__auto.blk.18.ffn_norm.weight : tensor<4096xf32>
%168 = torch_c.from_builtin_tensor %__auto.blk.18.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.18.ffn_gate.weight = util.global.load @__auto.blk.18.ffn_gate.weight : tensor<14336x4096xf16>
%169 = torch_c.from_builtin_tensor %__auto.blk.18.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.18.ffn_up.weight = util.global.load @__auto.blk.18.ffn_up.weight : tensor<14336x4096xf16>
%170 = torch_c.from_builtin_tensor %__auto.blk.18.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.18.ffn_down.weight = util.global.load @__auto.blk.18.ffn_down.weight : tensor<4096x14336xf16>
%171 = torch_c.from_builtin_tensor %__auto.blk.18.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.19.attn_norm.weight = util.global.load @__auto.blk.19.attn_norm.weight : tensor<4096xf32>
%172 = torch_c.from_builtin_tensor %__auto.blk.19.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.19.attn_q.weight = util.global.load @__auto.blk.19.attn_q.weight : tensor<4096x4096xf16>
%173 = torch_c.from_builtin_tensor %__auto.blk.19.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.19.attn_k.weight = util.global.load @__auto.blk.19.attn_k.weight : tensor<1024x4096xf16>
%174 = torch_c.from_builtin_tensor %__auto.blk.19.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.19.attn_v.weight = util.global.load @__auto.blk.19.attn_v.weight : tensor<1024x4096xf16>
%175 = torch_c.from_builtin_tensor %__auto.blk.19.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.19.attn_output.weight = util.global.load @__auto.blk.19.attn_output.weight : tensor<4096x4096xf16>
%176 = torch_c.from_builtin_tensor %__auto.blk.19.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.19.ffn_norm.weight = util.global.load @__auto.blk.19.ffn_norm.weight : tensor<4096xf32>
%177 = torch_c.from_builtin_tensor %__auto.blk.19.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.19.ffn_gate.weight = util.global.load @__auto.blk.19.ffn_gate.weight : tensor<14336x4096xf16>
%178 = torch_c.from_builtin_tensor %__auto.blk.19.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.19.ffn_up.weight = util.global.load @__auto.blk.19.ffn_up.weight : tensor<14336x4096xf16>
%179 = torch_c.from_builtin_tensor %__auto.blk.19.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.19.ffn_down.weight = util.global.load @__auto.blk.19.ffn_down.weight : tensor<4096x14336xf16>
%180 = torch_c.from_builtin_tensor %__auto.blk.19.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.20.attn_norm.weight = util.global.load @__auto.blk.20.attn_norm.weight : tensor<4096xf32>
%181 = torch_c.from_builtin_tensor %__auto.blk.20.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.20.attn_q.weight = util.global.load @__auto.blk.20.attn_q.weight : tensor<4096x4096xf16>
%182 = torch_c.from_builtin_tensor %__auto.blk.20.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.20.attn_k.weight = util.global.load @__auto.blk.20.attn_k.weight : tensor<1024x4096xf16>
%183 = torch_c.from_builtin_tensor %__auto.blk.20.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.20.attn_v.weight = util.global.load @__auto.blk.20.attn_v.weight : tensor<1024x4096xf16>
%184 = torch_c.from_builtin_tensor %__auto.blk.20.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.20.attn_output.weight = util.global.load @__auto.blk.20.attn_output.weight : tensor<4096x4096xf16>
%185 = torch_c.from_builtin_tensor %__auto.blk.20.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.20.ffn_norm.weight = util.global.load @__auto.blk.20.ffn_norm.weight : tensor<4096xf32>
%186 = torch_c.from_builtin_tensor %__auto.blk.20.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.20.ffn_gate.weight = util.global.load @__auto.blk.20.ffn_gate.weight : tensor<14336x4096xf16>
%187 = torch_c.from_builtin_tensor %__auto.blk.20.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.20.ffn_up.weight = util.global.load @__auto.blk.20.ffn_up.weight : tensor<14336x4096xf16>
%188 = torch_c.from_builtin_tensor %__auto.blk.20.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.20.ffn_down.weight = util.global.load @__auto.blk.20.ffn_down.weight : tensor<4096x14336xf16>
%189 = torch_c.from_builtin_tensor %__auto.blk.20.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.21.attn_norm.weight = util.global.load @__auto.blk.21.attn_norm.weight : tensor<4096xf32>
%190 = torch_c.from_builtin_tensor %__auto.blk.21.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.21.attn_q.weight = util.global.load @__auto.blk.21.attn_q.weight : tensor<4096x4096xf16>
%191 = torch_c.from_builtin_tensor %__auto.blk.21.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.21.attn_k.weight = util.global.load @__auto.blk.21.attn_k.weight : tensor<1024x4096xf16>
%192 = torch_c.from_builtin_tensor %__auto.blk.21.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.21.attn_v.weight = util.global.load @__auto.blk.21.attn_v.weight : tensor<1024x4096xf16>
%193 = torch_c.from_builtin_tensor %__auto.blk.21.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.21.attn_output.weight = util.global.load @__auto.blk.21.attn_output.weight : tensor<4096x4096xf16>
%194 = torch_c.from_builtin_tensor %__auto.blk.21.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.21.ffn_norm.weight = util.global.load @__auto.blk.21.ffn_norm.weight : tensor<4096xf32>
%195 = torch_c.from_builtin_tensor %__auto.blk.21.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.21.ffn_gate.weight = util.global.load @__auto.blk.21.ffn_gate.weight : tensor<14336x4096xf16>
%196 = torch_c.from_builtin_tensor %__auto.blk.21.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.21.ffn_up.weight = util.global.load @__auto.blk.21.ffn_up.weight : tensor<14336x4096xf16>
%197 = torch_c.from_builtin_tensor %__auto.blk.21.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.21.ffn_down.weight = util.global.load @__auto.blk.21.ffn_down.weight : tensor<4096x14336xf16>
%198 = torch_c.from_builtin_tensor %__auto.blk.21.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.22.attn_norm.weight = util.global.load @__auto.blk.22.attn_norm.weight : tensor<4096xf32>
%199 = torch_c.from_builtin_tensor %__auto.blk.22.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.22.attn_q.weight = util.global.load @__auto.blk.22.attn_q.weight : tensor<4096x4096xf16>
%200 = torch_c.from_builtin_tensor %__auto.blk.22.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.22.attn_k.weight = util.global.load @__auto.blk.22.attn_k.weight : tensor<1024x4096xf16>
%201 = torch_c.from_builtin_tensor %__auto.blk.22.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.22.attn_v.weight = util.global.load @__auto.blk.22.attn_v.weight : tensor<1024x4096xf16>
%202 = torch_c.from_builtin_tensor %__auto.blk.22.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.22.attn_output.weight = util.global.load @__auto.blk.22.attn_output.weight : tensor<4096x4096xf16>
%203 = torch_c.from_builtin_tensor %__auto.blk.22.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.22.ffn_norm.weight = util.global.load @__auto.blk.22.ffn_norm.weight : tensor<4096xf32>
%204 = torch_c.from_builtin_tensor %__auto.blk.22.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.22.ffn_gate.weight = util.global.load @__auto.blk.22.ffn_gate.weight : tensor<14336x4096xf16>
%205 = torch_c.from_builtin_tensor %__auto.blk.22.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.22.ffn_up.weight = util.global.load @__auto.blk.22.ffn_up.weight : tensor<14336x4096xf16>
%206 = torch_c.from_builtin_tensor %__auto.blk.22.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.22.ffn_down.weight = util.global.load @__auto.blk.22.ffn_down.weight : tensor<4096x14336xf16>
%207 = torch_c.from_builtin_tensor %__auto.blk.22.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.23.attn_norm.weight = util.global.load @__auto.blk.23.attn_norm.weight : tensor<4096xf32>
%208 = torch_c.from_builtin_tensor %__auto.blk.23.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.23.attn_q.weight = util.global.load @__auto.blk.23.attn_q.weight : tensor<4096x4096xf16>
%209 = torch_c.from_builtin_tensor %__auto.blk.23.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.23.attn_k.weight = util.global.load @__auto.blk.23.attn_k.weight : tensor<1024x4096xf16>
%210 = torch_c.from_builtin_tensor %__auto.blk.23.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.23.attn_v.weight = util.global.load @__auto.blk.23.attn_v.weight : tensor<1024x4096xf16>
%211 = torch_c.from_builtin_tensor %__auto.blk.23.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.23.attn_output.weight = util.global.load @__auto.blk.23.attn_output.weight : tensor<4096x4096xf16>
%212 = torch_c.from_builtin_tensor %__auto.blk.23.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.23.ffn_norm.weight = util.global.load @__auto.blk.23.ffn_norm.weight : tensor<4096xf32>
%213 = torch_c.from_builtin_tensor %__auto.blk.23.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.23.ffn_gate.weight = util.global.load @__auto.blk.23.ffn_gate.weight : tensor<14336x4096xf16>
%214 = torch_c.from_builtin_tensor %__auto.blk.23.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.23.ffn_up.weight = util.global.load @__auto.blk.23.ffn_up.weight : tensor<14336x4096xf16>
%215 = torch_c.from_builtin_tensor %__auto.blk.23.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.23.ffn_down.weight = util.global.load @__auto.blk.23.ffn_down.weight : tensor<4096x14336xf16>
%216 = torch_c.from_builtin_tensor %__auto.blk.23.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.24.attn_norm.weight = util.global.load @__auto.blk.24.attn_norm.weight : tensor<4096xf32>
%217 = torch_c.from_builtin_tensor %__auto.blk.24.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.24.attn_q.weight = util.global.load @__auto.blk.24.attn_q.weight : tensor<4096x4096xf16>
%218 = torch_c.from_builtin_tensor %__auto.blk.24.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.24.attn_k.weight = util.global.load @__auto.blk.24.attn_k.weight : tensor<1024x4096xf16>
%219 = torch_c.from_builtin_tensor %__auto.blk.24.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.24.attn_v.weight = util.global.load @__auto.blk.24.attn_v.weight : tensor<1024x4096xf16>
%220 = torch_c.from_builtin_tensor %__auto.blk.24.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.24.attn_output.weight = util.global.load @__auto.blk.24.attn_output.weight : tensor<4096x4096xf16>
%221 = torch_c.from_builtin_tensor %__auto.blk.24.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.24.ffn_norm.weight = util.global.load @__auto.blk.24.ffn_norm.weight : tensor<4096xf32>
%222 = torch_c.from_builtin_tensor %__auto.blk.24.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.24.ffn_gate.weight = util.global.load @__auto.blk.24.ffn_gate.weight : tensor<14336x4096xf16>
%223 = torch_c.from_builtin_tensor %__auto.blk.24.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.24.ffn_up.weight = util.global.load @__auto.blk.24.ffn_up.weight : tensor<14336x4096xf16>
%224 = torch_c.from_builtin_tensor %__auto.blk.24.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.24.ffn_down.weight = util.global.load @__auto.blk.24.ffn_down.weight : tensor<4096x14336xf16>
%225 = torch_c.from_builtin_tensor %__auto.blk.24.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.25.attn_norm.weight = util.global.load @__auto.blk.25.attn_norm.weight : tensor<4096xf32>
%226 = torch_c.from_builtin_tensor %__auto.blk.25.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.25.attn_q.weight = util.global.load @__auto.blk.25.attn_q.weight : tensor<4096x4096xf16>
%227 = torch_c.from_builtin_tensor %__auto.blk.25.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.25.attn_k.weight = util.global.load @__auto.blk.25.attn_k.weight : tensor<1024x4096xf16>
%228 = torch_c.from_builtin_tensor %__auto.blk.25.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.25.attn_v.weight = util.global.load @__auto.blk.25.attn_v.weight : tensor<1024x4096xf16>
%229 = torch_c.from_builtin_tensor %__auto.blk.25.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.25.attn_output.weight = util.global.load @__auto.blk.25.attn_output.weight : tensor<4096x4096xf16>
%230 = torch_c.from_builtin_tensor %__auto.blk.25.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.25.ffn_norm.weight = util.global.load @__auto.blk.25.ffn_norm.weight : tensor<4096xf32>
%231 = torch_c.from_builtin_tensor %__auto.blk.25.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.25.ffn_gate.weight = util.global.load @__auto.blk.25.ffn_gate.weight : tensor<14336x4096xf16>
%232 = torch_c.from_builtin_tensor %__auto.blk.25.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.25.ffn_up.weight = util.global.load @__auto.blk.25.ffn_up.weight : tensor<14336x4096xf16>
%233 = torch_c.from_builtin_tensor %__auto.blk.25.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.25.ffn_down.weight = util.global.load @__auto.blk.25.ffn_down.weight : tensor<4096x14336xf16>
%234 = torch_c.from_builtin_tensor %__auto.blk.25.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.26.attn_norm.weight = util.global.load @__auto.blk.26.attn_norm.weight : tensor<4096xf32>
%235 = torch_c.from_builtin_tensor %__auto.blk.26.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.26.attn_q.weight = util.global.load @__auto.blk.26.attn_q.weight : tensor<4096x4096xf16>
%236 = torch_c.from_builtin_tensor %__auto.blk.26.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.26.attn_k.weight = util.global.load @__auto.blk.26.attn_k.weight : tensor<1024x4096xf16>
%237 = torch_c.from_builtin_tensor %__auto.blk.26.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.26.attn_v.weight = util.global.load @__auto.blk.26.attn_v.weight : tensor<1024x4096xf16>
%238 = torch_c.from_builtin_tensor %__auto.blk.26.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.26.attn_output.weight = util.global.load @__auto.blk.26.attn_output.weight : tensor<4096x4096xf16>
%239 = torch_c.from_builtin_tensor %__auto.blk.26.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.26.ffn_norm.weight = util.global.load @__auto.blk.26.ffn_norm.weight : tensor<4096xf32>
%240 = torch_c.from_builtin_tensor %__auto.blk.26.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.26.ffn_gate.weight = util.global.load @__auto.blk.26.ffn_gate.weight : tensor<14336x4096xf16>
%241 = torch_c.from_builtin_tensor %__auto.blk.26.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.26.ffn_up.weight = util.global.load @__auto.blk.26.ffn_up.weight : tensor<14336x4096xf16>
%242 = torch_c.from_builtin_tensor %__auto.blk.26.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.26.ffn_down.weight = util.global.load @__auto.blk.26.ffn_down.weight : tensor<4096x14336xf16>
%243 = torch_c.from_builtin_tensor %__auto.blk.26.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.27.attn_norm.weight = util.global.load @__auto.blk.27.attn_norm.weight : tensor<4096xf32>
%244 = torch_c.from_builtin_tensor %__auto.blk.27.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.27.attn_q.weight = util.global.load @__auto.blk.27.attn_q.weight : tensor<4096x4096xf16>
%245 = torch_c.from_builtin_tensor %__auto.blk.27.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.27.attn_k.weight = util.global.load @__auto.blk.27.attn_k.weight : tensor<1024x4096xf16>
%246 = torch_c.from_builtin_tensor %__auto.blk.27.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.27.attn_v.weight = util.global.load @__auto.blk.27.attn_v.weight : tensor<1024x4096xf16>
%247 = torch_c.from_builtin_tensor %__auto.blk.27.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.27.attn_output.weight = util.global.load @__auto.blk.27.attn_output.weight : tensor<4096x4096xf16>
%248 = torch_c.from_builtin_tensor %__auto.blk.27.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.27.ffn_norm.weight = util.global.load @__auto.blk.27.ffn_norm.weight : tensor<4096xf32>
%249 = torch_c.from_builtin_tensor %__auto.blk.27.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.27.ffn_gate.weight = util.global.load @__auto.blk.27.ffn_gate.weight : tensor<14336x4096xf16>
%250 = torch_c.from_builtin_tensor %__auto.blk.27.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.27.ffn_up.weight = util.global.load @__auto.blk.27.ffn_up.weight : tensor<14336x4096xf16>
%251 = torch_c.from_builtin_tensor %__auto.blk.27.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.27.ffn_down.weight = util.global.load @__auto.blk.27.ffn_down.weight : tensor<4096x14336xf16>
%252 = torch_c.from_builtin_tensor %__auto.blk.27.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.28.attn_norm.weight = util.global.load @__auto.blk.28.attn_norm.weight : tensor<4096xf32>
%253 = torch_c.from_builtin_tensor %__auto.blk.28.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.28.attn_q.weight = util.global.load @__auto.blk.28.attn_q.weight : tensor<4096x4096xf16>
%254 = torch_c.from_builtin_tensor %__auto.blk.28.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.28.attn_k.weight = util.global.load @__auto.blk.28.attn_k.weight : tensor<1024x4096xf16>
%255 = torch_c.from_builtin_tensor %__auto.blk.28.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.28.attn_v.weight = util.global.load @__auto.blk.28.attn_v.weight : tensor<1024x4096xf16>
%256 = torch_c.from_builtin_tensor %__auto.blk.28.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.28.attn_output.weight = util.global.load @__auto.blk.28.attn_output.weight : tensor<4096x4096xf16>
%257 = torch_c.from_builtin_tensor %__auto.blk.28.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.28.ffn_norm.weight = util.global.load @__auto.blk.28.ffn_norm.weight : tensor<4096xf32>
%258 = torch_c.from_builtin_tensor %__auto.blk.28.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.28.ffn_gate.weight = util.global.load @__auto.blk.28.ffn_gate.weight : tensor<14336x4096xf16>
%259 = torch_c.from_builtin_tensor %__auto.blk.28.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.28.ffn_up.weight = util.global.load @__auto.blk.28.ffn_up.weight : tensor<14336x4096xf16>
%260 = torch_c.from_builtin_tensor %__auto.blk.28.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.28.ffn_down.weight = util.global.load @__auto.blk.28.ffn_down.weight : tensor<4096x14336xf16>
%261 = torch_c.from_builtin_tensor %__auto.blk.28.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.29.attn_norm.weight = util.global.load @__auto.blk.29.attn_norm.weight : tensor<4096xf32>
%262 = torch_c.from_builtin_tensor %__auto.blk.29.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.29.attn_q.weight = util.global.load @__auto.blk.29.attn_q.weight : tensor<4096x4096xf16>
%263 = torch_c.from_builtin_tensor %__auto.blk.29.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.29.attn_k.weight = util.global.load @__auto.blk.29.attn_k.weight : tensor<1024x4096xf16>
%264 = torch_c.from_builtin_tensor %__auto.blk.29.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.29.attn_v.weight = util.global.load @__auto.blk.29.attn_v.weight : tensor<1024x4096xf16>
%265 = torch_c.from_builtin_tensor %__auto.blk.29.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.29.attn_output.weight = util.global.load @__auto.blk.29.attn_output.weight : tensor<4096x4096xf16>
%266 = torch_c.from_builtin_tensor %__auto.blk.29.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.29.ffn_norm.weight = util.global.load @__auto.blk.29.ffn_norm.weight : tensor<4096xf32>
%267 = torch_c.from_builtin_tensor %__auto.blk.29.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.29.ffn_gate.weight = util.global.load @__auto.blk.29.ffn_gate.weight : tensor<14336x4096xf16>
%268 = torch_c.from_builtin_tensor %__auto.blk.29.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.29.ffn_up.weight = util.global.load @__auto.blk.29.ffn_up.weight : tensor<14336x4096xf16>
%269 = torch_c.from_builtin_tensor %__auto.blk.29.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.29.ffn_down.weight = util.global.load @__auto.blk.29.ffn_down.weight : tensor<4096x14336xf16>
%270 = torch_c.from_builtin_tensor %__auto.blk.29.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.30.attn_norm.weight = util.global.load @__auto.blk.30.attn_norm.weight : tensor<4096xf32>
%271 = torch_c.from_builtin_tensor %__auto.blk.30.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.30.attn_q.weight = util.global.load @__auto.blk.30.attn_q.weight : tensor<4096x4096xf16>
%272 = torch_c.from_builtin_tensor %__auto.blk.30.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.30.attn_k.weight = util.global.load @__auto.blk.30.attn_k.weight : tensor<1024x4096xf16>
%273 = torch_c.from_builtin_tensor %__auto.blk.30.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.30.attn_v.weight = util.global.load @__auto.blk.30.attn_v.weight : tensor<1024x4096xf16>
%274 = torch_c.from_builtin_tensor %__auto.blk.30.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.30.attn_output.weight = util.global.load @__auto.blk.30.attn_output.weight : tensor<4096x4096xf16>
%275 = torch_c.from_builtin_tensor %__auto.blk.30.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.30.ffn_norm.weight = util.global.load @__auto.blk.30.ffn_norm.weight : tensor<4096xf32>
%276 = torch_c.from_builtin_tensor %__auto.blk.30.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.30.ffn_gate.weight = util.global.load @__auto.blk.30.ffn_gate.weight : tensor<14336x4096xf16>
%277 = torch_c.from_builtin_tensor %__auto.blk.30.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.30.ffn_up.weight = util.global.load @__auto.blk.30.ffn_up.weight : tensor<14336x4096xf16>
%278 = torch_c.from_builtin_tensor %__auto.blk.30.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.30.ffn_down.weight = util.global.load @__auto.blk.30.ffn_down.weight : tensor<4096x14336xf16>
%279 = torch_c.from_builtin_tensor %__auto.blk.30.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.blk.31.attn_norm.weight = util.global.load @__auto.blk.31.attn_norm.weight : tensor<4096xf32>
%280 = torch_c.from_builtin_tensor %__auto.blk.31.attn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.31.attn_q.weight = util.global.load @__auto.blk.31.attn_q.weight : tensor<4096x4096xf16>
%281 = torch_c.from_builtin_tensor %__auto.blk.31.attn_q.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.31.attn_k.weight = util.global.load @__auto.blk.31.attn_k.weight : tensor<1024x4096xf16>
%282 = torch_c.from_builtin_tensor %__auto.blk.31.attn_k.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.31.attn_v.weight = util.global.load @__auto.blk.31.attn_v.weight : tensor<1024x4096xf16>
%283 = torch_c.from_builtin_tensor %__auto.blk.31.attn_v.weight : tensor<1024x4096xf16> -> !torch.vtensor<[1024,4096],f16>
%__auto.blk.31.attn_output.weight = util.global.load @__auto.blk.31.attn_output.weight : tensor<4096x4096xf16>
%284 = torch_c.from_builtin_tensor %__auto.blk.31.attn_output.weight : tensor<4096x4096xf16> -> !torch.vtensor<[4096,4096],f16>
%__auto.blk.31.ffn_norm.weight = util.global.load @__auto.blk.31.ffn_norm.weight : tensor<4096xf32>
%285 = torch_c.from_builtin_tensor %__auto.blk.31.ffn_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.blk.31.ffn_gate.weight = util.global.load @__auto.blk.31.ffn_gate.weight : tensor<14336x4096xf16>
%286 = torch_c.from_builtin_tensor %__auto.blk.31.ffn_gate.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.31.ffn_up.weight = util.global.load @__auto.blk.31.ffn_up.weight : tensor<14336x4096xf16>
%287 = torch_c.from_builtin_tensor %__auto.blk.31.ffn_up.weight : tensor<14336x4096xf16> -> !torch.vtensor<[14336,4096],f16>
%__auto.blk.31.ffn_down.weight = util.global.load @__auto.blk.31.ffn_down.weight : tensor<4096x14336xf16>
%288 = torch_c.from_builtin_tensor %__auto.blk.31.ffn_down.weight : tensor<4096x14336xf16> -> !torch.vtensor<[4096,14336],f16>
%__auto.output_norm.weight = util.global.load @__auto.output_norm.weight : tensor<4096xf32>
%289 = torch_c.from_builtin_tensor %__auto.output_norm.weight : tensor<4096xf32> -> !torch.vtensor<[4096],f32>
%__auto.output.weight = util.global.load @__auto.output.weight : tensor<128256x4096xf16>
%290 = torch_c.from_builtin_tensor %__auto.output.weight : tensor<128256x4096xf16> -> !torch.vtensor<[128256,4096],f16>
%291 = torch.copy.to_vtensor %arg3 : !torch.vtensor<[?,2097152],f16>
%292 = torch.symbolic_int "s1" {min_val = 2, max_val = 4095} : !torch.int
%293 = torch.symbolic_int "s2" {min_val = 2, max_val = 9223372036854775806} : !torch.int
torch.bind_symbolic_shape %arg0, [%292], affine_map<()[s0] -> (4, s0 * 32)> : !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %arg2, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %291, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int1 = torch.constant.int 1
%294 = torch.aten.size.int %arg0, %int1 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.int
%int0 = torch.constant.int 0
%int1_0 = torch.constant.int 1
%none = torch.constant.none
%none_1 = torch.constant.none
%cpu = torch.constant.device "cpu"
%false = torch.constant.bool false
%295 = torch.aten.arange.start_step %int0, %294, %int1_0, %none, %none_1, %cpu, %false : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %295, [%292], affine_map<()[s0] -> (s0 * 32)> : !torch.vtensor<[?],si64>
%int-1 = torch.constant.int -1
%296 = torch.aten.unsqueeze %arg1, %int-1 : !torch.vtensor<[4],si64>, !torch.int -> !torch.vtensor<[4,1],si64>
%297 = torch.aten.ge.Tensor %295, %296 : !torch.vtensor<[?],si64>, !torch.vtensor<[4,1],si64> -> !torch.vtensor<[4,?],i1>
torch.bind_symbolic_shape %297, [%292], affine_map<()[s0] -> (4, s0 * 32)> : !torch.vtensor<[4,?],i1>
%int1_2 = torch.constant.int 1
%int1_3 = torch.constant.int 1
%298 = torch.prim.ListConstruct %int1_2, %int1_3 : (!torch.int, !torch.int) -> !torch.list<int>
%int11 = torch.constant.int 11
%none_4 = torch.constant.none
%cpu_5 = torch.constant.device "cpu"
%false_6 = torch.constant.bool false
%299 = torch.aten.ones %298, %int11, %none_4, %cpu_5, %false_6 : !torch.list<int>, !torch.int, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[1,1],i1>
%int131072 = torch.constant.int 131072
%int131072_7 = torch.constant.int 131072
%300 = torch.prim.ListConstruct %int131072, %int131072_7 : (!torch.int, !torch.int) -> !torch.list<int>
%false_8 = torch.constant.bool false
%301 = torch.aten.expand %299, %300, %false_8 : !torch.vtensor<[1,1],i1>, !torch.list<int>, !torch.bool -> !torch.vtensor<[131072,131072],i1>
%int1_9 = torch.constant.int 1
%302 = torch.aten.triu %301, %int1_9 : !torch.vtensor<[131072,131072],i1>, !torch.int -> !torch.vtensor<[131072,131072],i1>
%int0_10 = torch.constant.int 0
%303 = torch.aten.unsqueeze %302, %int0_10 : !torch.vtensor<[131072,131072],i1>, !torch.int -> !torch.vtensor<[1,131072,131072],i1>
%int1_11 = torch.constant.int 1
%304 = torch.aten.unsqueeze %303, %int1_11 : !torch.vtensor<[1,131072,131072],i1>, !torch.int -> !torch.vtensor<[1,1,131072,131072],i1>
%int2 = torch.constant.int 2
%int0_12 = torch.constant.int 0
%int9223372036854775807 = torch.constant.int 9223372036854775807
%int1_13 = torch.constant.int 1
%305 = torch.aten.slice.Tensor %304, %int2, %int0_12, %int9223372036854775807, %int1_13 : !torch.vtensor<[1,1,131072,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,131072,131072],i1>
%int3 = torch.constant.int 3
%int0_14 = torch.constant.int 0
%int9223372036854775807_15 = torch.constant.int 9223372036854775807
%int1_16 = torch.constant.int 1
%306 = torch.aten.slice.Tensor %305, %int3, %int0_14, %int9223372036854775807_15, %int1_16 : !torch.vtensor<[1,1,131072,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,131072,131072],i1>
%int0_17 = torch.constant.int 0
%int0_18 = torch.constant.int 0
%int9223372036854775807_19 = torch.constant.int 9223372036854775807
%int1_20 = torch.constant.int 1
%307 = torch.aten.slice.Tensor %306, %int0_17, %int0_18, %int9223372036854775807_19, %int1_20 : !torch.vtensor<[1,1,131072,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,131072,131072],i1>
%int1_21 = torch.constant.int 1
%int0_22 = torch.constant.int 0
%int9223372036854775807_23 = torch.constant.int 9223372036854775807
%int1_24 = torch.constant.int 1
%308 = torch.aten.slice.Tensor %307, %int1_21, %int0_22, %int9223372036854775807_23, %int1_24 : !torch.vtensor<[1,1,131072,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,131072,131072],i1>
%int0_25 = torch.constant.int 0
%309 = torch.aten.size.int %295, %int0_25 : !torch.vtensor<[?],si64>, !torch.int -> !torch.int
%int2_26 = torch.constant.int 2
%int0_27 = torch.constant.int 0
%int1_28 = torch.constant.int 1
%310 = torch.aten.slice.Tensor %308, %int2_26, %int0_27, %309, %int1_28 : !torch.vtensor<[1,1,131072,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,?,131072],i1>
torch.bind_symbolic_shape %310, [%292], affine_map<()[s0] -> (1, 1, s0 * 32, 131072)> : !torch.vtensor<[1,1,?,131072],i1>
%int3_29 = torch.constant.int 3
%int0_30 = torch.constant.int 0
%int1_31 = torch.constant.int 1
%311 = torch.aten.slice.Tensor %310, %int3_29, %int0_30, %309, %int1_31 : !torch.vtensor<[1,1,?,131072],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,1,?,?],i1>
torch.bind_symbolic_shape %311, [%292], affine_map<()[s0] -> (1, 1, s0 * 32, s0 * 32)> : !torch.vtensor<[1,1,?,?],i1>
%int0_32 = torch.constant.int 0
%int0_33 = torch.constant.int 0
%int9223372036854775807_34 = torch.constant.int 9223372036854775807
%int1_35 = torch.constant.int 1
%312 = torch.aten.slice.Tensor %297, %int0_32, %int0_33, %int9223372036854775807_34, %int1_35 : !torch.vtensor<[4,?],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[4,?],i1>
torch.bind_symbolic_shape %312, [%292], affine_map<()[s0] -> (4, s0 * 32)> : !torch.vtensor<[4,?],i1>
%int1_36 = torch.constant.int 1
%313 = torch.aten.unsqueeze %312, %int1_36 : !torch.vtensor<[4,?],i1>, !torch.int -> !torch.vtensor<[4,1,?],i1>
torch.bind_symbolic_shape %313, [%292], affine_map<()[s0] -> (4, 1, s0 * 32)> : !torch.vtensor<[4,1,?],i1>
%int2_37 = torch.constant.int 2
%314 = torch.aten.unsqueeze %313, %int2_37 : !torch.vtensor<[4,1,?],i1>, !torch.int -> !torch.vtensor<[4,1,1,?],i1>
torch.bind_symbolic_shape %314, [%292], affine_map<()[s0] -> (4, 1, 1, s0 * 32)> : !torch.vtensor<[4,1,1,?],i1>
%int3_38 = torch.constant.int 3
%int0_39 = torch.constant.int 0
%int9223372036854775807_40 = torch.constant.int 9223372036854775807
%int1_41 = torch.constant.int 1
%315 = torch.aten.slice.Tensor %314, %int3_38, %int0_39, %int9223372036854775807_40, %int1_41 : !torch.vtensor<[4,1,1,?],i1>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[4,1,1,?],i1>
torch.bind_symbolic_shape %315, [%292], affine_map<()[s0] -> (4, 1, 1, s0 * 32)> : !torch.vtensor<[4,1,1,?],i1>
%316 = torch.aten.logical_or %311, %315 : !torch.vtensor<[1,1,?,?],i1>, !torch.vtensor<[4,1,1,?],i1> -> !torch.vtensor<[4,1,?,?],i1>
torch.bind_symbolic_shape %316, [%292], affine_map<()[s0] -> (4, 1, s0 * 32, s0 * 32)> : !torch.vtensor<[4,1,?,?],i1>
%int0_42 = torch.constant.int 0
%int6 = torch.constant.int 6
%int0_43 = torch.constant.int 0
%cpu_44 = torch.constant.device "cpu"
%none_45 = torch.constant.none
%317 = torch.aten.scalar_tensor %int0_42, %int6, %int0_43, %cpu_44, %none_45 : !torch.int, !torch.int, !torch.int, !torch.Device, !torch.none -> !torch.vtensor<[],f32>
%float-Inf = torch.constant.float 0xFFF0000000000000
%int6_46 = torch.constant.int 6
%int0_47 = torch.constant.int 0
%cpu_48 = torch.constant.device "cpu"
%none_49 = torch.constant.none
%318 = torch.aten.scalar_tensor %float-Inf, %int6_46, %int0_47, %cpu_48, %none_49 : !torch.float, !torch.int, !torch.int, !torch.Device, !torch.none -> !torch.vtensor<[],f32>
%319 = torch.aten.where.self %316, %318, %317 : !torch.vtensor<[4,1,?,?],i1>, !torch.vtensor<[],f32>, !torch.vtensor<[],f32> -> !torch.vtensor<[4,1,?,?],f32>
torch.bind_symbolic_shape %319, [%292], affine_map<()[s0] -> (4, 1, s0 * 32, s0 * 32)> : !torch.vtensor<[4,1,?,?],f32>
%int5 = torch.constant.int 5
%320 = torch.prims.convert_element_type %319, %int5 : !torch.vtensor<[4,1,?,?],f32>, !torch.int -> !torch.vtensor<[4,1,?,?],f16>
torch.bind_symbolic_shape %320, [%292], affine_map<()[s0] -> (4, 1, s0 * 32, s0 * 32)> : !torch.vtensor<[4,1,?,?],f16>
%int-1_50 = torch.constant.int -1
%false_51 = torch.constant.bool false
%false_52 = torch.constant.bool false
%321 = torch.aten.embedding %0, %arg0, %int-1_50, %false_51, %false_52 : !torch.vtensor<[128256,4096],f16>, !torch.vtensor<[4,?],si64>, !torch.int, !torch.bool, !torch.bool -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %321, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_53 = torch.constant.int 6
%322 = torch.prims.convert_element_type %321, %int6_53 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %322, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_54 = torch.constant.int 2
%323 = torch.aten.pow.Tensor_Scalar %322, %int2_54 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %323, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_55 = torch.constant.int -1
%324 = torch.prim.ListConstruct %int-1_55 : (!torch.int) -> !torch.list<int>
%true = torch.constant.bool true
%none_56 = torch.constant.none
%325 = torch.aten.mean.dim %323, %324, %true, %none_56 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %325, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06 = torch.constant.float 9.9999997473787516E-6
%int1_57 = torch.constant.int 1
%326 = torch.aten.add.Scalar %325, %float9.999990e-06, %int1_57 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %326, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%327 = torch.aten.rsqrt %326 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %327, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%328 = torch.aten.mul.Tensor %322, %327 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %328, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%329 = torch.aten.mul.Tensor %1, %328 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %329, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_58 = torch.constant.int 5
%330 = torch.prims.convert_element_type %329, %int5_58 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %330, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2 = torch.constant.int -2
%int-1_59 = torch.constant.int -1
%331 = torch.aten.transpose.int %2, %int-2, %int-1_59 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4 = torch.constant.int 4
%332 = torch.aten.mul.int %int4, %294 : !torch.int, !torch.int -> !torch.int
%int4096 = torch.constant.int 4096
%333 = torch.prim.ListConstruct %332, %int4096 : (!torch.int, !torch.int) -> !torch.list<int>
%334 = torch.aten.view %330, %333 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %334, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%335 = torch.aten.mm %334, %331 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %335, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_60 = torch.constant.int 4
%int4096_61 = torch.constant.int 4096
%336 = torch.prim.ListConstruct %int4_60, %294, %int4096_61 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%337 = torch.aten.view %335, %336 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %337, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_62 = torch.constant.int -2
%int-1_63 = torch.constant.int -1
%338 = torch.aten.transpose.int %3, %int-2_62, %int-1_63 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_64 = torch.constant.int 4
%339 = torch.aten.mul.int %int4_64, %294 : !torch.int, !torch.int -> !torch.int
%int4096_65 = torch.constant.int 4096
%340 = torch.prim.ListConstruct %339, %int4096_65 : (!torch.int, !torch.int) -> !torch.list<int>
%341 = torch.aten.view %330, %340 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %341, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%342 = torch.aten.mm %341, %338 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %342, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_66 = torch.constant.int 4
%int1024 = torch.constant.int 1024
%343 = torch.prim.ListConstruct %int4_66, %294, %int1024 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%344 = torch.aten.view %342, %343 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %344, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_67 = torch.constant.int -2
%int-1_68 = torch.constant.int -1
%345 = torch.aten.transpose.int %4, %int-2_67, %int-1_68 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_69 = torch.constant.int 4
%346 = torch.aten.mul.int %int4_69, %294 : !torch.int, !torch.int -> !torch.int
%int4096_70 = torch.constant.int 4096
%347 = torch.prim.ListConstruct %346, %int4096_70 : (!torch.int, !torch.int) -> !torch.list<int>
%348 = torch.aten.view %330, %347 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %348, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%349 = torch.aten.mm %348, %345 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %349, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_71 = torch.constant.int 4
%int1024_72 = torch.constant.int 1024
%350 = torch.prim.ListConstruct %int4_71, %294, %int1024_72 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%351 = torch.aten.view %349, %350 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %351, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_73 = torch.constant.int 4
%int32 = torch.constant.int 32
%int128 = torch.constant.int 128
%352 = torch.prim.ListConstruct %int4_73, %294, %int32, %int128 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%353 = torch.aten.view %337, %352 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %353, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_74 = torch.constant.int 4
%int8 = torch.constant.int 8
%int128_75 = torch.constant.int 128
%354 = torch.prim.ListConstruct %int4_74, %294, %int8, %int128_75 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%355 = torch.aten.view %344, %354 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %355, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_76 = torch.constant.int 4
%int8_77 = torch.constant.int 8
%int128_78 = torch.constant.int 128
%356 = torch.prim.ListConstruct %int4_76, %294, %int8_77, %int128_78 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%357 = torch.aten.view %351, %356 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %357, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_79 = torch.constant.int 131072
%none_80 = torch.constant.none
%none_81 = torch.constant.none
%cpu_82 = torch.constant.device "cpu"
%false_83 = torch.constant.bool false
%358 = torch.aten.arange %int131072_79, %none_80, %none_81, %cpu_82, %false_83 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_84 = torch.constant.int 0
%int128_85 = torch.constant.int 128
%int2_86 = torch.constant.int 2
%none_87 = torch.constant.none
%none_88 = torch.constant.none
%cpu_89 = torch.constant.device "cpu"
%false_90 = torch.constant.bool false
%359 = torch.aten.arange.start_step %int0_84, %int128_85, %int2_86, %none_87, %none_88, %cpu_89, %false_90 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_91 = torch.constant.int 0
%int0_92 = torch.constant.int 0
%int64 = torch.constant.int 64
%int1_93 = torch.constant.int 1
%360 = torch.aten.slice.Tensor %359, %int0_91, %int0_92, %int64, %int1_93 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_94 = torch.constant.int 6
%361 = torch.prims.convert_element_type %360, %int6_94 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_95 = torch.constant.int 128
%362 = torch.aten.div.Scalar %361, %int128_95 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05 = torch.constant.float 5.000000e+05
%363 = torch.aten.pow.Scalar %float5.000000e05, %362 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%364 = torch.aten.reciprocal %363 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00 = torch.constant.float 1.000000e+00
%365 = torch.aten.mul.Scalar %364, %float1.000000e00 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_96 = torch.constant.int 131072
%int1_97 = torch.constant.int 1
%366 = torch.prim.ListConstruct %int131072_96, %int1_97 : (!torch.int, !torch.int) -> !torch.list<int>
%367 = torch.aten.view %358, %366 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%368 = torch.aten.mul.Tensor %367, %365 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%369 = torch.aten.cos %368 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%370 = torch.aten.sin %368 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%371 = torch.aten.complex %369, %370 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_98 = torch.constant.int 1
%372 = torch.aten.size.int %337, %int1_98 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_99 = torch.constant.int 0
%373 = torch.aten.add.int %int0_99, %372 : !torch.int, !torch.int -> !torch.int
%int0_100 = torch.constant.int 0
%int0_101 = torch.constant.int 0
%int1_102 = torch.constant.int 1
%374 = torch.aten.slice.Tensor %371, %int0_100, %int0_101, %373, %int1_102 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %374, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_103 = torch.constant.int 1
%int0_104 = torch.constant.int 0
%int9223372036854775807_105 = torch.constant.int 9223372036854775807
%int1_106 = torch.constant.int 1
%375 = torch.aten.slice.Tensor %374, %int1_103, %int0_104, %int9223372036854775807_105, %int1_106 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %375, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_107 = torch.constant.int 0
%376 = torch.aten.unsqueeze %375, %int0_107 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %376, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_108 = torch.constant.int 2
%377 = torch.aten.unsqueeze %376, %int2_108 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %377, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_109 = torch.constant.int 3
%int0_110 = torch.constant.int 0
%int9223372036854775807_111 = torch.constant.int 9223372036854775807
%int1_112 = torch.constant.int 1
%378 = torch.aten.slice.Tensor %377, %int3_109, %int0_110, %int9223372036854775807_111, %int1_112 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %378, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%379 = torch_c.to_builtin_tensor %353 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1 = arith.constant 1 : index
%dim = tensor.dim %379, %c1 : tensor<4x?x32x128xf16>
%380 = flow.tensor.bitcast %379 : tensor<4x?x32x128xf16>{%dim} -> tensor<4x?x32x64xcomplex<f16>>{%dim}
%381 = torch_c.from_builtin_tensor %380 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %381, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%382 = torch.aten.mul.Tensor %381, %378 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %382, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%383 = torch_c.to_builtin_tensor %382 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_113 = arith.constant 1 : index
%dim_114 = tensor.dim %383, %c1_113 : tensor<4x?x32x64xcomplex<f32>>
%384 = flow.tensor.bitcast %383 : tensor<4x?x32x64xcomplex<f32>>{%dim_114} -> tensor<4x?x32x128xf32>{%dim_114}
%385 = torch_c.from_builtin_tensor %384 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %385, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_115 = torch.constant.int 5
%386 = torch.prims.convert_element_type %385, %int5_115 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %386, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_116 = torch.constant.int 131072
%none_117 = torch.constant.none
%none_118 = torch.constant.none
%cpu_119 = torch.constant.device "cpu"
%false_120 = torch.constant.bool false
%387 = torch.aten.arange %int131072_116, %none_117, %none_118, %cpu_119, %false_120 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_121 = torch.constant.int 0
%int128_122 = torch.constant.int 128
%int2_123 = torch.constant.int 2
%none_124 = torch.constant.none
%none_125 = torch.constant.none
%cpu_126 = torch.constant.device "cpu"
%false_127 = torch.constant.bool false
%388 = torch.aten.arange.start_step %int0_121, %int128_122, %int2_123, %none_124, %none_125, %cpu_126, %false_127 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_128 = torch.constant.int 0
%int0_129 = torch.constant.int 0
%int64_130 = torch.constant.int 64
%int1_131 = torch.constant.int 1
%389 = torch.aten.slice.Tensor %388, %int0_128, %int0_129, %int64_130, %int1_131 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_132 = torch.constant.int 6
%390 = torch.prims.convert_element_type %389, %int6_132 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_133 = torch.constant.int 128
%391 = torch.aten.div.Scalar %390, %int128_133 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_134 = torch.constant.float 5.000000e+05
%392 = torch.aten.pow.Scalar %float5.000000e05_134, %391 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%393 = torch.aten.reciprocal %392 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_135 = torch.constant.float 1.000000e+00
%394 = torch.aten.mul.Scalar %393, %float1.000000e00_135 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_136 = torch.constant.int 131072
%int1_137 = torch.constant.int 1
%395 = torch.prim.ListConstruct %int131072_136, %int1_137 : (!torch.int, !torch.int) -> !torch.list<int>
%396 = torch.aten.view %387, %395 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%397 = torch.aten.mul.Tensor %396, %394 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%398 = torch.aten.cos %397 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%399 = torch.aten.sin %397 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%400 = torch.aten.complex %398, %399 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_138 = torch.constant.int 1
%401 = torch.aten.size.int %344, %int1_138 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_139 = torch.constant.int 0
%402 = torch.aten.add.int %int0_139, %401 : !torch.int, !torch.int -> !torch.int
%int0_140 = torch.constant.int 0
%int0_141 = torch.constant.int 0
%int1_142 = torch.constant.int 1
%403 = torch.aten.slice.Tensor %400, %int0_140, %int0_141, %402, %int1_142 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %403, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_143 = torch.constant.int 1
%int0_144 = torch.constant.int 0
%int9223372036854775807_145 = torch.constant.int 9223372036854775807
%int1_146 = torch.constant.int 1
%404 = torch.aten.slice.Tensor %403, %int1_143, %int0_144, %int9223372036854775807_145, %int1_146 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %404, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_147 = torch.constant.int 0
%405 = torch.aten.unsqueeze %404, %int0_147 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %405, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_148 = torch.constant.int 2
%406 = torch.aten.unsqueeze %405, %int2_148 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %406, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_149 = torch.constant.int 3
%int0_150 = torch.constant.int 0
%int9223372036854775807_151 = torch.constant.int 9223372036854775807
%int1_152 = torch.constant.int 1
%407 = torch.aten.slice.Tensor %406, %int3_149, %int0_150, %int9223372036854775807_151, %int1_152 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %407, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%408 = torch_c.to_builtin_tensor %355 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_153 = arith.constant 1 : index
%dim_154 = tensor.dim %408, %c1_153 : tensor<4x?x8x128xf16>
%409 = flow.tensor.bitcast %408 : tensor<4x?x8x128xf16>{%dim_154} -> tensor<4x?x8x64xcomplex<f16>>{%dim_154}
%410 = torch_c.from_builtin_tensor %409 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %410, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%411 = torch.aten.mul.Tensor %410, %407 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %411, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%412 = torch_c.to_builtin_tensor %411 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_155 = arith.constant 1 : index
%dim_156 = tensor.dim %412, %c1_155 : tensor<4x?x8x64xcomplex<f32>>
%413 = flow.tensor.bitcast %412 : tensor<4x?x8x64xcomplex<f32>>{%dim_156} -> tensor<4x?x8x128xf32>{%dim_156}
%414 = torch_c.from_builtin_tensor %413 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %414, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_157 = torch.constant.int 5
%415 = torch.prims.convert_element_type %414, %int5_157 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %415, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int0_158 = torch.constant.int 0
%416 = torch.aten.size.int %291, %int0_158 : !torch.vtensor<[?,2097152],f16>, !torch.int -> !torch.int
%int32_159 = torch.constant.int 32
%int2_160 = torch.constant.int 2
%int32_161 = torch.constant.int 32
%int8_162 = torch.constant.int 8
%int128_163 = torch.constant.int 128
%417 = torch.prim.ListConstruct %416, %int32_159, %int2_160, %int32_161, %int8_162, %int128_163 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%418 = torch.aten.view %291, %417 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %418, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_164 = torch.constant.int 32
%419 = torch.aten.mul.int %416, %int32_164 : !torch.int, !torch.int -> !torch.int
%int2_165 = torch.constant.int 2
%420 = torch.aten.mul.int %419, %int2_165 : !torch.int, !torch.int -> !torch.int
%int32_166 = torch.constant.int 32
%int8_167 = torch.constant.int 8
%int128_168 = torch.constant.int 128
%421 = torch.prim.ListConstruct %420, %int32_166, %int8_167, %int128_168 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%422 = torch.aten.view %418, %421 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %422, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int64_169 = torch.constant.int 64
%423 = torch.aten.mul.Scalar %arg2, %int64_169 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %423, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int0_170 = torch.constant.int 0
%int1_171 = torch.constant.int 1
%424 = torch.aten.add.Scalar %423, %int0_170, %int1_171 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %424, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int1_172 = torch.constant.int 1
%425 = torch.aten.size.int %arg2, %int1_172 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.int
%int4_173 = torch.constant.int 4
%int32_174 = torch.constant.int 32
%int8_175 = torch.constant.int 8
%int128_176 = torch.constant.int 128
%426 = torch.prim.ListConstruct %int4_173, %425, %int32_174, %int8_175, %int128_176 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%427 = torch.aten.view %415, %426 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %427, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_177 = torch.constant.int 4
%428 = torch.aten.mul.int %int4_177, %425 : !torch.int, !torch.int -> !torch.int
%int32_178 = torch.constant.int 32
%int8_179 = torch.constant.int 8
%int128_180 = torch.constant.int 128
%429 = torch.prim.ListConstruct %428, %int32_178, %int8_179, %int128_180 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%430 = torch.aten.view %427, %429 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %430, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_181 = torch.constant.int 4
%431 = torch.aten.mul.int %int4_181, %425 : !torch.int, !torch.int -> !torch.int
%432 = torch.prim.ListConstruct %431 : (!torch.int) -> !torch.list<int>
%433 = torch.aten.view %424, %432 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %433, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%434 = torch.prim.ListConstruct %433 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_182 = torch.constant.bool false
%435 = torch.aten.index_put %422, %434, %430, %false_182 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %435, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_183 = torch.constant.int 32
%int2_184 = torch.constant.int 2
%int32_185 = torch.constant.int 32
%int8_186 = torch.constant.int 8
%int128_187 = torch.constant.int 128
%436 = torch.prim.ListConstruct %416, %int32_183, %int2_184, %int32_185, %int8_186, %int128_187 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%437 = torch.aten.view %435, %436 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %437, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152 = torch.constant.int 2097152
%438 = torch.prim.ListConstruct %416, %int2097152 : (!torch.int, !torch.int) -> !torch.list<int>
%439 = torch.aten.view %437, %438 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %439, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_188 = torch.constant.int 32
%int2_189 = torch.constant.int 2
%int32_190 = torch.constant.int 32
%int8_191 = torch.constant.int 8
%int128_192 = torch.constant.int 128
%440 = torch.prim.ListConstruct %416, %int32_188, %int2_189, %int32_190, %int8_191, %int128_192 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%441 = torch.aten.view %439, %440 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %441, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_193 = torch.constant.int 32
%int8_194 = torch.constant.int 8
%int128_195 = torch.constant.int 128
%442 = torch.prim.ListConstruct %420, %int32_193, %int8_194, %int128_195 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%443 = torch.aten.view %441, %442 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %443, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_196 = torch.constant.int 4
%int32_197 = torch.constant.int 32
%int8_198 = torch.constant.int 8
%int128_199 = torch.constant.int 128
%444 = torch.prim.ListConstruct %int4_196, %425, %int32_197, %int8_198, %int128_199 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%445 = torch.aten.view %357, %444 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %445, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_200 = torch.constant.int 4
%446 = torch.aten.mul.int %int4_200, %425 : !torch.int, !torch.int -> !torch.int
%int32_201 = torch.constant.int 32
%int8_202 = torch.constant.int 8
%int128_203 = torch.constant.int 128
%447 = torch.prim.ListConstruct %446, %int32_201, %int8_202, %int128_203 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%448 = torch.aten.view %445, %447 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %448, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_204 = torch.constant.int 1
%int1_205 = torch.constant.int 1
%449 = torch.aten.add.Scalar %424, %int1_204, %int1_205 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %449, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_206 = torch.constant.int 4
%450 = torch.aten.mul.int %int4_206, %425 : !torch.int, !torch.int -> !torch.int
%451 = torch.prim.ListConstruct %450 : (!torch.int) -> !torch.list<int>
%452 = torch.aten.view %449, %451 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %452, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%453 = torch.prim.ListConstruct %452 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_207 = torch.constant.bool false
%454 = torch.aten.index_put %443, %453, %448, %false_207 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %454, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_208 = torch.constant.int 32
%int2_209 = torch.constant.int 2
%int32_210 = torch.constant.int 32
%int8_211 = torch.constant.int 8
%int128_212 = torch.constant.int 128
%455 = torch.prim.ListConstruct %416, %int32_208, %int2_209, %int32_210, %int8_211, %int128_212 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%456 = torch.aten.view %454, %455 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %456, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_213 = torch.constant.int 2097152
%457 = torch.prim.ListConstruct %416, %int2097152_213 : (!torch.int, !torch.int) -> !torch.list<int>
%458 = torch.aten.view %456, %457 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %458, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_214 = torch.constant.int -2
%459 = torch.aten.unsqueeze %415, %int-2_214 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %459, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_215 = torch.constant.int 4
%int8_216 = torch.constant.int 8
%int4_217 = torch.constant.int 4
%int128_218 = torch.constant.int 128
%460 = torch.prim.ListConstruct %int4_215, %401, %int8_216, %int4_217, %int128_218 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_219 = torch.constant.bool false
%461 = torch.aten.expand %459, %460, %false_219 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %461, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_220 = torch.constant.int 0
%462 = torch.aten.clone %461, %int0_220 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %462, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_221 = torch.constant.int 4
%int32_222 = torch.constant.int 32
%int128_223 = torch.constant.int 128
%463 = torch.prim.ListConstruct %int4_221, %401, %int32_222, %int128_223 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%464 = torch.aten._unsafe_view %462, %463 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %464, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_224 = torch.constant.int -2
%465 = torch.aten.unsqueeze %357, %int-2_224 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %465, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_225 = torch.constant.int 1
%466 = torch.aten.size.int %351, %int1_225 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_226 = torch.constant.int 4
%int8_227 = torch.constant.int 8
%int4_228 = torch.constant.int 4
%int128_229 = torch.constant.int 128
%467 = torch.prim.ListConstruct %int4_226, %466, %int8_227, %int4_228, %int128_229 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_230 = torch.constant.bool false
%468 = torch.aten.expand %465, %467, %false_230 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %468, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_231 = torch.constant.int 0
%469 = torch.aten.clone %468, %int0_231 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %469, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_232 = torch.constant.int 4
%int32_233 = torch.constant.int 32
%int128_234 = torch.constant.int 128
%470 = torch.prim.ListConstruct %int4_232, %466, %int32_233, %int128_234 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%471 = torch.aten._unsafe_view %469, %470 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %471, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_235 = torch.constant.int 1
%int2_236 = torch.constant.int 2
%472 = torch.aten.transpose.int %386, %int1_235, %int2_236 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %472, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_237 = torch.constant.int 1
%int2_238 = torch.constant.int 2
%473 = torch.aten.transpose.int %464, %int1_237, %int2_238 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %473, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_239 = torch.constant.int 1
%int2_240 = torch.constant.int 2
%474 = torch.aten.transpose.int %471, %int1_239, %int2_240 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %474, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00 = torch.constant.float 0.000000e+00
%false_241 = torch.constant.bool false
%none_242 = torch.constant.none
%475:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%472, %473, %474, %float0.000000e00, %false_241, %320, %none_242) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %475#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_243 = torch.constant.int 1
%int2_244 = torch.constant.int 2
%476 = torch.aten.transpose.int %475#0, %int1_243, %int2_244 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %476, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_245 = torch.constant.int 4
%int4096_246 = torch.constant.int 4096
%477 = torch.prim.ListConstruct %int4_245, %372, %int4096_246 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%478 = torch.aten.view %476, %477 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %478, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_247 = torch.constant.int -2
%int-1_248 = torch.constant.int -1
%479 = torch.aten.transpose.int %5, %int-2_247, %int-1_248 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_249 = torch.constant.int 4
%480 = torch.aten.mul.int %int4_249, %372 : !torch.int, !torch.int -> !torch.int
%int4096_250 = torch.constant.int 4096
%481 = torch.prim.ListConstruct %480, %int4096_250 : (!torch.int, !torch.int) -> !torch.list<int>
%482 = torch.aten.view %478, %481 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %482, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%483 = torch.aten.mm %482, %479 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %483, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_251 = torch.constant.int 4
%int4096_252 = torch.constant.int 4096
%484 = torch.prim.ListConstruct %int4_251, %372, %int4096_252 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%485 = torch.aten.view %483, %484 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %485, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_253 = torch.constant.int 1
%486 = torch.aten.add.Tensor %321, %485, %int1_253 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %486, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_254 = torch.constant.int 6
%487 = torch.prims.convert_element_type %486, %int6_254 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %487, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_255 = torch.constant.int 2
%488 = torch.aten.pow.Tensor_Scalar %487, %int2_255 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %488, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_256 = torch.constant.int -1
%489 = torch.prim.ListConstruct %int-1_256 : (!torch.int) -> !torch.list<int>
%true_257 = torch.constant.bool true
%none_258 = torch.constant.none
%490 = torch.aten.mean.dim %488, %489, %true_257, %none_258 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %490, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_259 = torch.constant.float 9.9999997473787516E-6
%int1_260 = torch.constant.int 1
%491 = torch.aten.add.Scalar %490, %float9.999990e-06_259, %int1_260 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %491, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%492 = torch.aten.rsqrt %491 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %492, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%493 = torch.aten.mul.Tensor %487, %492 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %493, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%494 = torch.aten.mul.Tensor %6, %493 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %494, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_261 = torch.constant.int 5
%495 = torch.prims.convert_element_type %494, %int5_261 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %495, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_262 = torch.constant.int -2
%int-1_263 = torch.constant.int -1
%496 = torch.aten.transpose.int %7, %int-2_262, %int-1_263 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_264 = torch.constant.int 4
%497 = torch.aten.mul.int %int4_264, %294 : !torch.int, !torch.int -> !torch.int
%int4096_265 = torch.constant.int 4096
%498 = torch.prim.ListConstruct %497, %int4096_265 : (!torch.int, !torch.int) -> !torch.list<int>
%499 = torch.aten.view %495, %498 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %499, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%500 = torch.aten.mm %499, %496 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %500, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_266 = torch.constant.int 4
%int14336 = torch.constant.int 14336
%501 = torch.prim.ListConstruct %int4_266, %294, %int14336 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%502 = torch.aten.view %500, %501 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %502, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%503 = torch.aten.silu %502 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %503, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_267 = torch.constant.int -2
%int-1_268 = torch.constant.int -1
%504 = torch.aten.transpose.int %8, %int-2_267, %int-1_268 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_269 = torch.constant.int 4
%505 = torch.aten.mul.int %int4_269, %294 : !torch.int, !torch.int -> !torch.int
%int4096_270 = torch.constant.int 4096
%506 = torch.prim.ListConstruct %505, %int4096_270 : (!torch.int, !torch.int) -> !torch.list<int>
%507 = torch.aten.view %495, %506 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %507, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%508 = torch.aten.mm %507, %504 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %508, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_271 = torch.constant.int 4
%int14336_272 = torch.constant.int 14336
%509 = torch.prim.ListConstruct %int4_271, %294, %int14336_272 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%510 = torch.aten.view %508, %509 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %510, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%511 = torch.aten.mul.Tensor %503, %510 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %511, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_273 = torch.constant.int -2
%int-1_274 = torch.constant.int -1
%512 = torch.aten.transpose.int %9, %int-2_273, %int-1_274 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_275 = torch.constant.int 1
%513 = torch.aten.size.int %502, %int1_275 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_276 = torch.constant.int 4
%514 = torch.aten.mul.int %int4_276, %513 : !torch.int, !torch.int -> !torch.int
%int14336_277 = torch.constant.int 14336
%515 = torch.prim.ListConstruct %514, %int14336_277 : (!torch.int, !torch.int) -> !torch.list<int>
%516 = torch.aten.view %511, %515 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %516, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%517 = torch.aten.mm %516, %512 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %517, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_278 = torch.constant.int 4
%int4096_279 = torch.constant.int 4096
%518 = torch.prim.ListConstruct %int4_278, %513, %int4096_279 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%519 = torch.aten.view %517, %518 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %519, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_280 = torch.constant.int 1
%520 = torch.aten.add.Tensor %486, %519, %int1_280 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %520, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_281 = torch.constant.int 6
%521 = torch.prims.convert_element_type %520, %int6_281 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %521, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_282 = torch.constant.int 2
%522 = torch.aten.pow.Tensor_Scalar %521, %int2_282 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %522, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_283 = torch.constant.int -1
%523 = torch.prim.ListConstruct %int-1_283 : (!torch.int) -> !torch.list<int>
%true_284 = torch.constant.bool true
%none_285 = torch.constant.none
%524 = torch.aten.mean.dim %522, %523, %true_284, %none_285 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %524, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_286 = torch.constant.float 9.9999997473787516E-6
%int1_287 = torch.constant.int 1
%525 = torch.aten.add.Scalar %524, %float9.999990e-06_286, %int1_287 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %525, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%526 = torch.aten.rsqrt %525 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %526, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%527 = torch.aten.mul.Tensor %521, %526 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %527, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%528 = torch.aten.mul.Tensor %10, %527 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %528, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_288 = torch.constant.int 5
%529 = torch.prims.convert_element_type %528, %int5_288 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %529, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_289 = torch.constant.int -2
%int-1_290 = torch.constant.int -1
%530 = torch.aten.transpose.int %11, %int-2_289, %int-1_290 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_291 = torch.constant.int 4
%531 = torch.aten.mul.int %int4_291, %294 : !torch.int, !torch.int -> !torch.int
%int4096_292 = torch.constant.int 4096
%532 = torch.prim.ListConstruct %531, %int4096_292 : (!torch.int, !torch.int) -> !torch.list<int>
%533 = torch.aten.view %529, %532 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %533, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%534 = torch.aten.mm %533, %530 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %534, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_293 = torch.constant.int 4
%int4096_294 = torch.constant.int 4096
%535 = torch.prim.ListConstruct %int4_293, %294, %int4096_294 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%536 = torch.aten.view %534, %535 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %536, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_295 = torch.constant.int -2
%int-1_296 = torch.constant.int -1
%537 = torch.aten.transpose.int %12, %int-2_295, %int-1_296 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_297 = torch.constant.int 4
%538 = torch.aten.mul.int %int4_297, %294 : !torch.int, !torch.int -> !torch.int
%int4096_298 = torch.constant.int 4096
%539 = torch.prim.ListConstruct %538, %int4096_298 : (!torch.int, !torch.int) -> !torch.list<int>
%540 = torch.aten.view %529, %539 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %540, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%541 = torch.aten.mm %540, %537 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %541, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_299 = torch.constant.int 4
%int1024_300 = torch.constant.int 1024
%542 = torch.prim.ListConstruct %int4_299, %294, %int1024_300 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%543 = torch.aten.view %541, %542 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %543, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_301 = torch.constant.int -2
%int-1_302 = torch.constant.int -1
%544 = torch.aten.transpose.int %13, %int-2_301, %int-1_302 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_303 = torch.constant.int 4
%545 = torch.aten.mul.int %int4_303, %294 : !torch.int, !torch.int -> !torch.int
%int4096_304 = torch.constant.int 4096
%546 = torch.prim.ListConstruct %545, %int4096_304 : (!torch.int, !torch.int) -> !torch.list<int>
%547 = torch.aten.view %529, %546 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %547, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%548 = torch.aten.mm %547, %544 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %548, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_305 = torch.constant.int 4
%int1024_306 = torch.constant.int 1024
%549 = torch.prim.ListConstruct %int4_305, %294, %int1024_306 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%550 = torch.aten.view %548, %549 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %550, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_307 = torch.constant.int 4
%int32_308 = torch.constant.int 32
%int128_309 = torch.constant.int 128
%551 = torch.prim.ListConstruct %int4_307, %294, %int32_308, %int128_309 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%552 = torch.aten.view %536, %551 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %552, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_310 = torch.constant.int 4
%int8_311 = torch.constant.int 8
%int128_312 = torch.constant.int 128
%553 = torch.prim.ListConstruct %int4_310, %294, %int8_311, %int128_312 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%554 = torch.aten.view %543, %553 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %554, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_313 = torch.constant.int 4
%int8_314 = torch.constant.int 8
%int128_315 = torch.constant.int 128
%555 = torch.prim.ListConstruct %int4_313, %294, %int8_314, %int128_315 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%556 = torch.aten.view %550, %555 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %556, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_316 = torch.constant.int 131072
%none_317 = torch.constant.none
%none_318 = torch.constant.none
%cpu_319 = torch.constant.device "cpu"
%false_320 = torch.constant.bool false
%557 = torch.aten.arange %int131072_316, %none_317, %none_318, %cpu_319, %false_320 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_321 = torch.constant.int 0
%int128_322 = torch.constant.int 128
%int2_323 = torch.constant.int 2
%none_324 = torch.constant.none
%none_325 = torch.constant.none
%cpu_326 = torch.constant.device "cpu"
%false_327 = torch.constant.bool false
%558 = torch.aten.arange.start_step %int0_321, %int128_322, %int2_323, %none_324, %none_325, %cpu_326, %false_327 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_328 = torch.constant.int 0
%int0_329 = torch.constant.int 0
%int64_330 = torch.constant.int 64
%int1_331 = torch.constant.int 1
%559 = torch.aten.slice.Tensor %558, %int0_328, %int0_329, %int64_330, %int1_331 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_332 = torch.constant.int 6
%560 = torch.prims.convert_element_type %559, %int6_332 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_333 = torch.constant.int 128
%561 = torch.aten.div.Scalar %560, %int128_333 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_334 = torch.constant.float 5.000000e+05
%562 = torch.aten.pow.Scalar %float5.000000e05_334, %561 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%563 = torch.aten.reciprocal %562 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_335 = torch.constant.float 1.000000e+00
%564 = torch.aten.mul.Scalar %563, %float1.000000e00_335 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_336 = torch.constant.int 131072
%int1_337 = torch.constant.int 1
%565 = torch.prim.ListConstruct %int131072_336, %int1_337 : (!torch.int, !torch.int) -> !torch.list<int>
%566 = torch.aten.view %557, %565 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%567 = torch.aten.mul.Tensor %566, %564 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%568 = torch.aten.cos %567 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%569 = torch.aten.sin %567 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%570 = torch.aten.complex %568, %569 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_338 = torch.constant.int 1
%571 = torch.aten.size.int %536, %int1_338 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_339 = torch.constant.int 0
%572 = torch.aten.add.int %int0_339, %571 : !torch.int, !torch.int -> !torch.int
%int0_340 = torch.constant.int 0
%int0_341 = torch.constant.int 0
%int1_342 = torch.constant.int 1
%573 = torch.aten.slice.Tensor %570, %int0_340, %int0_341, %572, %int1_342 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %573, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_343 = torch.constant.int 1
%int0_344 = torch.constant.int 0
%int9223372036854775807_345 = torch.constant.int 9223372036854775807
%int1_346 = torch.constant.int 1
%574 = torch.aten.slice.Tensor %573, %int1_343, %int0_344, %int9223372036854775807_345, %int1_346 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %574, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_347 = torch.constant.int 0
%575 = torch.aten.unsqueeze %574, %int0_347 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %575, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_348 = torch.constant.int 2
%576 = torch.aten.unsqueeze %575, %int2_348 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %576, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_349 = torch.constant.int 3
%int0_350 = torch.constant.int 0
%int9223372036854775807_351 = torch.constant.int 9223372036854775807
%int1_352 = torch.constant.int 1
%577 = torch.aten.slice.Tensor %576, %int3_349, %int0_350, %int9223372036854775807_351, %int1_352 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %577, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%578 = torch_c.to_builtin_tensor %552 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_353 = arith.constant 1 : index
%dim_354 = tensor.dim %578, %c1_353 : tensor<4x?x32x128xf16>
%579 = flow.tensor.bitcast %578 : tensor<4x?x32x128xf16>{%dim_354} -> tensor<4x?x32x64xcomplex<f16>>{%dim_354}
%580 = torch_c.from_builtin_tensor %579 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %580, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%581 = torch.aten.mul.Tensor %580, %577 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %581, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%582 = torch_c.to_builtin_tensor %581 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_355 = arith.constant 1 : index
%dim_356 = tensor.dim %582, %c1_355 : tensor<4x?x32x64xcomplex<f32>>
%583 = flow.tensor.bitcast %582 : tensor<4x?x32x64xcomplex<f32>>{%dim_356} -> tensor<4x?x32x128xf32>{%dim_356}
%584 = torch_c.from_builtin_tensor %583 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %584, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_357 = torch.constant.int 5
%585 = torch.prims.convert_element_type %584, %int5_357 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %585, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_358 = torch.constant.int 131072
%none_359 = torch.constant.none
%none_360 = torch.constant.none
%cpu_361 = torch.constant.device "cpu"
%false_362 = torch.constant.bool false
%586 = torch.aten.arange %int131072_358, %none_359, %none_360, %cpu_361, %false_362 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_363 = torch.constant.int 0
%int128_364 = torch.constant.int 128
%int2_365 = torch.constant.int 2
%none_366 = torch.constant.none
%none_367 = torch.constant.none
%cpu_368 = torch.constant.device "cpu"
%false_369 = torch.constant.bool false
%587 = torch.aten.arange.start_step %int0_363, %int128_364, %int2_365, %none_366, %none_367, %cpu_368, %false_369 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_370 = torch.constant.int 0
%int0_371 = torch.constant.int 0
%int64_372 = torch.constant.int 64
%int1_373 = torch.constant.int 1
%588 = torch.aten.slice.Tensor %587, %int0_370, %int0_371, %int64_372, %int1_373 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_374 = torch.constant.int 6
%589 = torch.prims.convert_element_type %588, %int6_374 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_375 = torch.constant.int 128
%590 = torch.aten.div.Scalar %589, %int128_375 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_376 = torch.constant.float 5.000000e+05
%591 = torch.aten.pow.Scalar %float5.000000e05_376, %590 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%592 = torch.aten.reciprocal %591 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_377 = torch.constant.float 1.000000e+00
%593 = torch.aten.mul.Scalar %592, %float1.000000e00_377 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_378 = torch.constant.int 131072
%int1_379 = torch.constant.int 1
%594 = torch.prim.ListConstruct %int131072_378, %int1_379 : (!torch.int, !torch.int) -> !torch.list<int>
%595 = torch.aten.view %586, %594 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%596 = torch.aten.mul.Tensor %595, %593 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%597 = torch.aten.cos %596 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%598 = torch.aten.sin %596 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%599 = torch.aten.complex %597, %598 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_380 = torch.constant.int 1
%600 = torch.aten.size.int %543, %int1_380 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_381 = torch.constant.int 0
%601 = torch.aten.add.int %int0_381, %600 : !torch.int, !torch.int -> !torch.int
%int0_382 = torch.constant.int 0
%int0_383 = torch.constant.int 0
%int1_384 = torch.constant.int 1
%602 = torch.aten.slice.Tensor %599, %int0_382, %int0_383, %601, %int1_384 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %602, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_385 = torch.constant.int 1
%int0_386 = torch.constant.int 0
%int9223372036854775807_387 = torch.constant.int 9223372036854775807
%int1_388 = torch.constant.int 1
%603 = torch.aten.slice.Tensor %602, %int1_385, %int0_386, %int9223372036854775807_387, %int1_388 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %603, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_389 = torch.constant.int 0
%604 = torch.aten.unsqueeze %603, %int0_389 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %604, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_390 = torch.constant.int 2
%605 = torch.aten.unsqueeze %604, %int2_390 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %605, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_391 = torch.constant.int 3
%int0_392 = torch.constant.int 0
%int9223372036854775807_393 = torch.constant.int 9223372036854775807
%int1_394 = torch.constant.int 1
%606 = torch.aten.slice.Tensor %605, %int3_391, %int0_392, %int9223372036854775807_393, %int1_394 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %606, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%607 = torch_c.to_builtin_tensor %554 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_395 = arith.constant 1 : index
%dim_396 = tensor.dim %607, %c1_395 : tensor<4x?x8x128xf16>
%608 = flow.tensor.bitcast %607 : tensor<4x?x8x128xf16>{%dim_396} -> tensor<4x?x8x64xcomplex<f16>>{%dim_396}
%609 = torch_c.from_builtin_tensor %608 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %609, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%610 = torch.aten.mul.Tensor %609, %606 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %610, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%611 = torch_c.to_builtin_tensor %610 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_397 = arith.constant 1 : index
%dim_398 = tensor.dim %611, %c1_397 : tensor<4x?x8x64xcomplex<f32>>
%612 = flow.tensor.bitcast %611 : tensor<4x?x8x64xcomplex<f32>>{%dim_398} -> tensor<4x?x8x128xf32>{%dim_398}
%613 = torch_c.from_builtin_tensor %612 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %613, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_399 = torch.constant.int 5
%614 = torch.prims.convert_element_type %613, %int5_399 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %614, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_400 = torch.constant.int 64
%615 = torch.aten.mul.Scalar %arg2, %int64_400 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %615, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int2_401 = torch.constant.int 2
%int1_402 = torch.constant.int 1
%616 = torch.aten.add.Scalar %615, %int2_401, %int1_402 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %616, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_403 = torch.constant.int 4
%int32_404 = torch.constant.int 32
%int8_405 = torch.constant.int 8
%int128_406 = torch.constant.int 128
%617 = torch.prim.ListConstruct %int4_403, %425, %int32_404, %int8_405, %int128_406 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%618 = torch.aten.view %614, %617 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %618, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_407 = torch.constant.int 4
%619 = torch.aten.mul.int %int4_407, %425 : !torch.int, !torch.int -> !torch.int
%int32_408 = torch.constant.int 32
%int8_409 = torch.constant.int 8
%int128_410 = torch.constant.int 128
%620 = torch.prim.ListConstruct %619, %int32_408, %int8_409, %int128_410 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%621 = torch.aten.view %618, %620 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %621, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_411 = torch.constant.int 4
%622 = torch.aten.mul.int %int4_411, %425 : !torch.int, !torch.int -> !torch.int
%623 = torch.prim.ListConstruct %622 : (!torch.int) -> !torch.list<int>
%624 = torch.aten.view %616, %623 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %624, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_412 = torch.constant.int 32
%int2_413 = torch.constant.int 2
%int32_414 = torch.constant.int 32
%int8_415 = torch.constant.int 8
%int128_416 = torch.constant.int 128
%625 = torch.prim.ListConstruct %416, %int32_412, %int2_413, %int32_414, %int8_415, %int128_416 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%626 = torch.aten.view %458, %625 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %626, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_417 = torch.constant.int 32
%627 = torch.aten.mul.int %416, %int32_417 : !torch.int, !torch.int -> !torch.int
%int2_418 = torch.constant.int 2
%628 = torch.aten.mul.int %627, %int2_418 : !torch.int, !torch.int -> !torch.int
%int32_419 = torch.constant.int 32
%int8_420 = torch.constant.int 8
%int128_421 = torch.constant.int 128
%629 = torch.prim.ListConstruct %628, %int32_419, %int8_420, %int128_421 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%630 = torch.aten.view %626, %629 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %630, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%631 = torch.prim.ListConstruct %624 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_422 = torch.constant.bool false
%632 = torch.aten.index_put %630, %631, %621, %false_422 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %632, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_423 = torch.constant.int 32
%int2_424 = torch.constant.int 2
%int32_425 = torch.constant.int 32
%int8_426 = torch.constant.int 8
%int128_427 = torch.constant.int 128
%633 = torch.prim.ListConstruct %416, %int32_423, %int2_424, %int32_425, %int8_426, %int128_427 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%634 = torch.aten.view %632, %633 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %634, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_428 = torch.constant.int 2097152
%635 = torch.prim.ListConstruct %416, %int2097152_428 : (!torch.int, !torch.int) -> !torch.list<int>
%636 = torch.aten.view %634, %635 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %636, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_429 = torch.constant.int 32
%int2_430 = torch.constant.int 2
%int32_431 = torch.constant.int 32
%int8_432 = torch.constant.int 8
%int128_433 = torch.constant.int 128
%637 = torch.prim.ListConstruct %416, %int32_429, %int2_430, %int32_431, %int8_432, %int128_433 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%638 = torch.aten.view %636, %637 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %638, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_434 = torch.constant.int 32
%int8_435 = torch.constant.int 8
%int128_436 = torch.constant.int 128
%639 = torch.prim.ListConstruct %628, %int32_434, %int8_435, %int128_436 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%640 = torch.aten.view %638, %639 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %640, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_437 = torch.constant.int 4
%int32_438 = torch.constant.int 32
%int8_439 = torch.constant.int 8
%int128_440 = torch.constant.int 128
%641 = torch.prim.ListConstruct %int4_437, %425, %int32_438, %int8_439, %int128_440 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%642 = torch.aten.view %556, %641 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %642, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_441 = torch.constant.int 4
%643 = torch.aten.mul.int %int4_441, %425 : !torch.int, !torch.int -> !torch.int
%int32_442 = torch.constant.int 32
%int8_443 = torch.constant.int 8
%int128_444 = torch.constant.int 128
%644 = torch.prim.ListConstruct %643, %int32_442, %int8_443, %int128_444 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%645 = torch.aten.view %642, %644 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %645, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_445 = torch.constant.int 1
%int1_446 = torch.constant.int 1
%646 = torch.aten.add.Scalar %616, %int1_445, %int1_446 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %646, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_447 = torch.constant.int 4
%647 = torch.aten.mul.int %int4_447, %425 : !torch.int, !torch.int -> !torch.int
%648 = torch.prim.ListConstruct %647 : (!torch.int) -> !torch.list<int>
%649 = torch.aten.view %646, %648 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %649, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%650 = torch.prim.ListConstruct %649 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_448 = torch.constant.bool false
%651 = torch.aten.index_put %640, %650, %645, %false_448 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %651, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_449 = torch.constant.int 32
%int2_450 = torch.constant.int 2
%int32_451 = torch.constant.int 32
%int8_452 = torch.constant.int 8
%int128_453 = torch.constant.int 128
%652 = torch.prim.ListConstruct %416, %int32_449, %int2_450, %int32_451, %int8_452, %int128_453 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%653 = torch.aten.view %651, %652 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %653, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_454 = torch.constant.int 2097152
%654 = torch.prim.ListConstruct %416, %int2097152_454 : (!torch.int, !torch.int) -> !torch.list<int>
%655 = torch.aten.view %653, %654 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %655, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_455 = torch.constant.int -2
%656 = torch.aten.unsqueeze %614, %int-2_455 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %656, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_456 = torch.constant.int 4
%int8_457 = torch.constant.int 8
%int4_458 = torch.constant.int 4
%int128_459 = torch.constant.int 128
%657 = torch.prim.ListConstruct %int4_456, %600, %int8_457, %int4_458, %int128_459 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_460 = torch.constant.bool false
%658 = torch.aten.expand %656, %657, %false_460 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %658, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_461 = torch.constant.int 0
%659 = torch.aten.clone %658, %int0_461 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %659, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_462 = torch.constant.int 4
%int32_463 = torch.constant.int 32
%int128_464 = torch.constant.int 128
%660 = torch.prim.ListConstruct %int4_462, %600, %int32_463, %int128_464 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%661 = torch.aten._unsafe_view %659, %660 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %661, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_465 = torch.constant.int -2
%662 = torch.aten.unsqueeze %556, %int-2_465 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %662, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_466 = torch.constant.int 1
%663 = torch.aten.size.int %550, %int1_466 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_467 = torch.constant.int 4
%int8_468 = torch.constant.int 8
%int4_469 = torch.constant.int 4
%int128_470 = torch.constant.int 128
%664 = torch.prim.ListConstruct %int4_467, %663, %int8_468, %int4_469, %int128_470 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_471 = torch.constant.bool false
%665 = torch.aten.expand %662, %664, %false_471 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %665, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_472 = torch.constant.int 0
%666 = torch.aten.clone %665, %int0_472 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %666, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_473 = torch.constant.int 4
%int32_474 = torch.constant.int 32
%int128_475 = torch.constant.int 128
%667 = torch.prim.ListConstruct %int4_473, %663, %int32_474, %int128_475 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%668 = torch.aten._unsafe_view %666, %667 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %668, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_476 = torch.constant.int 1
%int2_477 = torch.constant.int 2
%669 = torch.aten.transpose.int %585, %int1_476, %int2_477 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %669, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_478 = torch.constant.int 1
%int2_479 = torch.constant.int 2
%670 = torch.aten.transpose.int %661, %int1_478, %int2_479 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %670, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_480 = torch.constant.int 1
%int2_481 = torch.constant.int 2
%671 = torch.aten.transpose.int %668, %int1_480, %int2_481 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %671, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_482 = torch.constant.float 0.000000e+00
%false_483 = torch.constant.bool false
%none_484 = torch.constant.none
%672:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%669, %670, %671, %float0.000000e00_482, %false_483, %320, %none_484) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %672#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_485 = torch.constant.int 1
%int2_486 = torch.constant.int 2
%673 = torch.aten.transpose.int %672#0, %int1_485, %int2_486 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %673, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_487 = torch.constant.int 4
%int4096_488 = torch.constant.int 4096
%674 = torch.prim.ListConstruct %int4_487, %571, %int4096_488 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%675 = torch.aten.view %673, %674 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %675, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_489 = torch.constant.int -2
%int-1_490 = torch.constant.int -1
%676 = torch.aten.transpose.int %14, %int-2_489, %int-1_490 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_491 = torch.constant.int 4
%677 = torch.aten.mul.int %int4_491, %571 : !torch.int, !torch.int -> !torch.int
%int4096_492 = torch.constant.int 4096
%678 = torch.prim.ListConstruct %677, %int4096_492 : (!torch.int, !torch.int) -> !torch.list<int>
%679 = torch.aten.view %675, %678 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %679, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%680 = torch.aten.mm %679, %676 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %680, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_493 = torch.constant.int 4
%int4096_494 = torch.constant.int 4096
%681 = torch.prim.ListConstruct %int4_493, %571, %int4096_494 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%682 = torch.aten.view %680, %681 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %682, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_495 = torch.constant.int 1
%683 = torch.aten.add.Tensor %520, %682, %int1_495 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %683, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_496 = torch.constant.int 6
%684 = torch.prims.convert_element_type %683, %int6_496 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %684, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_497 = torch.constant.int 2
%685 = torch.aten.pow.Tensor_Scalar %684, %int2_497 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %685, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_498 = torch.constant.int -1
%686 = torch.prim.ListConstruct %int-1_498 : (!torch.int) -> !torch.list<int>
%true_499 = torch.constant.bool true
%none_500 = torch.constant.none
%687 = torch.aten.mean.dim %685, %686, %true_499, %none_500 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %687, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_501 = torch.constant.float 9.9999997473787516E-6
%int1_502 = torch.constant.int 1
%688 = torch.aten.add.Scalar %687, %float9.999990e-06_501, %int1_502 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %688, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%689 = torch.aten.rsqrt %688 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %689, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%690 = torch.aten.mul.Tensor %684, %689 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %690, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%691 = torch.aten.mul.Tensor %15, %690 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %691, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_503 = torch.constant.int 5
%692 = torch.prims.convert_element_type %691, %int5_503 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %692, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_504 = torch.constant.int -2
%int-1_505 = torch.constant.int -1
%693 = torch.aten.transpose.int %16, %int-2_504, %int-1_505 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_506 = torch.constant.int 4
%694 = torch.aten.mul.int %int4_506, %294 : !torch.int, !torch.int -> !torch.int
%int4096_507 = torch.constant.int 4096
%695 = torch.prim.ListConstruct %694, %int4096_507 : (!torch.int, !torch.int) -> !torch.list<int>
%696 = torch.aten.view %692, %695 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %696, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%697 = torch.aten.mm %696, %693 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %697, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_508 = torch.constant.int 4
%int14336_509 = torch.constant.int 14336
%698 = torch.prim.ListConstruct %int4_508, %294, %int14336_509 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%699 = torch.aten.view %697, %698 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %699, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%700 = torch.aten.silu %699 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %700, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_510 = torch.constant.int -2
%int-1_511 = torch.constant.int -1
%701 = torch.aten.transpose.int %17, %int-2_510, %int-1_511 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_512 = torch.constant.int 4
%702 = torch.aten.mul.int %int4_512, %294 : !torch.int, !torch.int -> !torch.int
%int4096_513 = torch.constant.int 4096
%703 = torch.prim.ListConstruct %702, %int4096_513 : (!torch.int, !torch.int) -> !torch.list<int>
%704 = torch.aten.view %692, %703 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %704, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%705 = torch.aten.mm %704, %701 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %705, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_514 = torch.constant.int 4
%int14336_515 = torch.constant.int 14336
%706 = torch.prim.ListConstruct %int4_514, %294, %int14336_515 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%707 = torch.aten.view %705, %706 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %707, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%708 = torch.aten.mul.Tensor %700, %707 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %708, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_516 = torch.constant.int -2
%int-1_517 = torch.constant.int -1
%709 = torch.aten.transpose.int %18, %int-2_516, %int-1_517 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_518 = torch.constant.int 1
%710 = torch.aten.size.int %699, %int1_518 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_519 = torch.constant.int 4
%711 = torch.aten.mul.int %int4_519, %710 : !torch.int, !torch.int -> !torch.int
%int14336_520 = torch.constant.int 14336
%712 = torch.prim.ListConstruct %711, %int14336_520 : (!torch.int, !torch.int) -> !torch.list<int>
%713 = torch.aten.view %708, %712 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %713, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%714 = torch.aten.mm %713, %709 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %714, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_521 = torch.constant.int 4
%int4096_522 = torch.constant.int 4096
%715 = torch.prim.ListConstruct %int4_521, %710, %int4096_522 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%716 = torch.aten.view %714, %715 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %716, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_523 = torch.constant.int 1
%717 = torch.aten.add.Tensor %683, %716, %int1_523 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %717, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_524 = torch.constant.int 6
%718 = torch.prims.convert_element_type %717, %int6_524 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %718, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_525 = torch.constant.int 2
%719 = torch.aten.pow.Tensor_Scalar %718, %int2_525 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %719, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_526 = torch.constant.int -1
%720 = torch.prim.ListConstruct %int-1_526 : (!torch.int) -> !torch.list<int>
%true_527 = torch.constant.bool true
%none_528 = torch.constant.none
%721 = torch.aten.mean.dim %719, %720, %true_527, %none_528 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %721, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_529 = torch.constant.float 9.9999997473787516E-6
%int1_530 = torch.constant.int 1
%722 = torch.aten.add.Scalar %721, %float9.999990e-06_529, %int1_530 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %722, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%723 = torch.aten.rsqrt %722 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %723, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%724 = torch.aten.mul.Tensor %718, %723 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %724, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%725 = torch.aten.mul.Tensor %19, %724 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %725, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_531 = torch.constant.int 5
%726 = torch.prims.convert_element_type %725, %int5_531 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %726, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_532 = torch.constant.int -2
%int-1_533 = torch.constant.int -1
%727 = torch.aten.transpose.int %20, %int-2_532, %int-1_533 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_534 = torch.constant.int 4
%728 = torch.aten.mul.int %int4_534, %294 : !torch.int, !torch.int -> !torch.int
%int4096_535 = torch.constant.int 4096
%729 = torch.prim.ListConstruct %728, %int4096_535 : (!torch.int, !torch.int) -> !torch.list<int>
%730 = torch.aten.view %726, %729 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %730, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%731 = torch.aten.mm %730, %727 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %731, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_536 = torch.constant.int 4
%int4096_537 = torch.constant.int 4096
%732 = torch.prim.ListConstruct %int4_536, %294, %int4096_537 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%733 = torch.aten.view %731, %732 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %733, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_538 = torch.constant.int -2
%int-1_539 = torch.constant.int -1
%734 = torch.aten.transpose.int %21, %int-2_538, %int-1_539 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_540 = torch.constant.int 4
%735 = torch.aten.mul.int %int4_540, %294 : !torch.int, !torch.int -> !torch.int
%int4096_541 = torch.constant.int 4096
%736 = torch.prim.ListConstruct %735, %int4096_541 : (!torch.int, !torch.int) -> !torch.list<int>
%737 = torch.aten.view %726, %736 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %737, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%738 = torch.aten.mm %737, %734 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %738, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_542 = torch.constant.int 4
%int1024_543 = torch.constant.int 1024
%739 = torch.prim.ListConstruct %int4_542, %294, %int1024_543 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%740 = torch.aten.view %738, %739 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %740, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_544 = torch.constant.int -2
%int-1_545 = torch.constant.int -1
%741 = torch.aten.transpose.int %22, %int-2_544, %int-1_545 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_546 = torch.constant.int 4
%742 = torch.aten.mul.int %int4_546, %294 : !torch.int, !torch.int -> !torch.int
%int4096_547 = torch.constant.int 4096
%743 = torch.prim.ListConstruct %742, %int4096_547 : (!torch.int, !torch.int) -> !torch.list<int>
%744 = torch.aten.view %726, %743 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %744, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%745 = torch.aten.mm %744, %741 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %745, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_548 = torch.constant.int 4
%int1024_549 = torch.constant.int 1024
%746 = torch.prim.ListConstruct %int4_548, %294, %int1024_549 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%747 = torch.aten.view %745, %746 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %747, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_550 = torch.constant.int 4
%int32_551 = torch.constant.int 32
%int128_552 = torch.constant.int 128
%748 = torch.prim.ListConstruct %int4_550, %294, %int32_551, %int128_552 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%749 = torch.aten.view %733, %748 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %749, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_553 = torch.constant.int 4
%int8_554 = torch.constant.int 8
%int128_555 = torch.constant.int 128
%750 = torch.prim.ListConstruct %int4_553, %294, %int8_554, %int128_555 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%751 = torch.aten.view %740, %750 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %751, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_556 = torch.constant.int 4
%int8_557 = torch.constant.int 8
%int128_558 = torch.constant.int 128
%752 = torch.prim.ListConstruct %int4_556, %294, %int8_557, %int128_558 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%753 = torch.aten.view %747, %752 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %753, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_559 = torch.constant.int 131072
%none_560 = torch.constant.none
%none_561 = torch.constant.none
%cpu_562 = torch.constant.device "cpu"
%false_563 = torch.constant.bool false
%754 = torch.aten.arange %int131072_559, %none_560, %none_561, %cpu_562, %false_563 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_564 = torch.constant.int 0
%int128_565 = torch.constant.int 128
%int2_566 = torch.constant.int 2
%none_567 = torch.constant.none
%none_568 = torch.constant.none
%cpu_569 = torch.constant.device "cpu"
%false_570 = torch.constant.bool false
%755 = torch.aten.arange.start_step %int0_564, %int128_565, %int2_566, %none_567, %none_568, %cpu_569, %false_570 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_571 = torch.constant.int 0
%int0_572 = torch.constant.int 0
%int64_573 = torch.constant.int 64
%int1_574 = torch.constant.int 1
%756 = torch.aten.slice.Tensor %755, %int0_571, %int0_572, %int64_573, %int1_574 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_575 = torch.constant.int 6
%757 = torch.prims.convert_element_type %756, %int6_575 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_576 = torch.constant.int 128
%758 = torch.aten.div.Scalar %757, %int128_576 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_577 = torch.constant.float 5.000000e+05
%759 = torch.aten.pow.Scalar %float5.000000e05_577, %758 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%760 = torch.aten.reciprocal %759 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_578 = torch.constant.float 1.000000e+00
%761 = torch.aten.mul.Scalar %760, %float1.000000e00_578 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_579 = torch.constant.int 131072
%int1_580 = torch.constant.int 1
%762 = torch.prim.ListConstruct %int131072_579, %int1_580 : (!torch.int, !torch.int) -> !torch.list<int>
%763 = torch.aten.view %754, %762 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%764 = torch.aten.mul.Tensor %763, %761 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%765 = torch.aten.cos %764 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%766 = torch.aten.sin %764 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%767 = torch.aten.complex %765, %766 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_581 = torch.constant.int 1
%768 = torch.aten.size.int %733, %int1_581 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_582 = torch.constant.int 0
%769 = torch.aten.add.int %int0_582, %768 : !torch.int, !torch.int -> !torch.int
%int0_583 = torch.constant.int 0
%int0_584 = torch.constant.int 0
%int1_585 = torch.constant.int 1
%770 = torch.aten.slice.Tensor %767, %int0_583, %int0_584, %769, %int1_585 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %770, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_586 = torch.constant.int 1
%int0_587 = torch.constant.int 0
%int9223372036854775807_588 = torch.constant.int 9223372036854775807
%int1_589 = torch.constant.int 1
%771 = torch.aten.slice.Tensor %770, %int1_586, %int0_587, %int9223372036854775807_588, %int1_589 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %771, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_590 = torch.constant.int 0
%772 = torch.aten.unsqueeze %771, %int0_590 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %772, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_591 = torch.constant.int 2
%773 = torch.aten.unsqueeze %772, %int2_591 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %773, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_592 = torch.constant.int 3
%int0_593 = torch.constant.int 0
%int9223372036854775807_594 = torch.constant.int 9223372036854775807
%int1_595 = torch.constant.int 1
%774 = torch.aten.slice.Tensor %773, %int3_592, %int0_593, %int9223372036854775807_594, %int1_595 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %774, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%775 = torch_c.to_builtin_tensor %749 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_596 = arith.constant 1 : index
%dim_597 = tensor.dim %775, %c1_596 : tensor<4x?x32x128xf16>
%776 = flow.tensor.bitcast %775 : tensor<4x?x32x128xf16>{%dim_597} -> tensor<4x?x32x64xcomplex<f16>>{%dim_597}
%777 = torch_c.from_builtin_tensor %776 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %777, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%778 = torch.aten.mul.Tensor %777, %774 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %778, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%779 = torch_c.to_builtin_tensor %778 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_598 = arith.constant 1 : index
%dim_599 = tensor.dim %779, %c1_598 : tensor<4x?x32x64xcomplex<f32>>
%780 = flow.tensor.bitcast %779 : tensor<4x?x32x64xcomplex<f32>>{%dim_599} -> tensor<4x?x32x128xf32>{%dim_599}
%781 = torch_c.from_builtin_tensor %780 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %781, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_600 = torch.constant.int 5
%782 = torch.prims.convert_element_type %781, %int5_600 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %782, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_601 = torch.constant.int 131072
%none_602 = torch.constant.none
%none_603 = torch.constant.none
%cpu_604 = torch.constant.device "cpu"
%false_605 = torch.constant.bool false
%783 = torch.aten.arange %int131072_601, %none_602, %none_603, %cpu_604, %false_605 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_606 = torch.constant.int 0
%int128_607 = torch.constant.int 128
%int2_608 = torch.constant.int 2
%none_609 = torch.constant.none
%none_610 = torch.constant.none
%cpu_611 = torch.constant.device "cpu"
%false_612 = torch.constant.bool false
%784 = torch.aten.arange.start_step %int0_606, %int128_607, %int2_608, %none_609, %none_610, %cpu_611, %false_612 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_613 = torch.constant.int 0
%int0_614 = torch.constant.int 0
%int64_615 = torch.constant.int 64
%int1_616 = torch.constant.int 1
%785 = torch.aten.slice.Tensor %784, %int0_613, %int0_614, %int64_615, %int1_616 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_617 = torch.constant.int 6
%786 = torch.prims.convert_element_type %785, %int6_617 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_618 = torch.constant.int 128
%787 = torch.aten.div.Scalar %786, %int128_618 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_619 = torch.constant.float 5.000000e+05
%788 = torch.aten.pow.Scalar %float5.000000e05_619, %787 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%789 = torch.aten.reciprocal %788 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_620 = torch.constant.float 1.000000e+00
%790 = torch.aten.mul.Scalar %789, %float1.000000e00_620 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_621 = torch.constant.int 131072
%int1_622 = torch.constant.int 1
%791 = torch.prim.ListConstruct %int131072_621, %int1_622 : (!torch.int, !torch.int) -> !torch.list<int>
%792 = torch.aten.view %783, %791 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%793 = torch.aten.mul.Tensor %792, %790 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%794 = torch.aten.cos %793 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%795 = torch.aten.sin %793 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%796 = torch.aten.complex %794, %795 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_623 = torch.constant.int 1
%797 = torch.aten.size.int %740, %int1_623 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_624 = torch.constant.int 0
%798 = torch.aten.add.int %int0_624, %797 : !torch.int, !torch.int -> !torch.int
%int0_625 = torch.constant.int 0
%int0_626 = torch.constant.int 0
%int1_627 = torch.constant.int 1
%799 = torch.aten.slice.Tensor %796, %int0_625, %int0_626, %798, %int1_627 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %799, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_628 = torch.constant.int 1
%int0_629 = torch.constant.int 0
%int9223372036854775807_630 = torch.constant.int 9223372036854775807
%int1_631 = torch.constant.int 1
%800 = torch.aten.slice.Tensor %799, %int1_628, %int0_629, %int9223372036854775807_630, %int1_631 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %800, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_632 = torch.constant.int 0
%801 = torch.aten.unsqueeze %800, %int0_632 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %801, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_633 = torch.constant.int 2
%802 = torch.aten.unsqueeze %801, %int2_633 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %802, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_634 = torch.constant.int 3
%int0_635 = torch.constant.int 0
%int9223372036854775807_636 = torch.constant.int 9223372036854775807
%int1_637 = torch.constant.int 1
%803 = torch.aten.slice.Tensor %802, %int3_634, %int0_635, %int9223372036854775807_636, %int1_637 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %803, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%804 = torch_c.to_builtin_tensor %751 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_638 = arith.constant 1 : index
%dim_639 = tensor.dim %804, %c1_638 : tensor<4x?x8x128xf16>
%805 = flow.tensor.bitcast %804 : tensor<4x?x8x128xf16>{%dim_639} -> tensor<4x?x8x64xcomplex<f16>>{%dim_639}
%806 = torch_c.from_builtin_tensor %805 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %806, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%807 = torch.aten.mul.Tensor %806, %803 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %807, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%808 = torch_c.to_builtin_tensor %807 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_640 = arith.constant 1 : index
%dim_641 = tensor.dim %808, %c1_640 : tensor<4x?x8x64xcomplex<f32>>
%809 = flow.tensor.bitcast %808 : tensor<4x?x8x64xcomplex<f32>>{%dim_641} -> tensor<4x?x8x128xf32>{%dim_641}
%810 = torch_c.from_builtin_tensor %809 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %810, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_642 = torch.constant.int 5
%811 = torch.prims.convert_element_type %810, %int5_642 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %811, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_643 = torch.constant.int 64
%812 = torch.aten.mul.Scalar %arg2, %int64_643 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %812, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_644 = torch.constant.int 4
%int1_645 = torch.constant.int 1
%813 = torch.aten.add.Scalar %812, %int4_644, %int1_645 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %813, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_646 = torch.constant.int 4
%int32_647 = torch.constant.int 32
%int8_648 = torch.constant.int 8
%int128_649 = torch.constant.int 128
%814 = torch.prim.ListConstruct %int4_646, %425, %int32_647, %int8_648, %int128_649 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%815 = torch.aten.view %811, %814 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %815, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_650 = torch.constant.int 4
%816 = torch.aten.mul.int %int4_650, %425 : !torch.int, !torch.int -> !torch.int
%int32_651 = torch.constant.int 32
%int8_652 = torch.constant.int 8
%int128_653 = torch.constant.int 128
%817 = torch.prim.ListConstruct %816, %int32_651, %int8_652, %int128_653 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%818 = torch.aten.view %815, %817 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %818, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_654 = torch.constant.int 4
%819 = torch.aten.mul.int %int4_654, %425 : !torch.int, !torch.int -> !torch.int
%820 = torch.prim.ListConstruct %819 : (!torch.int) -> !torch.list<int>
%821 = torch.aten.view %813, %820 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %821, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_655 = torch.constant.int 32
%int2_656 = torch.constant.int 2
%int32_657 = torch.constant.int 32
%int8_658 = torch.constant.int 8
%int128_659 = torch.constant.int 128
%822 = torch.prim.ListConstruct %416, %int32_655, %int2_656, %int32_657, %int8_658, %int128_659 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%823 = torch.aten.view %655, %822 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %823, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_660 = torch.constant.int 32
%824 = torch.aten.mul.int %416, %int32_660 : !torch.int, !torch.int -> !torch.int
%int2_661 = torch.constant.int 2
%825 = torch.aten.mul.int %824, %int2_661 : !torch.int, !torch.int -> !torch.int
%int32_662 = torch.constant.int 32
%int8_663 = torch.constant.int 8
%int128_664 = torch.constant.int 128
%826 = torch.prim.ListConstruct %825, %int32_662, %int8_663, %int128_664 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%827 = torch.aten.view %823, %826 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %827, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%828 = torch.prim.ListConstruct %821 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_665 = torch.constant.bool false
%829 = torch.aten.index_put %827, %828, %818, %false_665 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %829, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_666 = torch.constant.int 32
%int2_667 = torch.constant.int 2
%int32_668 = torch.constant.int 32
%int8_669 = torch.constant.int 8
%int128_670 = torch.constant.int 128
%830 = torch.prim.ListConstruct %416, %int32_666, %int2_667, %int32_668, %int8_669, %int128_670 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%831 = torch.aten.view %829, %830 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %831, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_671 = torch.constant.int 2097152
%832 = torch.prim.ListConstruct %416, %int2097152_671 : (!torch.int, !torch.int) -> !torch.list<int>
%833 = torch.aten.view %831, %832 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %833, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_672 = torch.constant.int 32
%int2_673 = torch.constant.int 2
%int32_674 = torch.constant.int 32
%int8_675 = torch.constant.int 8
%int128_676 = torch.constant.int 128
%834 = torch.prim.ListConstruct %416, %int32_672, %int2_673, %int32_674, %int8_675, %int128_676 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%835 = torch.aten.view %833, %834 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %835, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_677 = torch.constant.int 32
%int8_678 = torch.constant.int 8
%int128_679 = torch.constant.int 128
%836 = torch.prim.ListConstruct %825, %int32_677, %int8_678, %int128_679 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%837 = torch.aten.view %835, %836 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %837, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_680 = torch.constant.int 4
%int32_681 = torch.constant.int 32
%int8_682 = torch.constant.int 8
%int128_683 = torch.constant.int 128
%838 = torch.prim.ListConstruct %int4_680, %425, %int32_681, %int8_682, %int128_683 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%839 = torch.aten.view %753, %838 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %839, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_684 = torch.constant.int 4
%840 = torch.aten.mul.int %int4_684, %425 : !torch.int, !torch.int -> !torch.int
%int32_685 = torch.constant.int 32
%int8_686 = torch.constant.int 8
%int128_687 = torch.constant.int 128
%841 = torch.prim.ListConstruct %840, %int32_685, %int8_686, %int128_687 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%842 = torch.aten.view %839, %841 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %842, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_688 = torch.constant.int 1
%int1_689 = torch.constant.int 1
%843 = torch.aten.add.Scalar %813, %int1_688, %int1_689 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %843, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_690 = torch.constant.int 4
%844 = torch.aten.mul.int %int4_690, %425 : !torch.int, !torch.int -> !torch.int
%845 = torch.prim.ListConstruct %844 : (!torch.int) -> !torch.list<int>
%846 = torch.aten.view %843, %845 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %846, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%847 = torch.prim.ListConstruct %846 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_691 = torch.constant.bool false
%848 = torch.aten.index_put %837, %847, %842, %false_691 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %848, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_692 = torch.constant.int 32
%int2_693 = torch.constant.int 2
%int32_694 = torch.constant.int 32
%int8_695 = torch.constant.int 8
%int128_696 = torch.constant.int 128
%849 = torch.prim.ListConstruct %416, %int32_692, %int2_693, %int32_694, %int8_695, %int128_696 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%850 = torch.aten.view %848, %849 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %850, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_697 = torch.constant.int 2097152
%851 = torch.prim.ListConstruct %416, %int2097152_697 : (!torch.int, !torch.int) -> !torch.list<int>
%852 = torch.aten.view %850, %851 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %852, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_698 = torch.constant.int -2
%853 = torch.aten.unsqueeze %811, %int-2_698 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %853, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_699 = torch.constant.int 4
%int8_700 = torch.constant.int 8
%int4_701 = torch.constant.int 4
%int128_702 = torch.constant.int 128
%854 = torch.prim.ListConstruct %int4_699, %797, %int8_700, %int4_701, %int128_702 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_703 = torch.constant.bool false
%855 = torch.aten.expand %853, %854, %false_703 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %855, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_704 = torch.constant.int 0
%856 = torch.aten.clone %855, %int0_704 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %856, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_705 = torch.constant.int 4
%int32_706 = torch.constant.int 32
%int128_707 = torch.constant.int 128
%857 = torch.prim.ListConstruct %int4_705, %797, %int32_706, %int128_707 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%858 = torch.aten._unsafe_view %856, %857 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %858, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_708 = torch.constant.int -2
%859 = torch.aten.unsqueeze %753, %int-2_708 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %859, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_709 = torch.constant.int 1
%860 = torch.aten.size.int %747, %int1_709 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_710 = torch.constant.int 4
%int8_711 = torch.constant.int 8
%int4_712 = torch.constant.int 4
%int128_713 = torch.constant.int 128
%861 = torch.prim.ListConstruct %int4_710, %860, %int8_711, %int4_712, %int128_713 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_714 = torch.constant.bool false
%862 = torch.aten.expand %859, %861, %false_714 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %862, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_715 = torch.constant.int 0
%863 = torch.aten.clone %862, %int0_715 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %863, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_716 = torch.constant.int 4
%int32_717 = torch.constant.int 32
%int128_718 = torch.constant.int 128
%864 = torch.prim.ListConstruct %int4_716, %860, %int32_717, %int128_718 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%865 = torch.aten._unsafe_view %863, %864 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %865, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_719 = torch.constant.int 1
%int2_720 = torch.constant.int 2
%866 = torch.aten.transpose.int %782, %int1_719, %int2_720 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %866, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_721 = torch.constant.int 1
%int2_722 = torch.constant.int 2
%867 = torch.aten.transpose.int %858, %int1_721, %int2_722 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %867, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_723 = torch.constant.int 1
%int2_724 = torch.constant.int 2
%868 = torch.aten.transpose.int %865, %int1_723, %int2_724 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %868, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_725 = torch.constant.float 0.000000e+00
%false_726 = torch.constant.bool false
%none_727 = torch.constant.none
%869:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%866, %867, %868, %float0.000000e00_725, %false_726, %320, %none_727) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %869#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_728 = torch.constant.int 1
%int2_729 = torch.constant.int 2
%870 = torch.aten.transpose.int %869#0, %int1_728, %int2_729 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %870, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_730 = torch.constant.int 4
%int4096_731 = torch.constant.int 4096
%871 = torch.prim.ListConstruct %int4_730, %768, %int4096_731 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%872 = torch.aten.view %870, %871 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %872, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_732 = torch.constant.int -2
%int-1_733 = torch.constant.int -1
%873 = torch.aten.transpose.int %23, %int-2_732, %int-1_733 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_734 = torch.constant.int 4
%874 = torch.aten.mul.int %int4_734, %768 : !torch.int, !torch.int -> !torch.int
%int4096_735 = torch.constant.int 4096
%875 = torch.prim.ListConstruct %874, %int4096_735 : (!torch.int, !torch.int) -> !torch.list<int>
%876 = torch.aten.view %872, %875 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %876, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%877 = torch.aten.mm %876, %873 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %877, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_736 = torch.constant.int 4
%int4096_737 = torch.constant.int 4096
%878 = torch.prim.ListConstruct %int4_736, %768, %int4096_737 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%879 = torch.aten.view %877, %878 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %879, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_738 = torch.constant.int 1
%880 = torch.aten.add.Tensor %717, %879, %int1_738 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %880, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_739 = torch.constant.int 6
%881 = torch.prims.convert_element_type %880, %int6_739 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %881, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_740 = torch.constant.int 2
%882 = torch.aten.pow.Tensor_Scalar %881, %int2_740 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %882, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_741 = torch.constant.int -1
%883 = torch.prim.ListConstruct %int-1_741 : (!torch.int) -> !torch.list<int>
%true_742 = torch.constant.bool true
%none_743 = torch.constant.none
%884 = torch.aten.mean.dim %882, %883, %true_742, %none_743 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %884, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_744 = torch.constant.float 9.9999997473787516E-6
%int1_745 = torch.constant.int 1
%885 = torch.aten.add.Scalar %884, %float9.999990e-06_744, %int1_745 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %885, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%886 = torch.aten.rsqrt %885 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %886, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%887 = torch.aten.mul.Tensor %881, %886 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %887, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%888 = torch.aten.mul.Tensor %24, %887 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %888, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_746 = torch.constant.int 5
%889 = torch.prims.convert_element_type %888, %int5_746 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %889, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_747 = torch.constant.int -2
%int-1_748 = torch.constant.int -1
%890 = torch.aten.transpose.int %25, %int-2_747, %int-1_748 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_749 = torch.constant.int 4
%891 = torch.aten.mul.int %int4_749, %294 : !torch.int, !torch.int -> !torch.int
%int4096_750 = torch.constant.int 4096
%892 = torch.prim.ListConstruct %891, %int4096_750 : (!torch.int, !torch.int) -> !torch.list<int>
%893 = torch.aten.view %889, %892 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %893, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%894 = torch.aten.mm %893, %890 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %894, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_751 = torch.constant.int 4
%int14336_752 = torch.constant.int 14336
%895 = torch.prim.ListConstruct %int4_751, %294, %int14336_752 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%896 = torch.aten.view %894, %895 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %896, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%897 = torch.aten.silu %896 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %897, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_753 = torch.constant.int -2
%int-1_754 = torch.constant.int -1
%898 = torch.aten.transpose.int %26, %int-2_753, %int-1_754 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_755 = torch.constant.int 4
%899 = torch.aten.mul.int %int4_755, %294 : !torch.int, !torch.int -> !torch.int
%int4096_756 = torch.constant.int 4096
%900 = torch.prim.ListConstruct %899, %int4096_756 : (!torch.int, !torch.int) -> !torch.list<int>
%901 = torch.aten.view %889, %900 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %901, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%902 = torch.aten.mm %901, %898 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %902, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_757 = torch.constant.int 4
%int14336_758 = torch.constant.int 14336
%903 = torch.prim.ListConstruct %int4_757, %294, %int14336_758 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%904 = torch.aten.view %902, %903 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %904, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%905 = torch.aten.mul.Tensor %897, %904 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %905, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_759 = torch.constant.int -2
%int-1_760 = torch.constant.int -1
%906 = torch.aten.transpose.int %27, %int-2_759, %int-1_760 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_761 = torch.constant.int 1
%907 = torch.aten.size.int %896, %int1_761 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_762 = torch.constant.int 4
%908 = torch.aten.mul.int %int4_762, %907 : !torch.int, !torch.int -> !torch.int
%int14336_763 = torch.constant.int 14336
%909 = torch.prim.ListConstruct %908, %int14336_763 : (!torch.int, !torch.int) -> !torch.list<int>
%910 = torch.aten.view %905, %909 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %910, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%911 = torch.aten.mm %910, %906 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %911, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_764 = torch.constant.int 4
%int4096_765 = torch.constant.int 4096
%912 = torch.prim.ListConstruct %int4_764, %907, %int4096_765 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%913 = torch.aten.view %911, %912 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %913, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_766 = torch.constant.int 1
%914 = torch.aten.add.Tensor %880, %913, %int1_766 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %914, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_767 = torch.constant.int 6
%915 = torch.prims.convert_element_type %914, %int6_767 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %915, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_768 = torch.constant.int 2
%916 = torch.aten.pow.Tensor_Scalar %915, %int2_768 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %916, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_769 = torch.constant.int -1
%917 = torch.prim.ListConstruct %int-1_769 : (!torch.int) -> !torch.list<int>
%true_770 = torch.constant.bool true
%none_771 = torch.constant.none
%918 = torch.aten.mean.dim %916, %917, %true_770, %none_771 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %918, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_772 = torch.constant.float 9.9999997473787516E-6
%int1_773 = torch.constant.int 1
%919 = torch.aten.add.Scalar %918, %float9.999990e-06_772, %int1_773 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %919, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%920 = torch.aten.rsqrt %919 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %920, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%921 = torch.aten.mul.Tensor %915, %920 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %921, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%922 = torch.aten.mul.Tensor %28, %921 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %922, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_774 = torch.constant.int 5
%923 = torch.prims.convert_element_type %922, %int5_774 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %923, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_775 = torch.constant.int -2
%int-1_776 = torch.constant.int -1
%924 = torch.aten.transpose.int %29, %int-2_775, %int-1_776 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_777 = torch.constant.int 4
%925 = torch.aten.mul.int %int4_777, %294 : !torch.int, !torch.int -> !torch.int
%int4096_778 = torch.constant.int 4096
%926 = torch.prim.ListConstruct %925, %int4096_778 : (!torch.int, !torch.int) -> !torch.list<int>
%927 = torch.aten.view %923, %926 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %927, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%928 = torch.aten.mm %927, %924 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %928, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_779 = torch.constant.int 4
%int4096_780 = torch.constant.int 4096
%929 = torch.prim.ListConstruct %int4_779, %294, %int4096_780 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%930 = torch.aten.view %928, %929 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %930, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_781 = torch.constant.int -2
%int-1_782 = torch.constant.int -1
%931 = torch.aten.transpose.int %30, %int-2_781, %int-1_782 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_783 = torch.constant.int 4
%932 = torch.aten.mul.int %int4_783, %294 : !torch.int, !torch.int -> !torch.int
%int4096_784 = torch.constant.int 4096
%933 = torch.prim.ListConstruct %932, %int4096_784 : (!torch.int, !torch.int) -> !torch.list<int>
%934 = torch.aten.view %923, %933 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %934, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%935 = torch.aten.mm %934, %931 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %935, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_785 = torch.constant.int 4
%int1024_786 = torch.constant.int 1024
%936 = torch.prim.ListConstruct %int4_785, %294, %int1024_786 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%937 = torch.aten.view %935, %936 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %937, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_787 = torch.constant.int -2
%int-1_788 = torch.constant.int -1
%938 = torch.aten.transpose.int %31, %int-2_787, %int-1_788 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_789 = torch.constant.int 4
%939 = torch.aten.mul.int %int4_789, %294 : !torch.int, !torch.int -> !torch.int
%int4096_790 = torch.constant.int 4096
%940 = torch.prim.ListConstruct %939, %int4096_790 : (!torch.int, !torch.int) -> !torch.list<int>
%941 = torch.aten.view %923, %940 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %941, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%942 = torch.aten.mm %941, %938 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %942, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_791 = torch.constant.int 4
%int1024_792 = torch.constant.int 1024
%943 = torch.prim.ListConstruct %int4_791, %294, %int1024_792 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%944 = torch.aten.view %942, %943 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %944, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_793 = torch.constant.int 4
%int32_794 = torch.constant.int 32
%int128_795 = torch.constant.int 128
%945 = torch.prim.ListConstruct %int4_793, %294, %int32_794, %int128_795 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%946 = torch.aten.view %930, %945 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %946, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_796 = torch.constant.int 4
%int8_797 = torch.constant.int 8
%int128_798 = torch.constant.int 128
%947 = torch.prim.ListConstruct %int4_796, %294, %int8_797, %int128_798 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%948 = torch.aten.view %937, %947 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %948, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_799 = torch.constant.int 4
%int8_800 = torch.constant.int 8
%int128_801 = torch.constant.int 128
%949 = torch.prim.ListConstruct %int4_799, %294, %int8_800, %int128_801 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%950 = torch.aten.view %944, %949 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %950, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_802 = torch.constant.int 131072
%none_803 = torch.constant.none
%none_804 = torch.constant.none
%cpu_805 = torch.constant.device "cpu"
%false_806 = torch.constant.bool false
%951 = torch.aten.arange %int131072_802, %none_803, %none_804, %cpu_805, %false_806 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_807 = torch.constant.int 0
%int128_808 = torch.constant.int 128
%int2_809 = torch.constant.int 2
%none_810 = torch.constant.none
%none_811 = torch.constant.none
%cpu_812 = torch.constant.device "cpu"
%false_813 = torch.constant.bool false
%952 = torch.aten.arange.start_step %int0_807, %int128_808, %int2_809, %none_810, %none_811, %cpu_812, %false_813 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_814 = torch.constant.int 0
%int0_815 = torch.constant.int 0
%int64_816 = torch.constant.int 64
%int1_817 = torch.constant.int 1
%953 = torch.aten.slice.Tensor %952, %int0_814, %int0_815, %int64_816, %int1_817 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_818 = torch.constant.int 6
%954 = torch.prims.convert_element_type %953, %int6_818 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_819 = torch.constant.int 128
%955 = torch.aten.div.Scalar %954, %int128_819 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_820 = torch.constant.float 5.000000e+05
%956 = torch.aten.pow.Scalar %float5.000000e05_820, %955 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%957 = torch.aten.reciprocal %956 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_821 = torch.constant.float 1.000000e+00
%958 = torch.aten.mul.Scalar %957, %float1.000000e00_821 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_822 = torch.constant.int 131072
%int1_823 = torch.constant.int 1
%959 = torch.prim.ListConstruct %int131072_822, %int1_823 : (!torch.int, !torch.int) -> !torch.list<int>
%960 = torch.aten.view %951, %959 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%961 = torch.aten.mul.Tensor %960, %958 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%962 = torch.aten.cos %961 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%963 = torch.aten.sin %961 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%964 = torch.aten.complex %962, %963 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_824 = torch.constant.int 1
%965 = torch.aten.size.int %930, %int1_824 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_825 = torch.constant.int 0
%966 = torch.aten.add.int %int0_825, %965 : !torch.int, !torch.int -> !torch.int
%int0_826 = torch.constant.int 0
%int0_827 = torch.constant.int 0
%int1_828 = torch.constant.int 1
%967 = torch.aten.slice.Tensor %964, %int0_826, %int0_827, %966, %int1_828 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %967, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_829 = torch.constant.int 1
%int0_830 = torch.constant.int 0
%int9223372036854775807_831 = torch.constant.int 9223372036854775807
%int1_832 = torch.constant.int 1
%968 = torch.aten.slice.Tensor %967, %int1_829, %int0_830, %int9223372036854775807_831, %int1_832 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %968, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_833 = torch.constant.int 0
%969 = torch.aten.unsqueeze %968, %int0_833 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %969, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_834 = torch.constant.int 2
%970 = torch.aten.unsqueeze %969, %int2_834 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %970, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_835 = torch.constant.int 3
%int0_836 = torch.constant.int 0
%int9223372036854775807_837 = torch.constant.int 9223372036854775807
%int1_838 = torch.constant.int 1
%971 = torch.aten.slice.Tensor %970, %int3_835, %int0_836, %int9223372036854775807_837, %int1_838 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %971, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%972 = torch_c.to_builtin_tensor %946 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_839 = arith.constant 1 : index
%dim_840 = tensor.dim %972, %c1_839 : tensor<4x?x32x128xf16>
%973 = flow.tensor.bitcast %972 : tensor<4x?x32x128xf16>{%dim_840} -> tensor<4x?x32x64xcomplex<f16>>{%dim_840}
%974 = torch_c.from_builtin_tensor %973 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %974, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%975 = torch.aten.mul.Tensor %974, %971 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %975, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%976 = torch_c.to_builtin_tensor %975 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_841 = arith.constant 1 : index
%dim_842 = tensor.dim %976, %c1_841 : tensor<4x?x32x64xcomplex<f32>>
%977 = flow.tensor.bitcast %976 : tensor<4x?x32x64xcomplex<f32>>{%dim_842} -> tensor<4x?x32x128xf32>{%dim_842}
%978 = torch_c.from_builtin_tensor %977 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %978, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_843 = torch.constant.int 5
%979 = torch.prims.convert_element_type %978, %int5_843 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %979, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_844 = torch.constant.int 131072
%none_845 = torch.constant.none
%none_846 = torch.constant.none
%cpu_847 = torch.constant.device "cpu"
%false_848 = torch.constant.bool false
%980 = torch.aten.arange %int131072_844, %none_845, %none_846, %cpu_847, %false_848 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_849 = torch.constant.int 0
%int128_850 = torch.constant.int 128
%int2_851 = torch.constant.int 2
%none_852 = torch.constant.none
%none_853 = torch.constant.none
%cpu_854 = torch.constant.device "cpu"
%false_855 = torch.constant.bool false
%981 = torch.aten.arange.start_step %int0_849, %int128_850, %int2_851, %none_852, %none_853, %cpu_854, %false_855 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_856 = torch.constant.int 0
%int0_857 = torch.constant.int 0
%int64_858 = torch.constant.int 64
%int1_859 = torch.constant.int 1
%982 = torch.aten.slice.Tensor %981, %int0_856, %int0_857, %int64_858, %int1_859 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_860 = torch.constant.int 6
%983 = torch.prims.convert_element_type %982, %int6_860 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_861 = torch.constant.int 128
%984 = torch.aten.div.Scalar %983, %int128_861 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_862 = torch.constant.float 5.000000e+05
%985 = torch.aten.pow.Scalar %float5.000000e05_862, %984 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%986 = torch.aten.reciprocal %985 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_863 = torch.constant.float 1.000000e+00
%987 = torch.aten.mul.Scalar %986, %float1.000000e00_863 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_864 = torch.constant.int 131072
%int1_865 = torch.constant.int 1
%988 = torch.prim.ListConstruct %int131072_864, %int1_865 : (!torch.int, !torch.int) -> !torch.list<int>
%989 = torch.aten.view %980, %988 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%990 = torch.aten.mul.Tensor %989, %987 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%991 = torch.aten.cos %990 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%992 = torch.aten.sin %990 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%993 = torch.aten.complex %991, %992 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_866 = torch.constant.int 1
%994 = torch.aten.size.int %937, %int1_866 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_867 = torch.constant.int 0
%995 = torch.aten.add.int %int0_867, %994 : !torch.int, !torch.int -> !torch.int
%int0_868 = torch.constant.int 0
%int0_869 = torch.constant.int 0
%int1_870 = torch.constant.int 1
%996 = torch.aten.slice.Tensor %993, %int0_868, %int0_869, %995, %int1_870 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %996, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_871 = torch.constant.int 1
%int0_872 = torch.constant.int 0
%int9223372036854775807_873 = torch.constant.int 9223372036854775807
%int1_874 = torch.constant.int 1
%997 = torch.aten.slice.Tensor %996, %int1_871, %int0_872, %int9223372036854775807_873, %int1_874 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %997, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_875 = torch.constant.int 0
%998 = torch.aten.unsqueeze %997, %int0_875 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %998, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_876 = torch.constant.int 2
%999 = torch.aten.unsqueeze %998, %int2_876 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %999, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_877 = torch.constant.int 3
%int0_878 = torch.constant.int 0
%int9223372036854775807_879 = torch.constant.int 9223372036854775807
%int1_880 = torch.constant.int 1
%1000 = torch.aten.slice.Tensor %999, %int3_877, %int0_878, %int9223372036854775807_879, %int1_880 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1000, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1001 = torch_c.to_builtin_tensor %948 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_881 = arith.constant 1 : index
%dim_882 = tensor.dim %1001, %c1_881 : tensor<4x?x8x128xf16>
%1002 = flow.tensor.bitcast %1001 : tensor<4x?x8x128xf16>{%dim_882} -> tensor<4x?x8x64xcomplex<f16>>{%dim_882}
%1003 = torch_c.from_builtin_tensor %1002 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1003, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1004 = torch.aten.mul.Tensor %1003, %1000 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1004, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1005 = torch_c.to_builtin_tensor %1004 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_883 = arith.constant 1 : index
%dim_884 = tensor.dim %1005, %c1_883 : tensor<4x?x8x64xcomplex<f32>>
%1006 = flow.tensor.bitcast %1005 : tensor<4x?x8x64xcomplex<f32>>{%dim_884} -> tensor<4x?x8x128xf32>{%dim_884}
%1007 = torch_c.from_builtin_tensor %1006 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1007, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_885 = torch.constant.int 5
%1008 = torch.prims.convert_element_type %1007, %int5_885 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1008, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_886 = torch.constant.int 64
%1009 = torch.aten.mul.Scalar %arg2, %int64_886 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1009, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int6_887 = torch.constant.int 6
%int1_888 = torch.constant.int 1
%1010 = torch.aten.add.Scalar %1009, %int6_887, %int1_888 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1010, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_889 = torch.constant.int 4
%int32_890 = torch.constant.int 32
%int8_891 = torch.constant.int 8
%int128_892 = torch.constant.int 128
%1011 = torch.prim.ListConstruct %int4_889, %425, %int32_890, %int8_891, %int128_892 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1012 = torch.aten.view %1008, %1011 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1012, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_893 = torch.constant.int 4
%1013 = torch.aten.mul.int %int4_893, %425 : !torch.int, !torch.int -> !torch.int
%int32_894 = torch.constant.int 32
%int8_895 = torch.constant.int 8
%int128_896 = torch.constant.int 128
%1014 = torch.prim.ListConstruct %1013, %int32_894, %int8_895, %int128_896 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1015 = torch.aten.view %1012, %1014 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1015, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_897 = torch.constant.int 4
%1016 = torch.aten.mul.int %int4_897, %425 : !torch.int, !torch.int -> !torch.int
%1017 = torch.prim.ListConstruct %1016 : (!torch.int) -> !torch.list<int>
%1018 = torch.aten.view %1010, %1017 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1018, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_898 = torch.constant.int 32
%int2_899 = torch.constant.int 2
%int32_900 = torch.constant.int 32
%int8_901 = torch.constant.int 8
%int128_902 = torch.constant.int 128
%1019 = torch.prim.ListConstruct %416, %int32_898, %int2_899, %int32_900, %int8_901, %int128_902 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1020 = torch.aten.view %852, %1019 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1020, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_903 = torch.constant.int 32
%1021 = torch.aten.mul.int %416, %int32_903 : !torch.int, !torch.int -> !torch.int
%int2_904 = torch.constant.int 2
%1022 = torch.aten.mul.int %1021, %int2_904 : !torch.int, !torch.int -> !torch.int
%int32_905 = torch.constant.int 32
%int8_906 = torch.constant.int 8
%int128_907 = torch.constant.int 128
%1023 = torch.prim.ListConstruct %1022, %int32_905, %int8_906, %int128_907 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1024 = torch.aten.view %1020, %1023 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1024, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%1025 = torch.prim.ListConstruct %1018 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_908 = torch.constant.bool false
%1026 = torch.aten.index_put %1024, %1025, %1015, %false_908 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1026, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_909 = torch.constant.int 32
%int2_910 = torch.constant.int 2
%int32_911 = torch.constant.int 32
%int8_912 = torch.constant.int 8
%int128_913 = torch.constant.int 128
%1027 = torch.prim.ListConstruct %416, %int32_909, %int2_910, %int32_911, %int8_912, %int128_913 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1028 = torch.aten.view %1026, %1027 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1028, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_914 = torch.constant.int 2097152
%1029 = torch.prim.ListConstruct %416, %int2097152_914 : (!torch.int, !torch.int) -> !torch.list<int>
%1030 = torch.aten.view %1028, %1029 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1030, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_915 = torch.constant.int 32
%int2_916 = torch.constant.int 2
%int32_917 = torch.constant.int 32
%int8_918 = torch.constant.int 8
%int128_919 = torch.constant.int 128
%1031 = torch.prim.ListConstruct %416, %int32_915, %int2_916, %int32_917, %int8_918, %int128_919 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1032 = torch.aten.view %1030, %1031 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1032, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_920 = torch.constant.int 32
%int8_921 = torch.constant.int 8
%int128_922 = torch.constant.int 128
%1033 = torch.prim.ListConstruct %1022, %int32_920, %int8_921, %int128_922 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1034 = torch.aten.view %1032, %1033 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1034, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_923 = torch.constant.int 4
%int32_924 = torch.constant.int 32
%int8_925 = torch.constant.int 8
%int128_926 = torch.constant.int 128
%1035 = torch.prim.ListConstruct %int4_923, %425, %int32_924, %int8_925, %int128_926 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1036 = torch.aten.view %950, %1035 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1036, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_927 = torch.constant.int 4
%1037 = torch.aten.mul.int %int4_927, %425 : !torch.int, !torch.int -> !torch.int
%int32_928 = torch.constant.int 32
%int8_929 = torch.constant.int 8
%int128_930 = torch.constant.int 128
%1038 = torch.prim.ListConstruct %1037, %int32_928, %int8_929, %int128_930 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1039 = torch.aten.view %1036, %1038 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1039, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_931 = torch.constant.int 1
%int1_932 = torch.constant.int 1
%1040 = torch.aten.add.Scalar %1010, %int1_931, %int1_932 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1040, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_933 = torch.constant.int 4
%1041 = torch.aten.mul.int %int4_933, %425 : !torch.int, !torch.int -> !torch.int
%1042 = torch.prim.ListConstruct %1041 : (!torch.int) -> !torch.list<int>
%1043 = torch.aten.view %1040, %1042 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1043, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%1044 = torch.prim.ListConstruct %1043 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_934 = torch.constant.bool false
%1045 = torch.aten.index_put %1034, %1044, %1039, %false_934 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1045, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_935 = torch.constant.int 32
%int2_936 = torch.constant.int 2
%int32_937 = torch.constant.int 32
%int8_938 = torch.constant.int 8
%int128_939 = torch.constant.int 128
%1046 = torch.prim.ListConstruct %416, %int32_935, %int2_936, %int32_937, %int8_938, %int128_939 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1047 = torch.aten.view %1045, %1046 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1047, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_940 = torch.constant.int 2097152
%1048 = torch.prim.ListConstruct %416, %int2097152_940 : (!torch.int, !torch.int) -> !torch.list<int>
%1049 = torch.aten.view %1047, %1048 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1049, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_941 = torch.constant.int -2
%1050 = torch.aten.unsqueeze %1008, %int-2_941 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1050, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_942 = torch.constant.int 4
%int8_943 = torch.constant.int 8
%int4_944 = torch.constant.int 4
%int128_945 = torch.constant.int 128
%1051 = torch.prim.ListConstruct %int4_942, %994, %int8_943, %int4_944, %int128_945 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_946 = torch.constant.bool false
%1052 = torch.aten.expand %1050, %1051, %false_946 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1052, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_947 = torch.constant.int 0
%1053 = torch.aten.clone %1052, %int0_947 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1053, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_948 = torch.constant.int 4
%int32_949 = torch.constant.int 32
%int128_950 = torch.constant.int 128
%1054 = torch.prim.ListConstruct %int4_948, %994, %int32_949, %int128_950 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1055 = torch.aten._unsafe_view %1053, %1054 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1055, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_951 = torch.constant.int -2
%1056 = torch.aten.unsqueeze %950, %int-2_951 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1056, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_952 = torch.constant.int 1
%1057 = torch.aten.size.int %944, %int1_952 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_953 = torch.constant.int 4
%int8_954 = torch.constant.int 8
%int4_955 = torch.constant.int 4
%int128_956 = torch.constant.int 128
%1058 = torch.prim.ListConstruct %int4_953, %1057, %int8_954, %int4_955, %int128_956 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_957 = torch.constant.bool false
%1059 = torch.aten.expand %1056, %1058, %false_957 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1059, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_958 = torch.constant.int 0
%1060 = torch.aten.clone %1059, %int0_958 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1060, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_959 = torch.constant.int 4
%int32_960 = torch.constant.int 32
%int128_961 = torch.constant.int 128
%1061 = torch.prim.ListConstruct %int4_959, %1057, %int32_960, %int128_961 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1062 = torch.aten._unsafe_view %1060, %1061 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1062, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_962 = torch.constant.int 1
%int2_963 = torch.constant.int 2
%1063 = torch.aten.transpose.int %979, %int1_962, %int2_963 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1063, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_964 = torch.constant.int 1
%int2_965 = torch.constant.int 2
%1064 = torch.aten.transpose.int %1055, %int1_964, %int2_965 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1064, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_966 = torch.constant.int 1
%int2_967 = torch.constant.int 2
%1065 = torch.aten.transpose.int %1062, %int1_966, %int2_967 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1065, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_968 = torch.constant.float 0.000000e+00
%false_969 = torch.constant.bool false
%none_970 = torch.constant.none
%1066:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%1063, %1064, %1065, %float0.000000e00_968, %false_969, %320, %none_970) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %1066#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_971 = torch.constant.int 1
%int2_972 = torch.constant.int 2
%1067 = torch.aten.transpose.int %1066#0, %int1_971, %int2_972 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1067, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_973 = torch.constant.int 4
%int4096_974 = torch.constant.int 4096
%1068 = torch.prim.ListConstruct %int4_973, %965, %int4096_974 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1069 = torch.aten.view %1067, %1068 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1069, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_975 = torch.constant.int -2
%int-1_976 = torch.constant.int -1
%1070 = torch.aten.transpose.int %32, %int-2_975, %int-1_976 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_977 = torch.constant.int 4
%1071 = torch.aten.mul.int %int4_977, %965 : !torch.int, !torch.int -> !torch.int
%int4096_978 = torch.constant.int 4096
%1072 = torch.prim.ListConstruct %1071, %int4096_978 : (!torch.int, !torch.int) -> !torch.list<int>
%1073 = torch.aten.view %1069, %1072 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1073, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1074 = torch.aten.mm %1073, %1070 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1074, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_979 = torch.constant.int 4
%int4096_980 = torch.constant.int 4096
%1075 = torch.prim.ListConstruct %int4_979, %965, %int4096_980 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1076 = torch.aten.view %1074, %1075 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1076, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_981 = torch.constant.int 1
%1077 = torch.aten.add.Tensor %914, %1076, %int1_981 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1077, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_982 = torch.constant.int 6
%1078 = torch.prims.convert_element_type %1077, %int6_982 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1078, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_983 = torch.constant.int 2
%1079 = torch.aten.pow.Tensor_Scalar %1078, %int2_983 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1079, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_984 = torch.constant.int -1
%1080 = torch.prim.ListConstruct %int-1_984 : (!torch.int) -> !torch.list<int>
%true_985 = torch.constant.bool true
%none_986 = torch.constant.none
%1081 = torch.aten.mean.dim %1079, %1080, %true_985, %none_986 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1081, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_987 = torch.constant.float 9.9999997473787516E-6
%int1_988 = torch.constant.int 1
%1082 = torch.aten.add.Scalar %1081, %float9.999990e-06_987, %int1_988 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1082, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1083 = torch.aten.rsqrt %1082 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1083, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1084 = torch.aten.mul.Tensor %1078, %1083 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1084, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1085 = torch.aten.mul.Tensor %33, %1084 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1085, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_989 = torch.constant.int 5
%1086 = torch.prims.convert_element_type %1085, %int5_989 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1086, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_990 = torch.constant.int -2
%int-1_991 = torch.constant.int -1
%1087 = torch.aten.transpose.int %34, %int-2_990, %int-1_991 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_992 = torch.constant.int 4
%1088 = torch.aten.mul.int %int4_992, %294 : !torch.int, !torch.int -> !torch.int
%int4096_993 = torch.constant.int 4096
%1089 = torch.prim.ListConstruct %1088, %int4096_993 : (!torch.int, !torch.int) -> !torch.list<int>
%1090 = torch.aten.view %1086, %1089 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1090, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1091 = torch.aten.mm %1090, %1087 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1091, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_994 = torch.constant.int 4
%int14336_995 = torch.constant.int 14336
%1092 = torch.prim.ListConstruct %int4_994, %294, %int14336_995 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1093 = torch.aten.view %1091, %1092 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1093, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1094 = torch.aten.silu %1093 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1094, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_996 = torch.constant.int -2
%int-1_997 = torch.constant.int -1
%1095 = torch.aten.transpose.int %35, %int-2_996, %int-1_997 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_998 = torch.constant.int 4
%1096 = torch.aten.mul.int %int4_998, %294 : !torch.int, !torch.int -> !torch.int
%int4096_999 = torch.constant.int 4096
%1097 = torch.prim.ListConstruct %1096, %int4096_999 : (!torch.int, !torch.int) -> !torch.list<int>
%1098 = torch.aten.view %1086, %1097 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1098, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1099 = torch.aten.mm %1098, %1095 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1099, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1000 = torch.constant.int 4
%int14336_1001 = torch.constant.int 14336
%1100 = torch.prim.ListConstruct %int4_1000, %294, %int14336_1001 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1101 = torch.aten.view %1099, %1100 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1101, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1102 = torch.aten.mul.Tensor %1094, %1101 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1102, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1002 = torch.constant.int -2
%int-1_1003 = torch.constant.int -1
%1103 = torch.aten.transpose.int %36, %int-2_1002, %int-1_1003 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_1004 = torch.constant.int 1
%1104 = torch.aten.size.int %1093, %int1_1004 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_1005 = torch.constant.int 4
%1105 = torch.aten.mul.int %int4_1005, %1104 : !torch.int, !torch.int -> !torch.int
%int14336_1006 = torch.constant.int 14336
%1106 = torch.prim.ListConstruct %1105, %int14336_1006 : (!torch.int, !torch.int) -> !torch.list<int>
%1107 = torch.aten.view %1102, %1106 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1107, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%1108 = torch.aten.mm %1107, %1103 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1108, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1007 = torch.constant.int 4
%int4096_1008 = torch.constant.int 4096
%1109 = torch.prim.ListConstruct %int4_1007, %1104, %int4096_1008 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1110 = torch.aten.view %1108, %1109 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1110, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1009 = torch.constant.int 1
%1111 = torch.aten.add.Tensor %1077, %1110, %int1_1009 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1111, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1010 = torch.constant.int 6
%1112 = torch.prims.convert_element_type %1111, %int6_1010 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1112, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1011 = torch.constant.int 2
%1113 = torch.aten.pow.Tensor_Scalar %1112, %int2_1011 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1113, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1012 = torch.constant.int -1
%1114 = torch.prim.ListConstruct %int-1_1012 : (!torch.int) -> !torch.list<int>
%true_1013 = torch.constant.bool true
%none_1014 = torch.constant.none
%1115 = torch.aten.mean.dim %1113, %1114, %true_1013, %none_1014 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1115, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1015 = torch.constant.float 9.9999997473787516E-6
%int1_1016 = torch.constant.int 1
%1116 = torch.aten.add.Scalar %1115, %float9.999990e-06_1015, %int1_1016 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1116, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1117 = torch.aten.rsqrt %1116 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1117, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1118 = torch.aten.mul.Tensor %1112, %1117 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1118, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1119 = torch.aten.mul.Tensor %37, %1118 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1119, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1017 = torch.constant.int 5
%1120 = torch.prims.convert_element_type %1119, %int5_1017 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1120, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1018 = torch.constant.int -2
%int-1_1019 = torch.constant.int -1
%1121 = torch.aten.transpose.int %38, %int-2_1018, %int-1_1019 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1020 = torch.constant.int 4
%1122 = torch.aten.mul.int %int4_1020, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1021 = torch.constant.int 4096
%1123 = torch.prim.ListConstruct %1122, %int4096_1021 : (!torch.int, !torch.int) -> !torch.list<int>
%1124 = torch.aten.view %1120, %1123 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1124, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1125 = torch.aten.mm %1124, %1121 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1125, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1022 = torch.constant.int 4
%int4096_1023 = torch.constant.int 4096
%1126 = torch.prim.ListConstruct %int4_1022, %294, %int4096_1023 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1127 = torch.aten.view %1125, %1126 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1127, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1024 = torch.constant.int -2
%int-1_1025 = torch.constant.int -1
%1128 = torch.aten.transpose.int %39, %int-2_1024, %int-1_1025 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1026 = torch.constant.int 4
%1129 = torch.aten.mul.int %int4_1026, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1027 = torch.constant.int 4096
%1130 = torch.prim.ListConstruct %1129, %int4096_1027 : (!torch.int, !torch.int) -> !torch.list<int>
%1131 = torch.aten.view %1120, %1130 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1131, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1132 = torch.aten.mm %1131, %1128 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1132, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1028 = torch.constant.int 4
%int1024_1029 = torch.constant.int 1024
%1133 = torch.prim.ListConstruct %int4_1028, %294, %int1024_1029 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1134 = torch.aten.view %1132, %1133 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1134, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_1030 = torch.constant.int -2
%int-1_1031 = torch.constant.int -1
%1135 = torch.aten.transpose.int %40, %int-2_1030, %int-1_1031 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1032 = torch.constant.int 4
%1136 = torch.aten.mul.int %int4_1032, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1033 = torch.constant.int 4096
%1137 = torch.prim.ListConstruct %1136, %int4096_1033 : (!torch.int, !torch.int) -> !torch.list<int>
%1138 = torch.aten.view %1120, %1137 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1138, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1139 = torch.aten.mm %1138, %1135 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1139, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1034 = torch.constant.int 4
%int1024_1035 = torch.constant.int 1024
%1140 = torch.prim.ListConstruct %int4_1034, %294, %int1024_1035 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1141 = torch.aten.view %1139, %1140 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1141, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_1036 = torch.constant.int 4
%int32_1037 = torch.constant.int 32
%int128_1038 = torch.constant.int 128
%1142 = torch.prim.ListConstruct %int4_1036, %294, %int32_1037, %int128_1038 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1143 = torch.aten.view %1127, %1142 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1143, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1039 = torch.constant.int 4
%int8_1040 = torch.constant.int 8
%int128_1041 = torch.constant.int 128
%1144 = torch.prim.ListConstruct %int4_1039, %294, %int8_1040, %int128_1041 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1145 = torch.aten.view %1134, %1144 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1145, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_1042 = torch.constant.int 4
%int8_1043 = torch.constant.int 8
%int128_1044 = torch.constant.int 128
%1146 = torch.prim.ListConstruct %int4_1042, %294, %int8_1043, %int128_1044 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1147 = torch.aten.view %1141, %1146 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1147, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_1045 = torch.constant.int 131072
%none_1046 = torch.constant.none
%none_1047 = torch.constant.none
%cpu_1048 = torch.constant.device "cpu"
%false_1049 = torch.constant.bool false
%1148 = torch.aten.arange %int131072_1045, %none_1046, %none_1047, %cpu_1048, %false_1049 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1050 = torch.constant.int 0
%int128_1051 = torch.constant.int 128
%int2_1052 = torch.constant.int 2
%none_1053 = torch.constant.none
%none_1054 = torch.constant.none
%cpu_1055 = torch.constant.device "cpu"
%false_1056 = torch.constant.bool false
%1149 = torch.aten.arange.start_step %int0_1050, %int128_1051, %int2_1052, %none_1053, %none_1054, %cpu_1055, %false_1056 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1057 = torch.constant.int 0
%int0_1058 = torch.constant.int 0
%int64_1059 = torch.constant.int 64
%int1_1060 = torch.constant.int 1
%1150 = torch.aten.slice.Tensor %1149, %int0_1057, %int0_1058, %int64_1059, %int1_1060 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1061 = torch.constant.int 6
%1151 = torch.prims.convert_element_type %1150, %int6_1061 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1062 = torch.constant.int 128
%1152 = torch.aten.div.Scalar %1151, %int128_1062 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1063 = torch.constant.float 5.000000e+05
%1153 = torch.aten.pow.Scalar %float5.000000e05_1063, %1152 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1154 = torch.aten.reciprocal %1153 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1064 = torch.constant.float 1.000000e+00
%1155 = torch.aten.mul.Scalar %1154, %float1.000000e00_1064 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1065 = torch.constant.int 131072
%int1_1066 = torch.constant.int 1
%1156 = torch.prim.ListConstruct %int131072_1065, %int1_1066 : (!torch.int, !torch.int) -> !torch.list<int>
%1157 = torch.aten.view %1148, %1156 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1158 = torch.aten.mul.Tensor %1157, %1155 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1159 = torch.aten.cos %1158 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1160 = torch.aten.sin %1158 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1161 = torch.aten.complex %1159, %1160 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1067 = torch.constant.int 1
%1162 = torch.aten.size.int %1127, %int1_1067 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_1068 = torch.constant.int 0
%1163 = torch.aten.add.int %int0_1068, %1162 : !torch.int, !torch.int -> !torch.int
%int0_1069 = torch.constant.int 0
%int0_1070 = torch.constant.int 0
%int1_1071 = torch.constant.int 1
%1164 = torch.aten.slice.Tensor %1161, %int0_1069, %int0_1070, %1163, %int1_1071 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1164, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1072 = torch.constant.int 1
%int0_1073 = torch.constant.int 0
%int9223372036854775807_1074 = torch.constant.int 9223372036854775807
%int1_1075 = torch.constant.int 1
%1165 = torch.aten.slice.Tensor %1164, %int1_1072, %int0_1073, %int9223372036854775807_1074, %int1_1075 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1165, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1076 = torch.constant.int 0
%1166 = torch.aten.unsqueeze %1165, %int0_1076 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1166, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1077 = torch.constant.int 2
%1167 = torch.aten.unsqueeze %1166, %int2_1077 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1167, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1078 = torch.constant.int 3
%int0_1079 = torch.constant.int 0
%int9223372036854775807_1080 = torch.constant.int 9223372036854775807
%int1_1081 = torch.constant.int 1
%1168 = torch.aten.slice.Tensor %1167, %int3_1078, %int0_1079, %int9223372036854775807_1080, %int1_1081 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1168, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1169 = torch_c.to_builtin_tensor %1143 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_1082 = arith.constant 1 : index
%dim_1083 = tensor.dim %1169, %c1_1082 : tensor<4x?x32x128xf16>
%1170 = flow.tensor.bitcast %1169 : tensor<4x?x32x128xf16>{%dim_1083} -> tensor<4x?x32x64xcomplex<f16>>{%dim_1083}
%1171 = torch_c.from_builtin_tensor %1170 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %1171, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%1172 = torch.aten.mul.Tensor %1171, %1168 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %1172, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%1173 = torch_c.to_builtin_tensor %1172 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_1084 = arith.constant 1 : index
%dim_1085 = tensor.dim %1173, %c1_1084 : tensor<4x?x32x64xcomplex<f32>>
%1174 = flow.tensor.bitcast %1173 : tensor<4x?x32x64xcomplex<f32>>{%dim_1085} -> tensor<4x?x32x128xf32>{%dim_1085}
%1175 = torch_c.from_builtin_tensor %1174 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %1175, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_1086 = torch.constant.int 5
%1176 = torch.prims.convert_element_type %1175, %int5_1086 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1176, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_1087 = torch.constant.int 131072
%none_1088 = torch.constant.none
%none_1089 = torch.constant.none
%cpu_1090 = torch.constant.device "cpu"
%false_1091 = torch.constant.bool false
%1177 = torch.aten.arange %int131072_1087, %none_1088, %none_1089, %cpu_1090, %false_1091 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1092 = torch.constant.int 0
%int128_1093 = torch.constant.int 128
%int2_1094 = torch.constant.int 2
%none_1095 = torch.constant.none
%none_1096 = torch.constant.none
%cpu_1097 = torch.constant.device "cpu"
%false_1098 = torch.constant.bool false
%1178 = torch.aten.arange.start_step %int0_1092, %int128_1093, %int2_1094, %none_1095, %none_1096, %cpu_1097, %false_1098 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1099 = torch.constant.int 0
%int0_1100 = torch.constant.int 0
%int64_1101 = torch.constant.int 64
%int1_1102 = torch.constant.int 1
%1179 = torch.aten.slice.Tensor %1178, %int0_1099, %int0_1100, %int64_1101, %int1_1102 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1103 = torch.constant.int 6
%1180 = torch.prims.convert_element_type %1179, %int6_1103 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1104 = torch.constant.int 128
%1181 = torch.aten.div.Scalar %1180, %int128_1104 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1105 = torch.constant.float 5.000000e+05
%1182 = torch.aten.pow.Scalar %float5.000000e05_1105, %1181 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1183 = torch.aten.reciprocal %1182 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1106 = torch.constant.float 1.000000e+00
%1184 = torch.aten.mul.Scalar %1183, %float1.000000e00_1106 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1107 = torch.constant.int 131072
%int1_1108 = torch.constant.int 1
%1185 = torch.prim.ListConstruct %int131072_1107, %int1_1108 : (!torch.int, !torch.int) -> !torch.list<int>
%1186 = torch.aten.view %1177, %1185 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1187 = torch.aten.mul.Tensor %1186, %1184 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1188 = torch.aten.cos %1187 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1189 = torch.aten.sin %1187 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1190 = torch.aten.complex %1188, %1189 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1109 = torch.constant.int 1
%1191 = torch.aten.size.int %1134, %int1_1109 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_1110 = torch.constant.int 0
%1192 = torch.aten.add.int %int0_1110, %1191 : !torch.int, !torch.int -> !torch.int
%int0_1111 = torch.constant.int 0
%int0_1112 = torch.constant.int 0
%int1_1113 = torch.constant.int 1
%1193 = torch.aten.slice.Tensor %1190, %int0_1111, %int0_1112, %1192, %int1_1113 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1193, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1114 = torch.constant.int 1
%int0_1115 = torch.constant.int 0
%int9223372036854775807_1116 = torch.constant.int 9223372036854775807
%int1_1117 = torch.constant.int 1
%1194 = torch.aten.slice.Tensor %1193, %int1_1114, %int0_1115, %int9223372036854775807_1116, %int1_1117 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1194, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1118 = torch.constant.int 0
%1195 = torch.aten.unsqueeze %1194, %int0_1118 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1195, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1119 = torch.constant.int 2
%1196 = torch.aten.unsqueeze %1195, %int2_1119 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1196, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1120 = torch.constant.int 3
%int0_1121 = torch.constant.int 0
%int9223372036854775807_1122 = torch.constant.int 9223372036854775807
%int1_1123 = torch.constant.int 1
%1197 = torch.aten.slice.Tensor %1196, %int3_1120, %int0_1121, %int9223372036854775807_1122, %int1_1123 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1197, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1198 = torch_c.to_builtin_tensor %1145 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_1124 = arith.constant 1 : index
%dim_1125 = tensor.dim %1198, %c1_1124 : tensor<4x?x8x128xf16>
%1199 = flow.tensor.bitcast %1198 : tensor<4x?x8x128xf16>{%dim_1125} -> tensor<4x?x8x64xcomplex<f16>>{%dim_1125}
%1200 = torch_c.from_builtin_tensor %1199 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1200, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1201 = torch.aten.mul.Tensor %1200, %1197 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1201, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1202 = torch_c.to_builtin_tensor %1201 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_1126 = arith.constant 1 : index
%dim_1127 = tensor.dim %1202, %c1_1126 : tensor<4x?x8x64xcomplex<f32>>
%1203 = flow.tensor.bitcast %1202 : tensor<4x?x8x64xcomplex<f32>>{%dim_1127} -> tensor<4x?x8x128xf32>{%dim_1127}
%1204 = torch_c.from_builtin_tensor %1203 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1204, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_1128 = torch.constant.int 5
%1205 = torch.prims.convert_element_type %1204, %int5_1128 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1205, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_1129 = torch.constant.int 64
%1206 = torch.aten.mul.Scalar %arg2, %int64_1129 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1206, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int8_1130 = torch.constant.int 8
%int1_1131 = torch.constant.int 1
%1207 = torch.aten.add.Scalar %1206, %int8_1130, %int1_1131 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1207, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1132 = torch.constant.int 4
%int32_1133 = torch.constant.int 32
%int8_1134 = torch.constant.int 8
%int128_1135 = torch.constant.int 128
%1208 = torch.prim.ListConstruct %int4_1132, %425, %int32_1133, %int8_1134, %int128_1135 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1209 = torch.aten.view %1205, %1208 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1209, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1136 = torch.constant.int 4
%1210 = torch.aten.mul.int %int4_1136, %425 : !torch.int, !torch.int -> !torch.int
%int32_1137 = torch.constant.int 32
%int8_1138 = torch.constant.int 8
%int128_1139 = torch.constant.int 128
%1211 = torch.prim.ListConstruct %1210, %int32_1137, %int8_1138, %int128_1139 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1212 = torch.aten.view %1209, %1211 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1212, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1140 = torch.constant.int 4
%1213 = torch.aten.mul.int %int4_1140, %425 : !torch.int, !torch.int -> !torch.int
%1214 = torch.prim.ListConstruct %1213 : (!torch.int) -> !torch.list<int>
%1215 = torch.aten.view %1207, %1214 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1215, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_1141 = torch.constant.int 32
%int2_1142 = torch.constant.int 2
%int32_1143 = torch.constant.int 32
%int8_1144 = torch.constant.int 8
%int128_1145 = torch.constant.int 128
%1216 = torch.prim.ListConstruct %416, %int32_1141, %int2_1142, %int32_1143, %int8_1144, %int128_1145 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1217 = torch.aten.view %1049, %1216 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1217, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1146 = torch.constant.int 32
%1218 = torch.aten.mul.int %416, %int32_1146 : !torch.int, !torch.int -> !torch.int
%int2_1147 = torch.constant.int 2
%1219 = torch.aten.mul.int %1218, %int2_1147 : !torch.int, !torch.int -> !torch.int
%int32_1148 = torch.constant.int 32
%int8_1149 = torch.constant.int 8
%int128_1150 = torch.constant.int 128
%1220 = torch.prim.ListConstruct %1219, %int32_1148, %int8_1149, %int128_1150 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1221 = torch.aten.view %1217, %1220 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1221, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%1222 = torch.prim.ListConstruct %1215 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1151 = torch.constant.bool false
%1223 = torch.aten.index_put %1221, %1222, %1212, %false_1151 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1223, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1152 = torch.constant.int 32
%int2_1153 = torch.constant.int 2
%int32_1154 = torch.constant.int 32
%int8_1155 = torch.constant.int 8
%int128_1156 = torch.constant.int 128
%1224 = torch.prim.ListConstruct %416, %int32_1152, %int2_1153, %int32_1154, %int8_1155, %int128_1156 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1225 = torch.aten.view %1223, %1224 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1225, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1157 = torch.constant.int 2097152
%1226 = torch.prim.ListConstruct %416, %int2097152_1157 : (!torch.int, !torch.int) -> !torch.list<int>
%1227 = torch.aten.view %1225, %1226 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1227, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_1158 = torch.constant.int 32
%int2_1159 = torch.constant.int 2
%int32_1160 = torch.constant.int 32
%int8_1161 = torch.constant.int 8
%int128_1162 = torch.constant.int 128
%1228 = torch.prim.ListConstruct %416, %int32_1158, %int2_1159, %int32_1160, %int8_1161, %int128_1162 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1229 = torch.aten.view %1227, %1228 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1229, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1163 = torch.constant.int 32
%int8_1164 = torch.constant.int 8
%int128_1165 = torch.constant.int 128
%1230 = torch.prim.ListConstruct %1219, %int32_1163, %int8_1164, %int128_1165 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1231 = torch.aten.view %1229, %1230 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1231, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1166 = torch.constant.int 4
%int32_1167 = torch.constant.int 32
%int8_1168 = torch.constant.int 8
%int128_1169 = torch.constant.int 128
%1232 = torch.prim.ListConstruct %int4_1166, %425, %int32_1167, %int8_1168, %int128_1169 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1233 = torch.aten.view %1147, %1232 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1233, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1170 = torch.constant.int 4
%1234 = torch.aten.mul.int %int4_1170, %425 : !torch.int, !torch.int -> !torch.int
%int32_1171 = torch.constant.int 32
%int8_1172 = torch.constant.int 8
%int128_1173 = torch.constant.int 128
%1235 = torch.prim.ListConstruct %1234, %int32_1171, %int8_1172, %int128_1173 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1236 = torch.aten.view %1233, %1235 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1236, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_1174 = torch.constant.int 1
%int1_1175 = torch.constant.int 1
%1237 = torch.aten.add.Scalar %1207, %int1_1174, %int1_1175 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1237, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1176 = torch.constant.int 4
%1238 = torch.aten.mul.int %int4_1176, %425 : !torch.int, !torch.int -> !torch.int
%1239 = torch.prim.ListConstruct %1238 : (!torch.int) -> !torch.list<int>
%1240 = torch.aten.view %1237, %1239 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1240, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%1241 = torch.prim.ListConstruct %1240 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1177 = torch.constant.bool false
%1242 = torch.aten.index_put %1231, %1241, %1236, %false_1177 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1242, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1178 = torch.constant.int 32
%int2_1179 = torch.constant.int 2
%int32_1180 = torch.constant.int 32
%int8_1181 = torch.constant.int 8
%int128_1182 = torch.constant.int 128
%1243 = torch.prim.ListConstruct %416, %int32_1178, %int2_1179, %int32_1180, %int8_1181, %int128_1182 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1244 = torch.aten.view %1242, %1243 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1244, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1183 = torch.constant.int 2097152
%1245 = torch.prim.ListConstruct %416, %int2097152_1183 : (!torch.int, !torch.int) -> !torch.list<int>
%1246 = torch.aten.view %1244, %1245 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1246, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_1184 = torch.constant.int -2
%1247 = torch.aten.unsqueeze %1205, %int-2_1184 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1247, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_1185 = torch.constant.int 4
%int8_1186 = torch.constant.int 8
%int4_1187 = torch.constant.int 4
%int128_1188 = torch.constant.int 128
%1248 = torch.prim.ListConstruct %int4_1185, %1191, %int8_1186, %int4_1187, %int128_1188 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1189 = torch.constant.bool false
%1249 = torch.aten.expand %1247, %1248, %false_1189 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1249, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1190 = torch.constant.int 0
%1250 = torch.aten.clone %1249, %int0_1190 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1250, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1191 = torch.constant.int 4
%int32_1192 = torch.constant.int 32
%int128_1193 = torch.constant.int 128
%1251 = torch.prim.ListConstruct %int4_1191, %1191, %int32_1192, %int128_1193 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1252 = torch.aten._unsafe_view %1250, %1251 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1252, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_1194 = torch.constant.int -2
%1253 = torch.aten.unsqueeze %1147, %int-2_1194 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1253, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_1195 = torch.constant.int 1
%1254 = torch.aten.size.int %1141, %int1_1195 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_1196 = torch.constant.int 4
%int8_1197 = torch.constant.int 8
%int4_1198 = torch.constant.int 4
%int128_1199 = torch.constant.int 128
%1255 = torch.prim.ListConstruct %int4_1196, %1254, %int8_1197, %int4_1198, %int128_1199 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1200 = torch.constant.bool false
%1256 = torch.aten.expand %1253, %1255, %false_1200 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1256, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1201 = torch.constant.int 0
%1257 = torch.aten.clone %1256, %int0_1201 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1257, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1202 = torch.constant.int 4
%int32_1203 = torch.constant.int 32
%int128_1204 = torch.constant.int 128
%1258 = torch.prim.ListConstruct %int4_1202, %1254, %int32_1203, %int128_1204 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1259 = torch.aten._unsafe_view %1257, %1258 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1259, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_1205 = torch.constant.int 1
%int2_1206 = torch.constant.int 2
%1260 = torch.aten.transpose.int %1176, %int1_1205, %int2_1206 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1260, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1207 = torch.constant.int 1
%int2_1208 = torch.constant.int 2
%1261 = torch.aten.transpose.int %1252, %int1_1207, %int2_1208 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1261, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1209 = torch.constant.int 1
%int2_1210 = torch.constant.int 2
%1262 = torch.aten.transpose.int %1259, %int1_1209, %int2_1210 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1262, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_1211 = torch.constant.float 0.000000e+00
%false_1212 = torch.constant.bool false
%none_1213 = torch.constant.none
%1263:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%1260, %1261, %1262, %float0.000000e00_1211, %false_1212, %320, %none_1213) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %1263#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1214 = torch.constant.int 1
%int2_1215 = torch.constant.int 2
%1264 = torch.aten.transpose.int %1263#0, %int1_1214, %int2_1215 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1264, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1216 = torch.constant.int 4
%int4096_1217 = torch.constant.int 4096
%1265 = torch.prim.ListConstruct %int4_1216, %1162, %int4096_1217 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1266 = torch.aten.view %1264, %1265 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1266, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1218 = torch.constant.int -2
%int-1_1219 = torch.constant.int -1
%1267 = torch.aten.transpose.int %41, %int-2_1218, %int-1_1219 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1220 = torch.constant.int 4
%1268 = torch.aten.mul.int %int4_1220, %1162 : !torch.int, !torch.int -> !torch.int
%int4096_1221 = torch.constant.int 4096
%1269 = torch.prim.ListConstruct %1268, %int4096_1221 : (!torch.int, !torch.int) -> !torch.list<int>
%1270 = torch.aten.view %1266, %1269 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1270, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1271 = torch.aten.mm %1270, %1267 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1271, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1222 = torch.constant.int 4
%int4096_1223 = torch.constant.int 4096
%1272 = torch.prim.ListConstruct %int4_1222, %1162, %int4096_1223 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1273 = torch.aten.view %1271, %1272 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1273, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1224 = torch.constant.int 1
%1274 = torch.aten.add.Tensor %1111, %1273, %int1_1224 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1274, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1225 = torch.constant.int 6
%1275 = torch.prims.convert_element_type %1274, %int6_1225 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1275, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1226 = torch.constant.int 2
%1276 = torch.aten.pow.Tensor_Scalar %1275, %int2_1226 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1276, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1227 = torch.constant.int -1
%1277 = torch.prim.ListConstruct %int-1_1227 : (!torch.int) -> !torch.list<int>
%true_1228 = torch.constant.bool true
%none_1229 = torch.constant.none
%1278 = torch.aten.mean.dim %1276, %1277, %true_1228, %none_1229 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1278, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1230 = torch.constant.float 9.9999997473787516E-6
%int1_1231 = torch.constant.int 1
%1279 = torch.aten.add.Scalar %1278, %float9.999990e-06_1230, %int1_1231 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1279, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1280 = torch.aten.rsqrt %1279 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1280, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1281 = torch.aten.mul.Tensor %1275, %1280 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1281, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1282 = torch.aten.mul.Tensor %42, %1281 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1282, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1232 = torch.constant.int 5
%1283 = torch.prims.convert_element_type %1282, %int5_1232 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1283, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1233 = torch.constant.int -2
%int-1_1234 = torch.constant.int -1
%1284 = torch.aten.transpose.int %43, %int-2_1233, %int-1_1234 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1235 = torch.constant.int 4
%1285 = torch.aten.mul.int %int4_1235, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1236 = torch.constant.int 4096
%1286 = torch.prim.ListConstruct %1285, %int4096_1236 : (!torch.int, !torch.int) -> !torch.list<int>
%1287 = torch.aten.view %1283, %1286 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1287, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1288 = torch.aten.mm %1287, %1284 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1288, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1237 = torch.constant.int 4
%int14336_1238 = torch.constant.int 14336
%1289 = torch.prim.ListConstruct %int4_1237, %294, %int14336_1238 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1290 = torch.aten.view %1288, %1289 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1290, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1291 = torch.aten.silu %1290 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1291, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1239 = torch.constant.int -2
%int-1_1240 = torch.constant.int -1
%1292 = torch.aten.transpose.int %44, %int-2_1239, %int-1_1240 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1241 = torch.constant.int 4
%1293 = torch.aten.mul.int %int4_1241, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1242 = torch.constant.int 4096
%1294 = torch.prim.ListConstruct %1293, %int4096_1242 : (!torch.int, !torch.int) -> !torch.list<int>
%1295 = torch.aten.view %1283, %1294 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1295, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1296 = torch.aten.mm %1295, %1292 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1296, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1243 = torch.constant.int 4
%int14336_1244 = torch.constant.int 14336
%1297 = torch.prim.ListConstruct %int4_1243, %294, %int14336_1244 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1298 = torch.aten.view %1296, %1297 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1298, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1299 = torch.aten.mul.Tensor %1291, %1298 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1299, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1245 = torch.constant.int -2
%int-1_1246 = torch.constant.int -1
%1300 = torch.aten.transpose.int %45, %int-2_1245, %int-1_1246 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_1247 = torch.constant.int 1
%1301 = torch.aten.size.int %1290, %int1_1247 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_1248 = torch.constant.int 4
%1302 = torch.aten.mul.int %int4_1248, %1301 : !torch.int, !torch.int -> !torch.int
%int14336_1249 = torch.constant.int 14336
%1303 = torch.prim.ListConstruct %1302, %int14336_1249 : (!torch.int, !torch.int) -> !torch.list<int>
%1304 = torch.aten.view %1299, %1303 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1304, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%1305 = torch.aten.mm %1304, %1300 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1305, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1250 = torch.constant.int 4
%int4096_1251 = torch.constant.int 4096
%1306 = torch.prim.ListConstruct %int4_1250, %1301, %int4096_1251 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1307 = torch.aten.view %1305, %1306 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1307, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1252 = torch.constant.int 1
%1308 = torch.aten.add.Tensor %1274, %1307, %int1_1252 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1308, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1253 = torch.constant.int 6
%1309 = torch.prims.convert_element_type %1308, %int6_1253 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1309, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1254 = torch.constant.int 2
%1310 = torch.aten.pow.Tensor_Scalar %1309, %int2_1254 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1310, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1255 = torch.constant.int -1
%1311 = torch.prim.ListConstruct %int-1_1255 : (!torch.int) -> !torch.list<int>
%true_1256 = torch.constant.bool true
%none_1257 = torch.constant.none
%1312 = torch.aten.mean.dim %1310, %1311, %true_1256, %none_1257 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1312, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1258 = torch.constant.float 9.9999997473787516E-6
%int1_1259 = torch.constant.int 1
%1313 = torch.aten.add.Scalar %1312, %float9.999990e-06_1258, %int1_1259 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1313, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1314 = torch.aten.rsqrt %1313 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1314, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1315 = torch.aten.mul.Tensor %1309, %1314 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1315, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1316 = torch.aten.mul.Tensor %46, %1315 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1316, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1260 = torch.constant.int 5
%1317 = torch.prims.convert_element_type %1316, %int5_1260 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1317, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1261 = torch.constant.int -2
%int-1_1262 = torch.constant.int -1
%1318 = torch.aten.transpose.int %47, %int-2_1261, %int-1_1262 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1263 = torch.constant.int 4
%1319 = torch.aten.mul.int %int4_1263, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1264 = torch.constant.int 4096
%1320 = torch.prim.ListConstruct %1319, %int4096_1264 : (!torch.int, !torch.int) -> !torch.list<int>
%1321 = torch.aten.view %1317, %1320 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1321, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1322 = torch.aten.mm %1321, %1318 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1322, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1265 = torch.constant.int 4
%int4096_1266 = torch.constant.int 4096
%1323 = torch.prim.ListConstruct %int4_1265, %294, %int4096_1266 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1324 = torch.aten.view %1322, %1323 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1324, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1267 = torch.constant.int -2
%int-1_1268 = torch.constant.int -1
%1325 = torch.aten.transpose.int %48, %int-2_1267, %int-1_1268 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1269 = torch.constant.int 4
%1326 = torch.aten.mul.int %int4_1269, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1270 = torch.constant.int 4096
%1327 = torch.prim.ListConstruct %1326, %int4096_1270 : (!torch.int, !torch.int) -> !torch.list<int>
%1328 = torch.aten.view %1317, %1327 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1328, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1329 = torch.aten.mm %1328, %1325 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1329, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1271 = torch.constant.int 4
%int1024_1272 = torch.constant.int 1024
%1330 = torch.prim.ListConstruct %int4_1271, %294, %int1024_1272 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1331 = torch.aten.view %1329, %1330 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1331, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_1273 = torch.constant.int -2
%int-1_1274 = torch.constant.int -1
%1332 = torch.aten.transpose.int %49, %int-2_1273, %int-1_1274 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1275 = torch.constant.int 4
%1333 = torch.aten.mul.int %int4_1275, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1276 = torch.constant.int 4096
%1334 = torch.prim.ListConstruct %1333, %int4096_1276 : (!torch.int, !torch.int) -> !torch.list<int>
%1335 = torch.aten.view %1317, %1334 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1335, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1336 = torch.aten.mm %1335, %1332 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1336, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1277 = torch.constant.int 4
%int1024_1278 = torch.constant.int 1024
%1337 = torch.prim.ListConstruct %int4_1277, %294, %int1024_1278 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1338 = torch.aten.view %1336, %1337 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1338, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_1279 = torch.constant.int 4
%int32_1280 = torch.constant.int 32
%int128_1281 = torch.constant.int 128
%1339 = torch.prim.ListConstruct %int4_1279, %294, %int32_1280, %int128_1281 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1340 = torch.aten.view %1324, %1339 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1340, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1282 = torch.constant.int 4
%int8_1283 = torch.constant.int 8
%int128_1284 = torch.constant.int 128
%1341 = torch.prim.ListConstruct %int4_1282, %294, %int8_1283, %int128_1284 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1342 = torch.aten.view %1331, %1341 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1342, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_1285 = torch.constant.int 4
%int8_1286 = torch.constant.int 8
%int128_1287 = torch.constant.int 128
%1343 = torch.prim.ListConstruct %int4_1285, %294, %int8_1286, %int128_1287 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1344 = torch.aten.view %1338, %1343 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1344, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_1288 = torch.constant.int 131072
%none_1289 = torch.constant.none
%none_1290 = torch.constant.none
%cpu_1291 = torch.constant.device "cpu"
%false_1292 = torch.constant.bool false
%1345 = torch.aten.arange %int131072_1288, %none_1289, %none_1290, %cpu_1291, %false_1292 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1293 = torch.constant.int 0
%int128_1294 = torch.constant.int 128
%int2_1295 = torch.constant.int 2
%none_1296 = torch.constant.none
%none_1297 = torch.constant.none
%cpu_1298 = torch.constant.device "cpu"
%false_1299 = torch.constant.bool false
%1346 = torch.aten.arange.start_step %int0_1293, %int128_1294, %int2_1295, %none_1296, %none_1297, %cpu_1298, %false_1299 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1300 = torch.constant.int 0
%int0_1301 = torch.constant.int 0
%int64_1302 = torch.constant.int 64
%int1_1303 = torch.constant.int 1
%1347 = torch.aten.slice.Tensor %1346, %int0_1300, %int0_1301, %int64_1302, %int1_1303 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1304 = torch.constant.int 6
%1348 = torch.prims.convert_element_type %1347, %int6_1304 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1305 = torch.constant.int 128
%1349 = torch.aten.div.Scalar %1348, %int128_1305 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1306 = torch.constant.float 5.000000e+05
%1350 = torch.aten.pow.Scalar %float5.000000e05_1306, %1349 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1351 = torch.aten.reciprocal %1350 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1307 = torch.constant.float 1.000000e+00
%1352 = torch.aten.mul.Scalar %1351, %float1.000000e00_1307 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1308 = torch.constant.int 131072
%int1_1309 = torch.constant.int 1
%1353 = torch.prim.ListConstruct %int131072_1308, %int1_1309 : (!torch.int, !torch.int) -> !torch.list<int>
%1354 = torch.aten.view %1345, %1353 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1355 = torch.aten.mul.Tensor %1354, %1352 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1356 = torch.aten.cos %1355 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1357 = torch.aten.sin %1355 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1358 = torch.aten.complex %1356, %1357 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1310 = torch.constant.int 1
%1359 = torch.aten.size.int %1324, %int1_1310 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_1311 = torch.constant.int 0
%1360 = torch.aten.add.int %int0_1311, %1359 : !torch.int, !torch.int -> !torch.int
%int0_1312 = torch.constant.int 0
%int0_1313 = torch.constant.int 0
%int1_1314 = torch.constant.int 1
%1361 = torch.aten.slice.Tensor %1358, %int0_1312, %int0_1313, %1360, %int1_1314 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1361, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1315 = torch.constant.int 1
%int0_1316 = torch.constant.int 0
%int9223372036854775807_1317 = torch.constant.int 9223372036854775807
%int1_1318 = torch.constant.int 1
%1362 = torch.aten.slice.Tensor %1361, %int1_1315, %int0_1316, %int9223372036854775807_1317, %int1_1318 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1362, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1319 = torch.constant.int 0
%1363 = torch.aten.unsqueeze %1362, %int0_1319 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1363, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1320 = torch.constant.int 2
%1364 = torch.aten.unsqueeze %1363, %int2_1320 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1364, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1321 = torch.constant.int 3
%int0_1322 = torch.constant.int 0
%int9223372036854775807_1323 = torch.constant.int 9223372036854775807
%int1_1324 = torch.constant.int 1
%1365 = torch.aten.slice.Tensor %1364, %int3_1321, %int0_1322, %int9223372036854775807_1323, %int1_1324 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1365, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1366 = torch_c.to_builtin_tensor %1340 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_1325 = arith.constant 1 : index
%dim_1326 = tensor.dim %1366, %c1_1325 : tensor<4x?x32x128xf16>
%1367 = flow.tensor.bitcast %1366 : tensor<4x?x32x128xf16>{%dim_1326} -> tensor<4x?x32x64xcomplex<f16>>{%dim_1326}
%1368 = torch_c.from_builtin_tensor %1367 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %1368, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%1369 = torch.aten.mul.Tensor %1368, %1365 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %1369, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%1370 = torch_c.to_builtin_tensor %1369 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_1327 = arith.constant 1 : index
%dim_1328 = tensor.dim %1370, %c1_1327 : tensor<4x?x32x64xcomplex<f32>>
%1371 = flow.tensor.bitcast %1370 : tensor<4x?x32x64xcomplex<f32>>{%dim_1328} -> tensor<4x?x32x128xf32>{%dim_1328}
%1372 = torch_c.from_builtin_tensor %1371 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %1372, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_1329 = torch.constant.int 5
%1373 = torch.prims.convert_element_type %1372, %int5_1329 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1373, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_1330 = torch.constant.int 131072
%none_1331 = torch.constant.none
%none_1332 = torch.constant.none
%cpu_1333 = torch.constant.device "cpu"
%false_1334 = torch.constant.bool false
%1374 = torch.aten.arange %int131072_1330, %none_1331, %none_1332, %cpu_1333, %false_1334 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1335 = torch.constant.int 0
%int128_1336 = torch.constant.int 128
%int2_1337 = torch.constant.int 2
%none_1338 = torch.constant.none
%none_1339 = torch.constant.none
%cpu_1340 = torch.constant.device "cpu"
%false_1341 = torch.constant.bool false
%1375 = torch.aten.arange.start_step %int0_1335, %int128_1336, %int2_1337, %none_1338, %none_1339, %cpu_1340, %false_1341 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1342 = torch.constant.int 0
%int0_1343 = torch.constant.int 0
%int64_1344 = torch.constant.int 64
%int1_1345 = torch.constant.int 1
%1376 = torch.aten.slice.Tensor %1375, %int0_1342, %int0_1343, %int64_1344, %int1_1345 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1346 = torch.constant.int 6
%1377 = torch.prims.convert_element_type %1376, %int6_1346 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1347 = torch.constant.int 128
%1378 = torch.aten.div.Scalar %1377, %int128_1347 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1348 = torch.constant.float 5.000000e+05
%1379 = torch.aten.pow.Scalar %float5.000000e05_1348, %1378 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1380 = torch.aten.reciprocal %1379 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1349 = torch.constant.float 1.000000e+00
%1381 = torch.aten.mul.Scalar %1380, %float1.000000e00_1349 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1350 = torch.constant.int 131072
%int1_1351 = torch.constant.int 1
%1382 = torch.prim.ListConstruct %int131072_1350, %int1_1351 : (!torch.int, !torch.int) -> !torch.list<int>
%1383 = torch.aten.view %1374, %1382 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1384 = torch.aten.mul.Tensor %1383, %1381 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1385 = torch.aten.cos %1384 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1386 = torch.aten.sin %1384 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1387 = torch.aten.complex %1385, %1386 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1352 = torch.constant.int 1
%1388 = torch.aten.size.int %1331, %int1_1352 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_1353 = torch.constant.int 0
%1389 = torch.aten.add.int %int0_1353, %1388 : !torch.int, !torch.int -> !torch.int
%int0_1354 = torch.constant.int 0
%int0_1355 = torch.constant.int 0
%int1_1356 = torch.constant.int 1
%1390 = torch.aten.slice.Tensor %1387, %int0_1354, %int0_1355, %1389, %int1_1356 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1390, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1357 = torch.constant.int 1
%int0_1358 = torch.constant.int 0
%int9223372036854775807_1359 = torch.constant.int 9223372036854775807
%int1_1360 = torch.constant.int 1
%1391 = torch.aten.slice.Tensor %1390, %int1_1357, %int0_1358, %int9223372036854775807_1359, %int1_1360 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1391, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1361 = torch.constant.int 0
%1392 = torch.aten.unsqueeze %1391, %int0_1361 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1392, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1362 = torch.constant.int 2
%1393 = torch.aten.unsqueeze %1392, %int2_1362 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1393, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1363 = torch.constant.int 3
%int0_1364 = torch.constant.int 0
%int9223372036854775807_1365 = torch.constant.int 9223372036854775807
%int1_1366 = torch.constant.int 1
%1394 = torch.aten.slice.Tensor %1393, %int3_1363, %int0_1364, %int9223372036854775807_1365, %int1_1366 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1394, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1395 = torch_c.to_builtin_tensor %1342 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_1367 = arith.constant 1 : index
%dim_1368 = tensor.dim %1395, %c1_1367 : tensor<4x?x8x128xf16>
%1396 = flow.tensor.bitcast %1395 : tensor<4x?x8x128xf16>{%dim_1368} -> tensor<4x?x8x64xcomplex<f16>>{%dim_1368}
%1397 = torch_c.from_builtin_tensor %1396 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1397, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1398 = torch.aten.mul.Tensor %1397, %1394 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1398, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1399 = torch_c.to_builtin_tensor %1398 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_1369 = arith.constant 1 : index
%dim_1370 = tensor.dim %1399, %c1_1369 : tensor<4x?x8x64xcomplex<f32>>
%1400 = flow.tensor.bitcast %1399 : tensor<4x?x8x64xcomplex<f32>>{%dim_1370} -> tensor<4x?x8x128xf32>{%dim_1370}
%1401 = torch_c.from_builtin_tensor %1400 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1401, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_1371 = torch.constant.int 5
%1402 = torch.prims.convert_element_type %1401, %int5_1371 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1402, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_1372 = torch.constant.int 64
%1403 = torch.aten.mul.Scalar %arg2, %int64_1372 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1403, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int10 = torch.constant.int 10
%int1_1373 = torch.constant.int 1
%1404 = torch.aten.add.Scalar %1403, %int10, %int1_1373 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1404, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1374 = torch.constant.int 4
%int32_1375 = torch.constant.int 32
%int8_1376 = torch.constant.int 8
%int128_1377 = torch.constant.int 128
%1405 = torch.prim.ListConstruct %int4_1374, %425, %int32_1375, %int8_1376, %int128_1377 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1406 = torch.aten.view %1402, %1405 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1406, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1378 = torch.constant.int 4
%1407 = torch.aten.mul.int %int4_1378, %425 : !torch.int, !torch.int -> !torch.int
%int32_1379 = torch.constant.int 32
%int8_1380 = torch.constant.int 8
%int128_1381 = torch.constant.int 128
%1408 = torch.prim.ListConstruct %1407, %int32_1379, %int8_1380, %int128_1381 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1409 = torch.aten.view %1406, %1408 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1409, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1382 = torch.constant.int 4
%1410 = torch.aten.mul.int %int4_1382, %425 : !torch.int, !torch.int -> !torch.int
%1411 = torch.prim.ListConstruct %1410 : (!torch.int) -> !torch.list<int>
%1412 = torch.aten.view %1404, %1411 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1412, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_1383 = torch.constant.int 32
%int2_1384 = torch.constant.int 2
%int32_1385 = torch.constant.int 32
%int8_1386 = torch.constant.int 8
%int128_1387 = torch.constant.int 128
%1413 = torch.prim.ListConstruct %416, %int32_1383, %int2_1384, %int32_1385, %int8_1386, %int128_1387 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1414 = torch.aten.view %1246, %1413 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1414, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1388 = torch.constant.int 32
%1415 = torch.aten.mul.int %416, %int32_1388 : !torch.int, !torch.int -> !torch.int
%int2_1389 = torch.constant.int 2
%1416 = torch.aten.mul.int %1415, %int2_1389 : !torch.int, !torch.int -> !torch.int
%int32_1390 = torch.constant.int 32
%int8_1391 = torch.constant.int 8
%int128_1392 = torch.constant.int 128
%1417 = torch.prim.ListConstruct %1416, %int32_1390, %int8_1391, %int128_1392 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1418 = torch.aten.view %1414, %1417 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1418, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%1419 = torch.prim.ListConstruct %1412 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1393 = torch.constant.bool false
%1420 = torch.aten.index_put %1418, %1419, %1409, %false_1393 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1420, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1394 = torch.constant.int 32
%int2_1395 = torch.constant.int 2
%int32_1396 = torch.constant.int 32
%int8_1397 = torch.constant.int 8
%int128_1398 = torch.constant.int 128
%1421 = torch.prim.ListConstruct %416, %int32_1394, %int2_1395, %int32_1396, %int8_1397, %int128_1398 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1422 = torch.aten.view %1420, %1421 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1422, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1399 = torch.constant.int 2097152
%1423 = torch.prim.ListConstruct %416, %int2097152_1399 : (!torch.int, !torch.int) -> !torch.list<int>
%1424 = torch.aten.view %1422, %1423 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1424, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_1400 = torch.constant.int 32
%int2_1401 = torch.constant.int 2
%int32_1402 = torch.constant.int 32
%int8_1403 = torch.constant.int 8
%int128_1404 = torch.constant.int 128
%1425 = torch.prim.ListConstruct %416, %int32_1400, %int2_1401, %int32_1402, %int8_1403, %int128_1404 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1426 = torch.aten.view %1424, %1425 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1426, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1405 = torch.constant.int 32
%int8_1406 = torch.constant.int 8
%int128_1407 = torch.constant.int 128
%1427 = torch.prim.ListConstruct %1416, %int32_1405, %int8_1406, %int128_1407 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1428 = torch.aten.view %1426, %1427 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1428, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1408 = torch.constant.int 4
%int32_1409 = torch.constant.int 32
%int8_1410 = torch.constant.int 8
%int128_1411 = torch.constant.int 128
%1429 = torch.prim.ListConstruct %int4_1408, %425, %int32_1409, %int8_1410, %int128_1411 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1430 = torch.aten.view %1344, %1429 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1430, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1412 = torch.constant.int 4
%1431 = torch.aten.mul.int %int4_1412, %425 : !torch.int, !torch.int -> !torch.int
%int32_1413 = torch.constant.int 32
%int8_1414 = torch.constant.int 8
%int128_1415 = torch.constant.int 128
%1432 = torch.prim.ListConstruct %1431, %int32_1413, %int8_1414, %int128_1415 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1433 = torch.aten.view %1430, %1432 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1433, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_1416 = torch.constant.int 1
%int1_1417 = torch.constant.int 1
%1434 = torch.aten.add.Scalar %1404, %int1_1416, %int1_1417 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1434, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1418 = torch.constant.int 4
%1435 = torch.aten.mul.int %int4_1418, %425 : !torch.int, !torch.int -> !torch.int
%1436 = torch.prim.ListConstruct %1435 : (!torch.int) -> !torch.list<int>
%1437 = torch.aten.view %1434, %1436 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1437, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%1438 = torch.prim.ListConstruct %1437 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1419 = torch.constant.bool false
%1439 = torch.aten.index_put %1428, %1438, %1433, %false_1419 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1439, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1420 = torch.constant.int 32
%int2_1421 = torch.constant.int 2
%int32_1422 = torch.constant.int 32
%int8_1423 = torch.constant.int 8
%int128_1424 = torch.constant.int 128
%1440 = torch.prim.ListConstruct %416, %int32_1420, %int2_1421, %int32_1422, %int8_1423, %int128_1424 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1441 = torch.aten.view %1439, %1440 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1441, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1425 = torch.constant.int 2097152
%1442 = torch.prim.ListConstruct %416, %int2097152_1425 : (!torch.int, !torch.int) -> !torch.list<int>
%1443 = torch.aten.view %1441, %1442 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1443, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_1426 = torch.constant.int -2
%1444 = torch.aten.unsqueeze %1402, %int-2_1426 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1444, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_1427 = torch.constant.int 4
%int8_1428 = torch.constant.int 8
%int4_1429 = torch.constant.int 4
%int128_1430 = torch.constant.int 128
%1445 = torch.prim.ListConstruct %int4_1427, %1388, %int8_1428, %int4_1429, %int128_1430 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1431 = torch.constant.bool false
%1446 = torch.aten.expand %1444, %1445, %false_1431 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1446, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1432 = torch.constant.int 0
%1447 = torch.aten.clone %1446, %int0_1432 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1447, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1433 = torch.constant.int 4
%int32_1434 = torch.constant.int 32
%int128_1435 = torch.constant.int 128
%1448 = torch.prim.ListConstruct %int4_1433, %1388, %int32_1434, %int128_1435 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1449 = torch.aten._unsafe_view %1447, %1448 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1449, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_1436 = torch.constant.int -2
%1450 = torch.aten.unsqueeze %1344, %int-2_1436 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1450, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_1437 = torch.constant.int 1
%1451 = torch.aten.size.int %1338, %int1_1437 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_1438 = torch.constant.int 4
%int8_1439 = torch.constant.int 8
%int4_1440 = torch.constant.int 4
%int128_1441 = torch.constant.int 128
%1452 = torch.prim.ListConstruct %int4_1438, %1451, %int8_1439, %int4_1440, %int128_1441 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1442 = torch.constant.bool false
%1453 = torch.aten.expand %1450, %1452, %false_1442 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1453, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1443 = torch.constant.int 0
%1454 = torch.aten.clone %1453, %int0_1443 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1454, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1444 = torch.constant.int 4
%int32_1445 = torch.constant.int 32
%int128_1446 = torch.constant.int 128
%1455 = torch.prim.ListConstruct %int4_1444, %1451, %int32_1445, %int128_1446 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1456 = torch.aten._unsafe_view %1454, %1455 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1456, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_1447 = torch.constant.int 1
%int2_1448 = torch.constant.int 2
%1457 = torch.aten.transpose.int %1373, %int1_1447, %int2_1448 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1457, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1449 = torch.constant.int 1
%int2_1450 = torch.constant.int 2
%1458 = torch.aten.transpose.int %1449, %int1_1449, %int2_1450 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1458, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1451 = torch.constant.int 1
%int2_1452 = torch.constant.int 2
%1459 = torch.aten.transpose.int %1456, %int1_1451, %int2_1452 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1459, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_1453 = torch.constant.float 0.000000e+00
%false_1454 = torch.constant.bool false
%none_1455 = torch.constant.none
%1460:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%1457, %1458, %1459, %float0.000000e00_1453, %false_1454, %320, %none_1455) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %1460#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1456 = torch.constant.int 1
%int2_1457 = torch.constant.int 2
%1461 = torch.aten.transpose.int %1460#0, %int1_1456, %int2_1457 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1461, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1458 = torch.constant.int 4
%int4096_1459 = torch.constant.int 4096
%1462 = torch.prim.ListConstruct %int4_1458, %1359, %int4096_1459 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1463 = torch.aten.view %1461, %1462 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1463, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1460 = torch.constant.int -2
%int-1_1461 = torch.constant.int -1
%1464 = torch.aten.transpose.int %50, %int-2_1460, %int-1_1461 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1462 = torch.constant.int 4
%1465 = torch.aten.mul.int %int4_1462, %1359 : !torch.int, !torch.int -> !torch.int
%int4096_1463 = torch.constant.int 4096
%1466 = torch.prim.ListConstruct %1465, %int4096_1463 : (!torch.int, !torch.int) -> !torch.list<int>
%1467 = torch.aten.view %1463, %1466 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1467, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1468 = torch.aten.mm %1467, %1464 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1468, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1464 = torch.constant.int 4
%int4096_1465 = torch.constant.int 4096
%1469 = torch.prim.ListConstruct %int4_1464, %1359, %int4096_1465 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1470 = torch.aten.view %1468, %1469 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1470, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1466 = torch.constant.int 1
%1471 = torch.aten.add.Tensor %1308, %1470, %int1_1466 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1471, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1467 = torch.constant.int 6
%1472 = torch.prims.convert_element_type %1471, %int6_1467 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1472, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1468 = torch.constant.int 2
%1473 = torch.aten.pow.Tensor_Scalar %1472, %int2_1468 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1473, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1469 = torch.constant.int -1
%1474 = torch.prim.ListConstruct %int-1_1469 : (!torch.int) -> !torch.list<int>
%true_1470 = torch.constant.bool true
%none_1471 = torch.constant.none
%1475 = torch.aten.mean.dim %1473, %1474, %true_1470, %none_1471 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1475, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1472 = torch.constant.float 9.9999997473787516E-6
%int1_1473 = torch.constant.int 1
%1476 = torch.aten.add.Scalar %1475, %float9.999990e-06_1472, %int1_1473 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1476, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1477 = torch.aten.rsqrt %1476 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1477, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1478 = torch.aten.mul.Tensor %1472, %1477 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1478, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1479 = torch.aten.mul.Tensor %51, %1478 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1479, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1474 = torch.constant.int 5
%1480 = torch.prims.convert_element_type %1479, %int5_1474 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1480, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1475 = torch.constant.int -2
%int-1_1476 = torch.constant.int -1
%1481 = torch.aten.transpose.int %52, %int-2_1475, %int-1_1476 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1477 = torch.constant.int 4
%1482 = torch.aten.mul.int %int4_1477, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1478 = torch.constant.int 4096
%1483 = torch.prim.ListConstruct %1482, %int4096_1478 : (!torch.int, !torch.int) -> !torch.list<int>
%1484 = torch.aten.view %1480, %1483 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1484, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1485 = torch.aten.mm %1484, %1481 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1485, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1479 = torch.constant.int 4
%int14336_1480 = torch.constant.int 14336
%1486 = torch.prim.ListConstruct %int4_1479, %294, %int14336_1480 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1487 = torch.aten.view %1485, %1486 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1487, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1488 = torch.aten.silu %1487 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1488, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1481 = torch.constant.int -2
%int-1_1482 = torch.constant.int -1
%1489 = torch.aten.transpose.int %53, %int-2_1481, %int-1_1482 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1483 = torch.constant.int 4
%1490 = torch.aten.mul.int %int4_1483, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1484 = torch.constant.int 4096
%1491 = torch.prim.ListConstruct %1490, %int4096_1484 : (!torch.int, !torch.int) -> !torch.list<int>
%1492 = torch.aten.view %1480, %1491 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1492, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1493 = torch.aten.mm %1492, %1489 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1493, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1485 = torch.constant.int 4
%int14336_1486 = torch.constant.int 14336
%1494 = torch.prim.ListConstruct %int4_1485, %294, %int14336_1486 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1495 = torch.aten.view %1493, %1494 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1495, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1496 = torch.aten.mul.Tensor %1488, %1495 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1496, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1487 = torch.constant.int -2
%int-1_1488 = torch.constant.int -1
%1497 = torch.aten.transpose.int %54, %int-2_1487, %int-1_1488 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_1489 = torch.constant.int 1
%1498 = torch.aten.size.int %1487, %int1_1489 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_1490 = torch.constant.int 4
%1499 = torch.aten.mul.int %int4_1490, %1498 : !torch.int, !torch.int -> !torch.int
%int14336_1491 = torch.constant.int 14336
%1500 = torch.prim.ListConstruct %1499, %int14336_1491 : (!torch.int, !torch.int) -> !torch.list<int>
%1501 = torch.aten.view %1496, %1500 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1501, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%1502 = torch.aten.mm %1501, %1497 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1502, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1492 = torch.constant.int 4
%int4096_1493 = torch.constant.int 4096
%1503 = torch.prim.ListConstruct %int4_1492, %1498, %int4096_1493 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1504 = torch.aten.view %1502, %1503 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1504, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1494 = torch.constant.int 1
%1505 = torch.aten.add.Tensor %1471, %1504, %int1_1494 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1505, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1495 = torch.constant.int 6
%1506 = torch.prims.convert_element_type %1505, %int6_1495 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1506, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1496 = torch.constant.int 2
%1507 = torch.aten.pow.Tensor_Scalar %1506, %int2_1496 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1507, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1497 = torch.constant.int -1
%1508 = torch.prim.ListConstruct %int-1_1497 : (!torch.int) -> !torch.list<int>
%true_1498 = torch.constant.bool true
%none_1499 = torch.constant.none
%1509 = torch.aten.mean.dim %1507, %1508, %true_1498, %none_1499 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1509, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1500 = torch.constant.float 9.9999997473787516E-6
%int1_1501 = torch.constant.int 1
%1510 = torch.aten.add.Scalar %1509, %float9.999990e-06_1500, %int1_1501 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1510, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1511 = torch.aten.rsqrt %1510 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1511, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1512 = torch.aten.mul.Tensor %1506, %1511 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1512, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1513 = torch.aten.mul.Tensor %55, %1512 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1513, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1502 = torch.constant.int 5
%1514 = torch.prims.convert_element_type %1513, %int5_1502 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1514, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1503 = torch.constant.int -2
%int-1_1504 = torch.constant.int -1
%1515 = torch.aten.transpose.int %56, %int-2_1503, %int-1_1504 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1505 = torch.constant.int 4
%1516 = torch.aten.mul.int %int4_1505, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1506 = torch.constant.int 4096
%1517 = torch.prim.ListConstruct %1516, %int4096_1506 : (!torch.int, !torch.int) -> !torch.list<int>
%1518 = torch.aten.view %1514, %1517 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1518, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1519 = torch.aten.mm %1518, %1515 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1519, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1507 = torch.constant.int 4
%int4096_1508 = torch.constant.int 4096
%1520 = torch.prim.ListConstruct %int4_1507, %294, %int4096_1508 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1521 = torch.aten.view %1519, %1520 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1521, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1509 = torch.constant.int -2
%int-1_1510 = torch.constant.int -1
%1522 = torch.aten.transpose.int %57, %int-2_1509, %int-1_1510 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1511 = torch.constant.int 4
%1523 = torch.aten.mul.int %int4_1511, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1512 = torch.constant.int 4096
%1524 = torch.prim.ListConstruct %1523, %int4096_1512 : (!torch.int, !torch.int) -> !torch.list<int>
%1525 = torch.aten.view %1514, %1524 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1525, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1526 = torch.aten.mm %1525, %1522 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1526, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1513 = torch.constant.int 4
%int1024_1514 = torch.constant.int 1024
%1527 = torch.prim.ListConstruct %int4_1513, %294, %int1024_1514 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1528 = torch.aten.view %1526, %1527 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1528, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_1515 = torch.constant.int -2
%int-1_1516 = torch.constant.int -1
%1529 = torch.aten.transpose.int %58, %int-2_1515, %int-1_1516 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1517 = torch.constant.int 4
%1530 = torch.aten.mul.int %int4_1517, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1518 = torch.constant.int 4096
%1531 = torch.prim.ListConstruct %1530, %int4096_1518 : (!torch.int, !torch.int) -> !torch.list<int>
%1532 = torch.aten.view %1514, %1531 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1532, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1533 = torch.aten.mm %1532, %1529 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1533, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1519 = torch.constant.int 4
%int1024_1520 = torch.constant.int 1024
%1534 = torch.prim.ListConstruct %int4_1519, %294, %int1024_1520 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1535 = torch.aten.view %1533, %1534 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1535, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_1521 = torch.constant.int 4
%int32_1522 = torch.constant.int 32
%int128_1523 = torch.constant.int 128
%1536 = torch.prim.ListConstruct %int4_1521, %294, %int32_1522, %int128_1523 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1537 = torch.aten.view %1521, %1536 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1537, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1524 = torch.constant.int 4
%int8_1525 = torch.constant.int 8
%int128_1526 = torch.constant.int 128
%1538 = torch.prim.ListConstruct %int4_1524, %294, %int8_1525, %int128_1526 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1539 = torch.aten.view %1528, %1538 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1539, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_1527 = torch.constant.int 4
%int8_1528 = torch.constant.int 8
%int128_1529 = torch.constant.int 128
%1540 = torch.prim.ListConstruct %int4_1527, %294, %int8_1528, %int128_1529 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1541 = torch.aten.view %1535, %1540 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1541, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_1530 = torch.constant.int 131072
%none_1531 = torch.constant.none
%none_1532 = torch.constant.none
%cpu_1533 = torch.constant.device "cpu"
%false_1534 = torch.constant.bool false
%1542 = torch.aten.arange %int131072_1530, %none_1531, %none_1532, %cpu_1533, %false_1534 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1535 = torch.constant.int 0
%int128_1536 = torch.constant.int 128
%int2_1537 = torch.constant.int 2
%none_1538 = torch.constant.none
%none_1539 = torch.constant.none
%cpu_1540 = torch.constant.device "cpu"
%false_1541 = torch.constant.bool false
%1543 = torch.aten.arange.start_step %int0_1535, %int128_1536, %int2_1537, %none_1538, %none_1539, %cpu_1540, %false_1541 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1542 = torch.constant.int 0
%int0_1543 = torch.constant.int 0
%int64_1544 = torch.constant.int 64
%int1_1545 = torch.constant.int 1
%1544 = torch.aten.slice.Tensor %1543, %int0_1542, %int0_1543, %int64_1544, %int1_1545 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1546 = torch.constant.int 6
%1545 = torch.prims.convert_element_type %1544, %int6_1546 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1547 = torch.constant.int 128
%1546 = torch.aten.div.Scalar %1545, %int128_1547 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1548 = torch.constant.float 5.000000e+05
%1547 = torch.aten.pow.Scalar %float5.000000e05_1548, %1546 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1548 = torch.aten.reciprocal %1547 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1549 = torch.constant.float 1.000000e+00
%1549 = torch.aten.mul.Scalar %1548, %float1.000000e00_1549 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1550 = torch.constant.int 131072
%int1_1551 = torch.constant.int 1
%1550 = torch.prim.ListConstruct %int131072_1550, %int1_1551 : (!torch.int, !torch.int) -> !torch.list<int>
%1551 = torch.aten.view %1542, %1550 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1552 = torch.aten.mul.Tensor %1551, %1549 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1553 = torch.aten.cos %1552 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1554 = torch.aten.sin %1552 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1555 = torch.aten.complex %1553, %1554 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1552 = torch.constant.int 1
%1556 = torch.aten.size.int %1521, %int1_1552 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_1553 = torch.constant.int 0
%1557 = torch.aten.add.int %int0_1553, %1556 : !torch.int, !torch.int -> !torch.int
%int0_1554 = torch.constant.int 0
%int0_1555 = torch.constant.int 0
%int1_1556 = torch.constant.int 1
%1558 = torch.aten.slice.Tensor %1555, %int0_1554, %int0_1555, %1557, %int1_1556 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1558, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1557 = torch.constant.int 1
%int0_1558 = torch.constant.int 0
%int9223372036854775807_1559 = torch.constant.int 9223372036854775807
%int1_1560 = torch.constant.int 1
%1559 = torch.aten.slice.Tensor %1558, %int1_1557, %int0_1558, %int9223372036854775807_1559, %int1_1560 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1559, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1561 = torch.constant.int 0
%1560 = torch.aten.unsqueeze %1559, %int0_1561 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1560, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1562 = torch.constant.int 2
%1561 = torch.aten.unsqueeze %1560, %int2_1562 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1561, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1563 = torch.constant.int 3
%int0_1564 = torch.constant.int 0
%int9223372036854775807_1565 = torch.constant.int 9223372036854775807
%int1_1566 = torch.constant.int 1
%1562 = torch.aten.slice.Tensor %1561, %int3_1563, %int0_1564, %int9223372036854775807_1565, %int1_1566 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1562, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1563 = torch_c.to_builtin_tensor %1537 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_1567 = arith.constant 1 : index
%dim_1568 = tensor.dim %1563, %c1_1567 : tensor<4x?x32x128xf16>
%1564 = flow.tensor.bitcast %1563 : tensor<4x?x32x128xf16>{%dim_1568} -> tensor<4x?x32x64xcomplex<f16>>{%dim_1568}
%1565 = torch_c.from_builtin_tensor %1564 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %1565, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%1566 = torch.aten.mul.Tensor %1565, %1562 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %1566, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%1567 = torch_c.to_builtin_tensor %1566 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_1569 = arith.constant 1 : index
%dim_1570 = tensor.dim %1567, %c1_1569 : tensor<4x?x32x64xcomplex<f32>>
%1568 = flow.tensor.bitcast %1567 : tensor<4x?x32x64xcomplex<f32>>{%dim_1570} -> tensor<4x?x32x128xf32>{%dim_1570}
%1569 = torch_c.from_builtin_tensor %1568 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %1569, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_1571 = torch.constant.int 5
%1570 = torch.prims.convert_element_type %1569, %int5_1571 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1570, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_1572 = torch.constant.int 131072
%none_1573 = torch.constant.none
%none_1574 = torch.constant.none
%cpu_1575 = torch.constant.device "cpu"
%false_1576 = torch.constant.bool false
%1571 = torch.aten.arange %int131072_1572, %none_1573, %none_1574, %cpu_1575, %false_1576 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1577 = torch.constant.int 0
%int128_1578 = torch.constant.int 128
%int2_1579 = torch.constant.int 2
%none_1580 = torch.constant.none
%none_1581 = torch.constant.none
%cpu_1582 = torch.constant.device "cpu"
%false_1583 = torch.constant.bool false
%1572 = torch.aten.arange.start_step %int0_1577, %int128_1578, %int2_1579, %none_1580, %none_1581, %cpu_1582, %false_1583 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1584 = torch.constant.int 0
%int0_1585 = torch.constant.int 0
%int64_1586 = torch.constant.int 64
%int1_1587 = torch.constant.int 1
%1573 = torch.aten.slice.Tensor %1572, %int0_1584, %int0_1585, %int64_1586, %int1_1587 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1588 = torch.constant.int 6
%1574 = torch.prims.convert_element_type %1573, %int6_1588 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1589 = torch.constant.int 128
%1575 = torch.aten.div.Scalar %1574, %int128_1589 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1590 = torch.constant.float 5.000000e+05
%1576 = torch.aten.pow.Scalar %float5.000000e05_1590, %1575 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1577 = torch.aten.reciprocal %1576 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1591 = torch.constant.float 1.000000e+00
%1578 = torch.aten.mul.Scalar %1577, %float1.000000e00_1591 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1592 = torch.constant.int 131072
%int1_1593 = torch.constant.int 1
%1579 = torch.prim.ListConstruct %int131072_1592, %int1_1593 : (!torch.int, !torch.int) -> !torch.list<int>
%1580 = torch.aten.view %1571, %1579 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1581 = torch.aten.mul.Tensor %1580, %1578 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1582 = torch.aten.cos %1581 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1583 = torch.aten.sin %1581 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1584 = torch.aten.complex %1582, %1583 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1594 = torch.constant.int 1
%1585 = torch.aten.size.int %1528, %int1_1594 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_1595 = torch.constant.int 0
%1586 = torch.aten.add.int %int0_1595, %1585 : !torch.int, !torch.int -> !torch.int
%int0_1596 = torch.constant.int 0
%int0_1597 = torch.constant.int 0
%int1_1598 = torch.constant.int 1
%1587 = torch.aten.slice.Tensor %1584, %int0_1596, %int0_1597, %1586, %int1_1598 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1587, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1599 = torch.constant.int 1
%int0_1600 = torch.constant.int 0
%int9223372036854775807_1601 = torch.constant.int 9223372036854775807
%int1_1602 = torch.constant.int 1
%1588 = torch.aten.slice.Tensor %1587, %int1_1599, %int0_1600, %int9223372036854775807_1601, %int1_1602 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1588, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1603 = torch.constant.int 0
%1589 = torch.aten.unsqueeze %1588, %int0_1603 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1589, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1604 = torch.constant.int 2
%1590 = torch.aten.unsqueeze %1589, %int2_1604 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1590, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1605 = torch.constant.int 3
%int0_1606 = torch.constant.int 0
%int9223372036854775807_1607 = torch.constant.int 9223372036854775807
%int1_1608 = torch.constant.int 1
%1591 = torch.aten.slice.Tensor %1590, %int3_1605, %int0_1606, %int9223372036854775807_1607, %int1_1608 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1591, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1592 = torch_c.to_builtin_tensor %1539 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_1609 = arith.constant 1 : index
%dim_1610 = tensor.dim %1592, %c1_1609 : tensor<4x?x8x128xf16>
%1593 = flow.tensor.bitcast %1592 : tensor<4x?x8x128xf16>{%dim_1610} -> tensor<4x?x8x64xcomplex<f16>>{%dim_1610}
%1594 = torch_c.from_builtin_tensor %1593 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1594, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1595 = torch.aten.mul.Tensor %1594, %1591 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1595, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1596 = torch_c.to_builtin_tensor %1595 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_1611 = arith.constant 1 : index
%dim_1612 = tensor.dim %1596, %c1_1611 : tensor<4x?x8x64xcomplex<f32>>
%1597 = flow.tensor.bitcast %1596 : tensor<4x?x8x64xcomplex<f32>>{%dim_1612} -> tensor<4x?x8x128xf32>{%dim_1612}
%1598 = torch_c.from_builtin_tensor %1597 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1598, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_1613 = torch.constant.int 5
%1599 = torch.prims.convert_element_type %1598, %int5_1613 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1599, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_1614 = torch.constant.int 64
%1600 = torch.aten.mul.Scalar %arg2, %int64_1614 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1600, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int12 = torch.constant.int 12
%int1_1615 = torch.constant.int 1
%1601 = torch.aten.add.Scalar %1600, %int12, %int1_1615 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1601, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1616 = torch.constant.int 4
%int32_1617 = torch.constant.int 32
%int8_1618 = torch.constant.int 8
%int128_1619 = torch.constant.int 128
%1602 = torch.prim.ListConstruct %int4_1616, %425, %int32_1617, %int8_1618, %int128_1619 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1603 = torch.aten.view %1599, %1602 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1603, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1620 = torch.constant.int 4
%1604 = torch.aten.mul.int %int4_1620, %425 : !torch.int, !torch.int -> !torch.int
%int32_1621 = torch.constant.int 32
%int8_1622 = torch.constant.int 8
%int128_1623 = torch.constant.int 128
%1605 = torch.prim.ListConstruct %1604, %int32_1621, %int8_1622, %int128_1623 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1606 = torch.aten.view %1603, %1605 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1606, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1624 = torch.constant.int 4
%1607 = torch.aten.mul.int %int4_1624, %425 : !torch.int, !torch.int -> !torch.int
%1608 = torch.prim.ListConstruct %1607 : (!torch.int) -> !torch.list<int>
%1609 = torch.aten.view %1601, %1608 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1609, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_1625 = torch.constant.int 32
%int2_1626 = torch.constant.int 2
%int32_1627 = torch.constant.int 32
%int8_1628 = torch.constant.int 8
%int128_1629 = torch.constant.int 128
%1610 = torch.prim.ListConstruct %416, %int32_1625, %int2_1626, %int32_1627, %int8_1628, %int128_1629 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1611 = torch.aten.view %1443, %1610 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1611, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1630 = torch.constant.int 32
%1612 = torch.aten.mul.int %416, %int32_1630 : !torch.int, !torch.int -> !torch.int
%int2_1631 = torch.constant.int 2
%1613 = torch.aten.mul.int %1612, %int2_1631 : !torch.int, !torch.int -> !torch.int
%int32_1632 = torch.constant.int 32
%int8_1633 = torch.constant.int 8
%int128_1634 = torch.constant.int 128
%1614 = torch.prim.ListConstruct %1613, %int32_1632, %int8_1633, %int128_1634 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1615 = torch.aten.view %1611, %1614 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1615, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%1616 = torch.prim.ListConstruct %1609 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1635 = torch.constant.bool false
%1617 = torch.aten.index_put %1615, %1616, %1606, %false_1635 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1617, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1636 = torch.constant.int 32
%int2_1637 = torch.constant.int 2
%int32_1638 = torch.constant.int 32
%int8_1639 = torch.constant.int 8
%int128_1640 = torch.constant.int 128
%1618 = torch.prim.ListConstruct %416, %int32_1636, %int2_1637, %int32_1638, %int8_1639, %int128_1640 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1619 = torch.aten.view %1617, %1618 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1619, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1641 = torch.constant.int 2097152
%1620 = torch.prim.ListConstruct %416, %int2097152_1641 : (!torch.int, !torch.int) -> !torch.list<int>
%1621 = torch.aten.view %1619, %1620 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1621, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_1642 = torch.constant.int 32
%int2_1643 = torch.constant.int 2
%int32_1644 = torch.constant.int 32
%int8_1645 = torch.constant.int 8
%int128_1646 = torch.constant.int 128
%1622 = torch.prim.ListConstruct %416, %int32_1642, %int2_1643, %int32_1644, %int8_1645, %int128_1646 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1623 = torch.aten.view %1621, %1622 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1623, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1647 = torch.constant.int 32
%int8_1648 = torch.constant.int 8
%int128_1649 = torch.constant.int 128
%1624 = torch.prim.ListConstruct %1613, %int32_1647, %int8_1648, %int128_1649 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1625 = torch.aten.view %1623, %1624 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1625, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1650 = torch.constant.int 4
%int32_1651 = torch.constant.int 32
%int8_1652 = torch.constant.int 8
%int128_1653 = torch.constant.int 128
%1626 = torch.prim.ListConstruct %int4_1650, %425, %int32_1651, %int8_1652, %int128_1653 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1627 = torch.aten.view %1541, %1626 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1627, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1654 = torch.constant.int 4
%1628 = torch.aten.mul.int %int4_1654, %425 : !torch.int, !torch.int -> !torch.int
%int32_1655 = torch.constant.int 32
%int8_1656 = torch.constant.int 8
%int128_1657 = torch.constant.int 128
%1629 = torch.prim.ListConstruct %1628, %int32_1655, %int8_1656, %int128_1657 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1630 = torch.aten.view %1627, %1629 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1630, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_1658 = torch.constant.int 1
%int1_1659 = torch.constant.int 1
%1631 = torch.aten.add.Scalar %1601, %int1_1658, %int1_1659 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1631, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1660 = torch.constant.int 4
%1632 = torch.aten.mul.int %int4_1660, %425 : !torch.int, !torch.int -> !torch.int
%1633 = torch.prim.ListConstruct %1632 : (!torch.int) -> !torch.list<int>
%1634 = torch.aten.view %1631, %1633 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1634, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%1635 = torch.prim.ListConstruct %1634 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1661 = torch.constant.bool false
%1636 = torch.aten.index_put %1625, %1635, %1630, %false_1661 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1636, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1662 = torch.constant.int 32
%int2_1663 = torch.constant.int 2
%int32_1664 = torch.constant.int 32
%int8_1665 = torch.constant.int 8
%int128_1666 = torch.constant.int 128
%1637 = torch.prim.ListConstruct %416, %int32_1662, %int2_1663, %int32_1664, %int8_1665, %int128_1666 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1638 = torch.aten.view %1636, %1637 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1638, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1667 = torch.constant.int 2097152
%1639 = torch.prim.ListConstruct %416, %int2097152_1667 : (!torch.int, !torch.int) -> !torch.list<int>
%1640 = torch.aten.view %1638, %1639 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1640, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_1668 = torch.constant.int -2
%1641 = torch.aten.unsqueeze %1599, %int-2_1668 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1641, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_1669 = torch.constant.int 4
%int8_1670 = torch.constant.int 8
%int4_1671 = torch.constant.int 4
%int128_1672 = torch.constant.int 128
%1642 = torch.prim.ListConstruct %int4_1669, %1585, %int8_1670, %int4_1671, %int128_1672 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1673 = torch.constant.bool false
%1643 = torch.aten.expand %1641, %1642, %false_1673 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1643, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1674 = torch.constant.int 0
%1644 = torch.aten.clone %1643, %int0_1674 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1644, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1675 = torch.constant.int 4
%int32_1676 = torch.constant.int 32
%int128_1677 = torch.constant.int 128
%1645 = torch.prim.ListConstruct %int4_1675, %1585, %int32_1676, %int128_1677 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1646 = torch.aten._unsafe_view %1644, %1645 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1646, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_1678 = torch.constant.int -2
%1647 = torch.aten.unsqueeze %1541, %int-2_1678 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1647, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_1679 = torch.constant.int 1
%1648 = torch.aten.size.int %1535, %int1_1679 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_1680 = torch.constant.int 4
%int8_1681 = torch.constant.int 8
%int4_1682 = torch.constant.int 4
%int128_1683 = torch.constant.int 128
%1649 = torch.prim.ListConstruct %int4_1680, %1648, %int8_1681, %int4_1682, %int128_1683 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1684 = torch.constant.bool false
%1650 = torch.aten.expand %1647, %1649, %false_1684 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1650, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1685 = torch.constant.int 0
%1651 = torch.aten.clone %1650, %int0_1685 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1651, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1686 = torch.constant.int 4
%int32_1687 = torch.constant.int 32
%int128_1688 = torch.constant.int 128
%1652 = torch.prim.ListConstruct %int4_1686, %1648, %int32_1687, %int128_1688 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1653 = torch.aten._unsafe_view %1651, %1652 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1653, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_1689 = torch.constant.int 1
%int2_1690 = torch.constant.int 2
%1654 = torch.aten.transpose.int %1570, %int1_1689, %int2_1690 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1654, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1691 = torch.constant.int 1
%int2_1692 = torch.constant.int 2
%1655 = torch.aten.transpose.int %1646, %int1_1691, %int2_1692 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1655, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1693 = torch.constant.int 1
%int2_1694 = torch.constant.int 2
%1656 = torch.aten.transpose.int %1653, %int1_1693, %int2_1694 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1656, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_1695 = torch.constant.float 0.000000e+00
%false_1696 = torch.constant.bool false
%none_1697 = torch.constant.none
%1657:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%1654, %1655, %1656, %float0.000000e00_1695, %false_1696, %320, %none_1697) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %1657#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1698 = torch.constant.int 1
%int2_1699 = torch.constant.int 2
%1658 = torch.aten.transpose.int %1657#0, %int1_1698, %int2_1699 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1658, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1700 = torch.constant.int 4
%int4096_1701 = torch.constant.int 4096
%1659 = torch.prim.ListConstruct %int4_1700, %1556, %int4096_1701 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1660 = torch.aten.view %1658, %1659 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1660, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1702 = torch.constant.int -2
%int-1_1703 = torch.constant.int -1
%1661 = torch.aten.transpose.int %59, %int-2_1702, %int-1_1703 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1704 = torch.constant.int 4
%1662 = torch.aten.mul.int %int4_1704, %1556 : !torch.int, !torch.int -> !torch.int
%int4096_1705 = torch.constant.int 4096
%1663 = torch.prim.ListConstruct %1662, %int4096_1705 : (!torch.int, !torch.int) -> !torch.list<int>
%1664 = torch.aten.view %1660, %1663 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1664, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1665 = torch.aten.mm %1664, %1661 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1665, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1706 = torch.constant.int 4
%int4096_1707 = torch.constant.int 4096
%1666 = torch.prim.ListConstruct %int4_1706, %1556, %int4096_1707 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1667 = torch.aten.view %1665, %1666 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1667, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1708 = torch.constant.int 1
%1668 = torch.aten.add.Tensor %1505, %1667, %int1_1708 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1668, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1709 = torch.constant.int 6
%1669 = torch.prims.convert_element_type %1668, %int6_1709 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1669, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1710 = torch.constant.int 2
%1670 = torch.aten.pow.Tensor_Scalar %1669, %int2_1710 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1670, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1711 = torch.constant.int -1
%1671 = torch.prim.ListConstruct %int-1_1711 : (!torch.int) -> !torch.list<int>
%true_1712 = torch.constant.bool true
%none_1713 = torch.constant.none
%1672 = torch.aten.mean.dim %1670, %1671, %true_1712, %none_1713 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1672, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1714 = torch.constant.float 9.9999997473787516E-6
%int1_1715 = torch.constant.int 1
%1673 = torch.aten.add.Scalar %1672, %float9.999990e-06_1714, %int1_1715 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1673, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1674 = torch.aten.rsqrt %1673 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1674, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1675 = torch.aten.mul.Tensor %1669, %1674 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1675, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1676 = torch.aten.mul.Tensor %60, %1675 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1676, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1716 = torch.constant.int 5
%1677 = torch.prims.convert_element_type %1676, %int5_1716 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1677, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1717 = torch.constant.int -2
%int-1_1718 = torch.constant.int -1
%1678 = torch.aten.transpose.int %61, %int-2_1717, %int-1_1718 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1719 = torch.constant.int 4
%1679 = torch.aten.mul.int %int4_1719, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1720 = torch.constant.int 4096
%1680 = torch.prim.ListConstruct %1679, %int4096_1720 : (!torch.int, !torch.int) -> !torch.list<int>
%1681 = torch.aten.view %1677, %1680 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1681, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1682 = torch.aten.mm %1681, %1678 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1682, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1721 = torch.constant.int 4
%int14336_1722 = torch.constant.int 14336
%1683 = torch.prim.ListConstruct %int4_1721, %294, %int14336_1722 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1684 = torch.aten.view %1682, %1683 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1684, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1685 = torch.aten.silu %1684 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1685, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1723 = torch.constant.int -2
%int-1_1724 = torch.constant.int -1
%1686 = torch.aten.transpose.int %62, %int-2_1723, %int-1_1724 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1725 = torch.constant.int 4
%1687 = torch.aten.mul.int %int4_1725, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1726 = torch.constant.int 4096
%1688 = torch.prim.ListConstruct %1687, %int4096_1726 : (!torch.int, !torch.int) -> !torch.list<int>
%1689 = torch.aten.view %1677, %1688 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1689, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1690 = torch.aten.mm %1689, %1686 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1690, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1727 = torch.constant.int 4
%int14336_1728 = torch.constant.int 14336
%1691 = torch.prim.ListConstruct %int4_1727, %294, %int14336_1728 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1692 = torch.aten.view %1690, %1691 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1692, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1693 = torch.aten.mul.Tensor %1685, %1692 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1693, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1729 = torch.constant.int -2
%int-1_1730 = torch.constant.int -1
%1694 = torch.aten.transpose.int %63, %int-2_1729, %int-1_1730 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_1731 = torch.constant.int 1
%1695 = torch.aten.size.int %1684, %int1_1731 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_1732 = torch.constant.int 4
%1696 = torch.aten.mul.int %int4_1732, %1695 : !torch.int, !torch.int -> !torch.int
%int14336_1733 = torch.constant.int 14336
%1697 = torch.prim.ListConstruct %1696, %int14336_1733 : (!torch.int, !torch.int) -> !torch.list<int>
%1698 = torch.aten.view %1693, %1697 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1698, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%1699 = torch.aten.mm %1698, %1694 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1699, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1734 = torch.constant.int 4
%int4096_1735 = torch.constant.int 4096
%1700 = torch.prim.ListConstruct %int4_1734, %1695, %int4096_1735 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1701 = torch.aten.view %1699, %1700 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1701, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1736 = torch.constant.int 1
%1702 = torch.aten.add.Tensor %1668, %1701, %int1_1736 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1702, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1737 = torch.constant.int 6
%1703 = torch.prims.convert_element_type %1702, %int6_1737 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1703, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1738 = torch.constant.int 2
%1704 = torch.aten.pow.Tensor_Scalar %1703, %int2_1738 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1704, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1739 = torch.constant.int -1
%1705 = torch.prim.ListConstruct %int-1_1739 : (!torch.int) -> !torch.list<int>
%true_1740 = torch.constant.bool true
%none_1741 = torch.constant.none
%1706 = torch.aten.mean.dim %1704, %1705, %true_1740, %none_1741 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1706, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1742 = torch.constant.float 9.9999997473787516E-6
%int1_1743 = torch.constant.int 1
%1707 = torch.aten.add.Scalar %1706, %float9.999990e-06_1742, %int1_1743 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1707, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1708 = torch.aten.rsqrt %1707 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1708, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1709 = torch.aten.mul.Tensor %1703, %1708 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1709, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1710 = torch.aten.mul.Tensor %64, %1709 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1710, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1744 = torch.constant.int 5
%1711 = torch.prims.convert_element_type %1710, %int5_1744 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1711, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1745 = torch.constant.int -2
%int-1_1746 = torch.constant.int -1
%1712 = torch.aten.transpose.int %65, %int-2_1745, %int-1_1746 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1747 = torch.constant.int 4
%1713 = torch.aten.mul.int %int4_1747, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1748 = torch.constant.int 4096
%1714 = torch.prim.ListConstruct %1713, %int4096_1748 : (!torch.int, !torch.int) -> !torch.list<int>
%1715 = torch.aten.view %1711, %1714 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1715, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1716 = torch.aten.mm %1715, %1712 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1716, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1749 = torch.constant.int 4
%int4096_1750 = torch.constant.int 4096
%1717 = torch.prim.ListConstruct %int4_1749, %294, %int4096_1750 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1718 = torch.aten.view %1716, %1717 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1718, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1751 = torch.constant.int -2
%int-1_1752 = torch.constant.int -1
%1719 = torch.aten.transpose.int %66, %int-2_1751, %int-1_1752 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1753 = torch.constant.int 4
%1720 = torch.aten.mul.int %int4_1753, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1754 = torch.constant.int 4096
%1721 = torch.prim.ListConstruct %1720, %int4096_1754 : (!torch.int, !torch.int) -> !torch.list<int>
%1722 = torch.aten.view %1711, %1721 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1722, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1723 = torch.aten.mm %1722, %1719 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1723, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1755 = torch.constant.int 4
%int1024_1756 = torch.constant.int 1024
%1724 = torch.prim.ListConstruct %int4_1755, %294, %int1024_1756 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1725 = torch.aten.view %1723, %1724 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1725, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_1757 = torch.constant.int -2
%int-1_1758 = torch.constant.int -1
%1726 = torch.aten.transpose.int %67, %int-2_1757, %int-1_1758 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1759 = torch.constant.int 4
%1727 = torch.aten.mul.int %int4_1759, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1760 = torch.constant.int 4096
%1728 = torch.prim.ListConstruct %1727, %int4096_1760 : (!torch.int, !torch.int) -> !torch.list<int>
%1729 = torch.aten.view %1711, %1728 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1729, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1730 = torch.aten.mm %1729, %1726 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1730, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1761 = torch.constant.int 4
%int1024_1762 = torch.constant.int 1024
%1731 = torch.prim.ListConstruct %int4_1761, %294, %int1024_1762 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1732 = torch.aten.view %1730, %1731 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1732, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_1763 = torch.constant.int 4
%int32_1764 = torch.constant.int 32
%int128_1765 = torch.constant.int 128
%1733 = torch.prim.ListConstruct %int4_1763, %294, %int32_1764, %int128_1765 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1734 = torch.aten.view %1718, %1733 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1734, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1766 = torch.constant.int 4
%int8_1767 = torch.constant.int 8
%int128_1768 = torch.constant.int 128
%1735 = torch.prim.ListConstruct %int4_1766, %294, %int8_1767, %int128_1768 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1736 = torch.aten.view %1725, %1735 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1736, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_1769 = torch.constant.int 4
%int8_1770 = torch.constant.int 8
%int128_1771 = torch.constant.int 128
%1737 = torch.prim.ListConstruct %int4_1769, %294, %int8_1770, %int128_1771 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1738 = torch.aten.view %1732, %1737 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1738, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_1772 = torch.constant.int 131072
%none_1773 = torch.constant.none
%none_1774 = torch.constant.none
%cpu_1775 = torch.constant.device "cpu"
%false_1776 = torch.constant.bool false
%1739 = torch.aten.arange %int131072_1772, %none_1773, %none_1774, %cpu_1775, %false_1776 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1777 = torch.constant.int 0
%int128_1778 = torch.constant.int 128
%int2_1779 = torch.constant.int 2
%none_1780 = torch.constant.none
%none_1781 = torch.constant.none
%cpu_1782 = torch.constant.device "cpu"
%false_1783 = torch.constant.bool false
%1740 = torch.aten.arange.start_step %int0_1777, %int128_1778, %int2_1779, %none_1780, %none_1781, %cpu_1782, %false_1783 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1784 = torch.constant.int 0
%int0_1785 = torch.constant.int 0
%int64_1786 = torch.constant.int 64
%int1_1787 = torch.constant.int 1
%1741 = torch.aten.slice.Tensor %1740, %int0_1784, %int0_1785, %int64_1786, %int1_1787 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1788 = torch.constant.int 6
%1742 = torch.prims.convert_element_type %1741, %int6_1788 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1789 = torch.constant.int 128
%1743 = torch.aten.div.Scalar %1742, %int128_1789 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1790 = torch.constant.float 5.000000e+05
%1744 = torch.aten.pow.Scalar %float5.000000e05_1790, %1743 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1745 = torch.aten.reciprocal %1744 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1791 = torch.constant.float 1.000000e+00
%1746 = torch.aten.mul.Scalar %1745, %float1.000000e00_1791 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1792 = torch.constant.int 131072
%int1_1793 = torch.constant.int 1
%1747 = torch.prim.ListConstruct %int131072_1792, %int1_1793 : (!torch.int, !torch.int) -> !torch.list<int>
%1748 = torch.aten.view %1739, %1747 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1749 = torch.aten.mul.Tensor %1748, %1746 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1750 = torch.aten.cos %1749 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1751 = torch.aten.sin %1749 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1752 = torch.aten.complex %1750, %1751 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1794 = torch.constant.int 1
%1753 = torch.aten.size.int %1718, %int1_1794 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_1795 = torch.constant.int 0
%1754 = torch.aten.add.int %int0_1795, %1753 : !torch.int, !torch.int -> !torch.int
%int0_1796 = torch.constant.int 0
%int0_1797 = torch.constant.int 0
%int1_1798 = torch.constant.int 1
%1755 = torch.aten.slice.Tensor %1752, %int0_1796, %int0_1797, %1754, %int1_1798 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1755, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1799 = torch.constant.int 1
%int0_1800 = torch.constant.int 0
%int9223372036854775807_1801 = torch.constant.int 9223372036854775807
%int1_1802 = torch.constant.int 1
%1756 = torch.aten.slice.Tensor %1755, %int1_1799, %int0_1800, %int9223372036854775807_1801, %int1_1802 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1756, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1803 = torch.constant.int 0
%1757 = torch.aten.unsqueeze %1756, %int0_1803 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1757, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1804 = torch.constant.int 2
%1758 = torch.aten.unsqueeze %1757, %int2_1804 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1758, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1805 = torch.constant.int 3
%int0_1806 = torch.constant.int 0
%int9223372036854775807_1807 = torch.constant.int 9223372036854775807
%int1_1808 = torch.constant.int 1
%1759 = torch.aten.slice.Tensor %1758, %int3_1805, %int0_1806, %int9223372036854775807_1807, %int1_1808 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1759, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1760 = torch_c.to_builtin_tensor %1734 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_1809 = arith.constant 1 : index
%dim_1810 = tensor.dim %1760, %c1_1809 : tensor<4x?x32x128xf16>
%1761 = flow.tensor.bitcast %1760 : tensor<4x?x32x128xf16>{%dim_1810} -> tensor<4x?x32x64xcomplex<f16>>{%dim_1810}
%1762 = torch_c.from_builtin_tensor %1761 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %1762, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%1763 = torch.aten.mul.Tensor %1762, %1759 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %1763, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%1764 = torch_c.to_builtin_tensor %1763 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_1811 = arith.constant 1 : index
%dim_1812 = tensor.dim %1764, %c1_1811 : tensor<4x?x32x64xcomplex<f32>>
%1765 = flow.tensor.bitcast %1764 : tensor<4x?x32x64xcomplex<f32>>{%dim_1812} -> tensor<4x?x32x128xf32>{%dim_1812}
%1766 = torch_c.from_builtin_tensor %1765 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %1766, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_1813 = torch.constant.int 5
%1767 = torch.prims.convert_element_type %1766, %int5_1813 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1767, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_1814 = torch.constant.int 131072
%none_1815 = torch.constant.none
%none_1816 = torch.constant.none
%cpu_1817 = torch.constant.device "cpu"
%false_1818 = torch.constant.bool false
%1768 = torch.aten.arange %int131072_1814, %none_1815, %none_1816, %cpu_1817, %false_1818 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_1819 = torch.constant.int 0
%int128_1820 = torch.constant.int 128
%int2_1821 = torch.constant.int 2
%none_1822 = torch.constant.none
%none_1823 = torch.constant.none
%cpu_1824 = torch.constant.device "cpu"
%false_1825 = torch.constant.bool false
%1769 = torch.aten.arange.start_step %int0_1819, %int128_1820, %int2_1821, %none_1822, %none_1823, %cpu_1824, %false_1825 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_1826 = torch.constant.int 0
%int0_1827 = torch.constant.int 0
%int64_1828 = torch.constant.int 64
%int1_1829 = torch.constant.int 1
%1770 = torch.aten.slice.Tensor %1769, %int0_1826, %int0_1827, %int64_1828, %int1_1829 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_1830 = torch.constant.int 6
%1771 = torch.prims.convert_element_type %1770, %int6_1830 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_1831 = torch.constant.int 128
%1772 = torch.aten.div.Scalar %1771, %int128_1831 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_1832 = torch.constant.float 5.000000e+05
%1773 = torch.aten.pow.Scalar %float5.000000e05_1832, %1772 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1774 = torch.aten.reciprocal %1773 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_1833 = torch.constant.float 1.000000e+00
%1775 = torch.aten.mul.Scalar %1774, %float1.000000e00_1833 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_1834 = torch.constant.int 131072
%int1_1835 = torch.constant.int 1
%1776 = torch.prim.ListConstruct %int131072_1834, %int1_1835 : (!torch.int, !torch.int) -> !torch.list<int>
%1777 = torch.aten.view %1768, %1776 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1778 = torch.aten.mul.Tensor %1777, %1775 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1779 = torch.aten.cos %1778 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1780 = torch.aten.sin %1778 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1781 = torch.aten.complex %1779, %1780 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_1836 = torch.constant.int 1
%1782 = torch.aten.size.int %1725, %int1_1836 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_1837 = torch.constant.int 0
%1783 = torch.aten.add.int %int0_1837, %1782 : !torch.int, !torch.int -> !torch.int
%int0_1838 = torch.constant.int 0
%int0_1839 = torch.constant.int 0
%int1_1840 = torch.constant.int 1
%1784 = torch.aten.slice.Tensor %1781, %int0_1838, %int0_1839, %1783, %int1_1840 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1784, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_1841 = torch.constant.int 1
%int0_1842 = torch.constant.int 0
%int9223372036854775807_1843 = torch.constant.int 9223372036854775807
%int1_1844 = torch.constant.int 1
%1785 = torch.aten.slice.Tensor %1784, %int1_1841, %int0_1842, %int9223372036854775807_1843, %int1_1844 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1785, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_1845 = torch.constant.int 0
%1786 = torch.aten.unsqueeze %1785, %int0_1845 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1786, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_1846 = torch.constant.int 2
%1787 = torch.aten.unsqueeze %1786, %int2_1846 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1787, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_1847 = torch.constant.int 3
%int0_1848 = torch.constant.int 0
%int9223372036854775807_1849 = torch.constant.int 9223372036854775807
%int1_1850 = torch.constant.int 1
%1788 = torch.aten.slice.Tensor %1787, %int3_1847, %int0_1848, %int9223372036854775807_1849, %int1_1850 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1788, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1789 = torch_c.to_builtin_tensor %1736 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_1851 = arith.constant 1 : index
%dim_1852 = tensor.dim %1789, %c1_1851 : tensor<4x?x8x128xf16>
%1790 = flow.tensor.bitcast %1789 : tensor<4x?x8x128xf16>{%dim_1852} -> tensor<4x?x8x64xcomplex<f16>>{%dim_1852}
%1791 = torch_c.from_builtin_tensor %1790 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1791, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1792 = torch.aten.mul.Tensor %1791, %1788 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1792, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1793 = torch_c.to_builtin_tensor %1792 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_1853 = arith.constant 1 : index
%dim_1854 = tensor.dim %1793, %c1_1853 : tensor<4x?x8x64xcomplex<f32>>
%1794 = flow.tensor.bitcast %1793 : tensor<4x?x8x64xcomplex<f32>>{%dim_1854} -> tensor<4x?x8x128xf32>{%dim_1854}
%1795 = torch_c.from_builtin_tensor %1794 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1795, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_1855 = torch.constant.int 5
%1796 = torch.prims.convert_element_type %1795, %int5_1855 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1796, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_1856 = torch.constant.int 64
%1797 = torch.aten.mul.Scalar %arg2, %int64_1856 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1797, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int14 = torch.constant.int 14
%int1_1857 = torch.constant.int 1
%1798 = torch.aten.add.Scalar %1797, %int14, %int1_1857 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1798, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1858 = torch.constant.int 4
%int32_1859 = torch.constant.int 32
%int8_1860 = torch.constant.int 8
%int128_1861 = torch.constant.int 128
%1799 = torch.prim.ListConstruct %int4_1858, %425, %int32_1859, %int8_1860, %int128_1861 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1800 = torch.aten.view %1796, %1799 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1800, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1862 = torch.constant.int 4
%1801 = torch.aten.mul.int %int4_1862, %425 : !torch.int, !torch.int -> !torch.int
%int32_1863 = torch.constant.int 32
%int8_1864 = torch.constant.int 8
%int128_1865 = torch.constant.int 128
%1802 = torch.prim.ListConstruct %1801, %int32_1863, %int8_1864, %int128_1865 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1803 = torch.aten.view %1800, %1802 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1803, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1866 = torch.constant.int 4
%1804 = torch.aten.mul.int %int4_1866, %425 : !torch.int, !torch.int -> !torch.int
%1805 = torch.prim.ListConstruct %1804 : (!torch.int) -> !torch.list<int>
%1806 = torch.aten.view %1798, %1805 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1806, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_1867 = torch.constant.int 32
%int2_1868 = torch.constant.int 2
%int32_1869 = torch.constant.int 32
%int8_1870 = torch.constant.int 8
%int128_1871 = torch.constant.int 128
%1807 = torch.prim.ListConstruct %416, %int32_1867, %int2_1868, %int32_1869, %int8_1870, %int128_1871 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1808 = torch.aten.view %1640, %1807 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1808, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1872 = torch.constant.int 32
%1809 = torch.aten.mul.int %416, %int32_1872 : !torch.int, !torch.int -> !torch.int
%int2_1873 = torch.constant.int 2
%1810 = torch.aten.mul.int %1809, %int2_1873 : !torch.int, !torch.int -> !torch.int
%int32_1874 = torch.constant.int 32
%int8_1875 = torch.constant.int 8
%int128_1876 = torch.constant.int 128
%1811 = torch.prim.ListConstruct %1810, %int32_1874, %int8_1875, %int128_1876 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1812 = torch.aten.view %1808, %1811 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1812, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%1813 = torch.prim.ListConstruct %1806 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1877 = torch.constant.bool false
%1814 = torch.aten.index_put %1812, %1813, %1803, %false_1877 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1814, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1878 = torch.constant.int 32
%int2_1879 = torch.constant.int 2
%int32_1880 = torch.constant.int 32
%int8_1881 = torch.constant.int 8
%int128_1882 = torch.constant.int 128
%1815 = torch.prim.ListConstruct %416, %int32_1878, %int2_1879, %int32_1880, %int8_1881, %int128_1882 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1816 = torch.aten.view %1814, %1815 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1816, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1883 = torch.constant.int 2097152
%1817 = torch.prim.ListConstruct %416, %int2097152_1883 : (!torch.int, !torch.int) -> !torch.list<int>
%1818 = torch.aten.view %1816, %1817 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1818, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_1884 = torch.constant.int 32
%int2_1885 = torch.constant.int 2
%int32_1886 = torch.constant.int 32
%int8_1887 = torch.constant.int 8
%int128_1888 = torch.constant.int 128
%1819 = torch.prim.ListConstruct %416, %int32_1884, %int2_1885, %int32_1886, %int8_1887, %int128_1888 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1820 = torch.aten.view %1818, %1819 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1820, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_1889 = torch.constant.int 32
%int8_1890 = torch.constant.int 8
%int128_1891 = torch.constant.int 128
%1821 = torch.prim.ListConstruct %1810, %int32_1889, %int8_1890, %int128_1891 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1822 = torch.aten.view %1820, %1821 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1822, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_1892 = torch.constant.int 4
%int32_1893 = torch.constant.int 32
%int8_1894 = torch.constant.int 8
%int128_1895 = torch.constant.int 128
%1823 = torch.prim.ListConstruct %int4_1892, %425, %int32_1893, %int8_1894, %int128_1895 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1824 = torch.aten.view %1738, %1823 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1824, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_1896 = torch.constant.int 4
%1825 = torch.aten.mul.int %int4_1896, %425 : !torch.int, !torch.int -> !torch.int
%int32_1897 = torch.constant.int 32
%int8_1898 = torch.constant.int 8
%int128_1899 = torch.constant.int 128
%1826 = torch.prim.ListConstruct %1825, %int32_1897, %int8_1898, %int128_1899 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1827 = torch.aten.view %1824, %1826 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1827, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_1900 = torch.constant.int 1
%int1_1901 = torch.constant.int 1
%1828 = torch.aten.add.Scalar %1798, %int1_1900, %int1_1901 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1828, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_1902 = torch.constant.int 4
%1829 = torch.aten.mul.int %int4_1902, %425 : !torch.int, !torch.int -> !torch.int
%1830 = torch.prim.ListConstruct %1829 : (!torch.int) -> !torch.list<int>
%1831 = torch.aten.view %1828, %1830 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %1831, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%1832 = torch.prim.ListConstruct %1831 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_1903 = torch.constant.bool false
%1833 = torch.aten.index_put %1822, %1832, %1827, %false_1903 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %1833, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_1904 = torch.constant.int 32
%int2_1905 = torch.constant.int 2
%int32_1906 = torch.constant.int 32
%int8_1907 = torch.constant.int 8
%int128_1908 = torch.constant.int 128
%1834 = torch.prim.ListConstruct %416, %int32_1904, %int2_1905, %int32_1906, %int8_1907, %int128_1908 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1835 = torch.aten.view %1833, %1834 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %1835, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_1909 = torch.constant.int 2097152
%1836 = torch.prim.ListConstruct %416, %int2097152_1909 : (!torch.int, !torch.int) -> !torch.list<int>
%1837 = torch.aten.view %1835, %1836 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %1837, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_1910 = torch.constant.int -2
%1838 = torch.aten.unsqueeze %1796, %int-2_1910 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1838, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_1911 = torch.constant.int 4
%int8_1912 = torch.constant.int 8
%int4_1913 = torch.constant.int 4
%int128_1914 = torch.constant.int 128
%1839 = torch.prim.ListConstruct %int4_1911, %1782, %int8_1912, %int4_1913, %int128_1914 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1915 = torch.constant.bool false
%1840 = torch.aten.expand %1838, %1839, %false_1915 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1840, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1916 = torch.constant.int 0
%1841 = torch.aten.clone %1840, %int0_1916 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1841, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1917 = torch.constant.int 4
%int32_1918 = torch.constant.int 32
%int128_1919 = torch.constant.int 128
%1842 = torch.prim.ListConstruct %int4_1917, %1782, %int32_1918, %int128_1919 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1843 = torch.aten._unsafe_view %1841, %1842 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1843, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_1920 = torch.constant.int -2
%1844 = torch.aten.unsqueeze %1738, %int-2_1920 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %1844, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_1921 = torch.constant.int 1
%1845 = torch.aten.size.int %1732, %int1_1921 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_1922 = torch.constant.int 4
%int8_1923 = torch.constant.int 8
%int4_1924 = torch.constant.int 4
%int128_1925 = torch.constant.int 128
%1846 = torch.prim.ListConstruct %int4_1922, %1845, %int8_1923, %int4_1924, %int128_1925 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_1926 = torch.constant.bool false
%1847 = torch.aten.expand %1844, %1846, %false_1926 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1847, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_1927 = torch.constant.int 0
%1848 = torch.aten.clone %1847, %int0_1927 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %1848, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_1928 = torch.constant.int 4
%int32_1929 = torch.constant.int 32
%int128_1930 = torch.constant.int 128
%1849 = torch.prim.ListConstruct %int4_1928, %1845, %int32_1929, %int128_1930 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1850 = torch.aten._unsafe_view %1848, %1849 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1850, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_1931 = torch.constant.int 1
%int2_1932 = torch.constant.int 2
%1851 = torch.aten.transpose.int %1767, %int1_1931, %int2_1932 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1851, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1933 = torch.constant.int 1
%int2_1934 = torch.constant.int 2
%1852 = torch.aten.transpose.int %1843, %int1_1933, %int2_1934 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1852, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1935 = torch.constant.int 1
%int2_1936 = torch.constant.int 2
%1853 = torch.aten.transpose.int %1850, %int1_1935, %int2_1936 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %1853, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_1937 = torch.constant.float 0.000000e+00
%false_1938 = torch.constant.bool false
%none_1939 = torch.constant.none
%1854:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%1851, %1852, %1853, %float0.000000e00_1937, %false_1938, %320, %none_1939) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %1854#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_1940 = torch.constant.int 1
%int2_1941 = torch.constant.int 2
%1855 = torch.aten.transpose.int %1854#0, %int1_1940, %int2_1941 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1855, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_1942 = torch.constant.int 4
%int4096_1943 = torch.constant.int 4096
%1856 = torch.prim.ListConstruct %int4_1942, %1753, %int4096_1943 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1857 = torch.aten.view %1855, %1856 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1857, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1944 = torch.constant.int -2
%int-1_1945 = torch.constant.int -1
%1858 = torch.aten.transpose.int %68, %int-2_1944, %int-1_1945 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1946 = torch.constant.int 4
%1859 = torch.aten.mul.int %int4_1946, %1753 : !torch.int, !torch.int -> !torch.int
%int4096_1947 = torch.constant.int 4096
%1860 = torch.prim.ListConstruct %1859, %int4096_1947 : (!torch.int, !torch.int) -> !torch.list<int>
%1861 = torch.aten.view %1857, %1860 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1861, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1862 = torch.aten.mm %1861, %1858 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1862, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1948 = torch.constant.int 4
%int4096_1949 = torch.constant.int 4096
%1863 = torch.prim.ListConstruct %int4_1948, %1753, %int4096_1949 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1864 = torch.aten.view %1862, %1863 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1864, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1950 = torch.constant.int 1
%1865 = torch.aten.add.Tensor %1702, %1864, %int1_1950 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1865, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1951 = torch.constant.int 6
%1866 = torch.prims.convert_element_type %1865, %int6_1951 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1866, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1952 = torch.constant.int 2
%1867 = torch.aten.pow.Tensor_Scalar %1866, %int2_1952 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1867, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1953 = torch.constant.int -1
%1868 = torch.prim.ListConstruct %int-1_1953 : (!torch.int) -> !torch.list<int>
%true_1954 = torch.constant.bool true
%none_1955 = torch.constant.none
%1869 = torch.aten.mean.dim %1867, %1868, %true_1954, %none_1955 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1869, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1956 = torch.constant.float 9.9999997473787516E-6
%int1_1957 = torch.constant.int 1
%1870 = torch.aten.add.Scalar %1869, %float9.999990e-06_1956, %int1_1957 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1870, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1871 = torch.aten.rsqrt %1870 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1871, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1872 = torch.aten.mul.Tensor %1866, %1871 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1872, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1873 = torch.aten.mul.Tensor %69, %1872 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1873, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1958 = torch.constant.int 5
%1874 = torch.prims.convert_element_type %1873, %int5_1958 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1874, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1959 = torch.constant.int -2
%int-1_1960 = torch.constant.int -1
%1875 = torch.aten.transpose.int %70, %int-2_1959, %int-1_1960 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1961 = torch.constant.int 4
%1876 = torch.aten.mul.int %int4_1961, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1962 = torch.constant.int 4096
%1877 = torch.prim.ListConstruct %1876, %int4096_1962 : (!torch.int, !torch.int) -> !torch.list<int>
%1878 = torch.aten.view %1874, %1877 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1878, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1879 = torch.aten.mm %1878, %1875 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1879, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1963 = torch.constant.int 4
%int14336_1964 = torch.constant.int 14336
%1880 = torch.prim.ListConstruct %int4_1963, %294, %int14336_1964 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1881 = torch.aten.view %1879, %1880 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1881, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1882 = torch.aten.silu %1881 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1882, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1965 = torch.constant.int -2
%int-1_1966 = torch.constant.int -1
%1883 = torch.aten.transpose.int %71, %int-2_1965, %int-1_1966 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_1967 = torch.constant.int 4
%1884 = torch.aten.mul.int %int4_1967, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1968 = torch.constant.int 4096
%1885 = torch.prim.ListConstruct %1884, %int4096_1968 : (!torch.int, !torch.int) -> !torch.list<int>
%1886 = torch.aten.view %1874, %1885 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1886, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1887 = torch.aten.mm %1886, %1883 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1887, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_1969 = torch.constant.int 4
%int14336_1970 = torch.constant.int 14336
%1888 = torch.prim.ListConstruct %int4_1969, %294, %int14336_1970 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1889 = torch.aten.view %1887, %1888 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1889, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%1890 = torch.aten.mul.Tensor %1882, %1889 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %1890, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_1971 = torch.constant.int -2
%int-1_1972 = torch.constant.int -1
%1891 = torch.aten.transpose.int %72, %int-2_1971, %int-1_1972 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_1973 = torch.constant.int 1
%1892 = torch.aten.size.int %1881, %int1_1973 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_1974 = torch.constant.int 4
%1893 = torch.aten.mul.int %int4_1974, %1892 : !torch.int, !torch.int -> !torch.int
%int14336_1975 = torch.constant.int 14336
%1894 = torch.prim.ListConstruct %1893, %int14336_1975 : (!torch.int, !torch.int) -> !torch.list<int>
%1895 = torch.aten.view %1890, %1894 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %1895, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%1896 = torch.aten.mm %1895, %1891 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1896, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1976 = torch.constant.int 4
%int4096_1977 = torch.constant.int 4096
%1897 = torch.prim.ListConstruct %int4_1976, %1892, %int4096_1977 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1898 = torch.aten.view %1896, %1897 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1898, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_1978 = torch.constant.int 1
%1899 = torch.aten.add.Tensor %1865, %1898, %int1_1978 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1899, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_1979 = torch.constant.int 6
%1900 = torch.prims.convert_element_type %1899, %int6_1979 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1900, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_1980 = torch.constant.int 2
%1901 = torch.aten.pow.Tensor_Scalar %1900, %int2_1980 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1901, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_1981 = torch.constant.int -1
%1902 = torch.prim.ListConstruct %int-1_1981 : (!torch.int) -> !torch.list<int>
%true_1982 = torch.constant.bool true
%none_1983 = torch.constant.none
%1903 = torch.aten.mean.dim %1901, %1902, %true_1982, %none_1983 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1903, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_1984 = torch.constant.float 9.9999997473787516E-6
%int1_1985 = torch.constant.int 1
%1904 = torch.aten.add.Scalar %1903, %float9.999990e-06_1984, %int1_1985 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1904, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1905 = torch.aten.rsqrt %1904 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %1905, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%1906 = torch.aten.mul.Tensor %1900, %1905 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1906, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%1907 = torch.aten.mul.Tensor %73, %1906 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %1907, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_1986 = torch.constant.int 5
%1908 = torch.prims.convert_element_type %1907, %int5_1986 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1908, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1987 = torch.constant.int -2
%int-1_1988 = torch.constant.int -1
%1909 = torch.aten.transpose.int %74, %int-2_1987, %int-1_1988 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_1989 = torch.constant.int 4
%1910 = torch.aten.mul.int %int4_1989, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1990 = torch.constant.int 4096
%1911 = torch.prim.ListConstruct %1910, %int4096_1990 : (!torch.int, !torch.int) -> !torch.list<int>
%1912 = torch.aten.view %1908, %1911 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1912, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1913 = torch.aten.mm %1912, %1909 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1913, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_1991 = torch.constant.int 4
%int4096_1992 = torch.constant.int 4096
%1914 = torch.prim.ListConstruct %int4_1991, %294, %int4096_1992 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1915 = torch.aten.view %1913, %1914 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %1915, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_1993 = torch.constant.int -2
%int-1_1994 = torch.constant.int -1
%1916 = torch.aten.transpose.int %75, %int-2_1993, %int-1_1994 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_1995 = torch.constant.int 4
%1917 = torch.aten.mul.int %int4_1995, %294 : !torch.int, !torch.int -> !torch.int
%int4096_1996 = torch.constant.int 4096
%1918 = torch.prim.ListConstruct %1917, %int4096_1996 : (!torch.int, !torch.int) -> !torch.list<int>
%1919 = torch.aten.view %1908, %1918 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1919, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1920 = torch.aten.mm %1919, %1916 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1920, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_1997 = torch.constant.int 4
%int1024_1998 = torch.constant.int 1024
%1921 = torch.prim.ListConstruct %int4_1997, %294, %int1024_1998 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1922 = torch.aten.view %1920, %1921 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1922, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_1999 = torch.constant.int -2
%int-1_2000 = torch.constant.int -1
%1923 = torch.aten.transpose.int %76, %int-2_1999, %int-1_2000 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2001 = torch.constant.int 4
%1924 = torch.aten.mul.int %int4_2001, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2002 = torch.constant.int 4096
%1925 = torch.prim.ListConstruct %1924, %int4096_2002 : (!torch.int, !torch.int) -> !torch.list<int>
%1926 = torch.aten.view %1908, %1925 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %1926, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%1927 = torch.aten.mm %1926, %1923 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %1927, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2003 = torch.constant.int 4
%int1024_2004 = torch.constant.int 1024
%1928 = torch.prim.ListConstruct %int4_2003, %294, %int1024_2004 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1929 = torch.aten.view %1927, %1928 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %1929, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_2005 = torch.constant.int 4
%int32_2006 = torch.constant.int 32
%int128_2007 = torch.constant.int 128
%1930 = torch.prim.ListConstruct %int4_2005, %294, %int32_2006, %int128_2007 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1931 = torch.aten.view %1915, %1930 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1931, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2008 = torch.constant.int 4
%int8_2009 = torch.constant.int 8
%int128_2010 = torch.constant.int 128
%1932 = torch.prim.ListConstruct %int4_2008, %294, %int8_2009, %int128_2010 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1933 = torch.aten.view %1922, %1932 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1933, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_2011 = torch.constant.int 4
%int8_2012 = torch.constant.int 8
%int128_2013 = torch.constant.int 128
%1934 = torch.prim.ListConstruct %int4_2011, %294, %int8_2012, %int128_2013 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1935 = torch.aten.view %1929, %1934 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1935, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_2014 = torch.constant.int 131072
%none_2015 = torch.constant.none
%none_2016 = torch.constant.none
%cpu_2017 = torch.constant.device "cpu"
%false_2018 = torch.constant.bool false
%1936 = torch.aten.arange %int131072_2014, %none_2015, %none_2016, %cpu_2017, %false_2018 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2019 = torch.constant.int 0
%int128_2020 = torch.constant.int 128
%int2_2021 = torch.constant.int 2
%none_2022 = torch.constant.none
%none_2023 = torch.constant.none
%cpu_2024 = torch.constant.device "cpu"
%false_2025 = torch.constant.bool false
%1937 = torch.aten.arange.start_step %int0_2019, %int128_2020, %int2_2021, %none_2022, %none_2023, %cpu_2024, %false_2025 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2026 = torch.constant.int 0
%int0_2027 = torch.constant.int 0
%int64_2028 = torch.constant.int 64
%int1_2029 = torch.constant.int 1
%1938 = torch.aten.slice.Tensor %1937, %int0_2026, %int0_2027, %int64_2028, %int1_2029 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2030 = torch.constant.int 6
%1939 = torch.prims.convert_element_type %1938, %int6_2030 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2031 = torch.constant.int 128
%1940 = torch.aten.div.Scalar %1939, %int128_2031 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2032 = torch.constant.float 5.000000e+05
%1941 = torch.aten.pow.Scalar %float5.000000e05_2032, %1940 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1942 = torch.aten.reciprocal %1941 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2033 = torch.constant.float 1.000000e+00
%1943 = torch.aten.mul.Scalar %1942, %float1.000000e00_2033 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2034 = torch.constant.int 131072
%int1_2035 = torch.constant.int 1
%1944 = torch.prim.ListConstruct %int131072_2034, %int1_2035 : (!torch.int, !torch.int) -> !torch.list<int>
%1945 = torch.aten.view %1936, %1944 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1946 = torch.aten.mul.Tensor %1945, %1943 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1947 = torch.aten.cos %1946 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1948 = torch.aten.sin %1946 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1949 = torch.aten.complex %1947, %1948 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2036 = torch.constant.int 1
%1950 = torch.aten.size.int %1915, %int1_2036 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_2037 = torch.constant.int 0
%1951 = torch.aten.add.int %int0_2037, %1950 : !torch.int, !torch.int -> !torch.int
%int0_2038 = torch.constant.int 0
%int0_2039 = torch.constant.int 0
%int1_2040 = torch.constant.int 1
%1952 = torch.aten.slice.Tensor %1949, %int0_2038, %int0_2039, %1951, %int1_2040 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1952, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2041 = torch.constant.int 1
%int0_2042 = torch.constant.int 0
%int9223372036854775807_2043 = torch.constant.int 9223372036854775807
%int1_2044 = torch.constant.int 1
%1953 = torch.aten.slice.Tensor %1952, %int1_2041, %int0_2042, %int9223372036854775807_2043, %int1_2044 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1953, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2045 = torch.constant.int 0
%1954 = torch.aten.unsqueeze %1953, %int0_2045 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1954, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2046 = torch.constant.int 2
%1955 = torch.aten.unsqueeze %1954, %int2_2046 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1955, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2047 = torch.constant.int 3
%int0_2048 = torch.constant.int 0
%int9223372036854775807_2049 = torch.constant.int 9223372036854775807
%int1_2050 = torch.constant.int 1
%1956 = torch.aten.slice.Tensor %1955, %int3_2047, %int0_2048, %int9223372036854775807_2049, %int1_2050 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1956, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1957 = torch_c.to_builtin_tensor %1931 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_2051 = arith.constant 1 : index
%dim_2052 = tensor.dim %1957, %c1_2051 : tensor<4x?x32x128xf16>
%1958 = flow.tensor.bitcast %1957 : tensor<4x?x32x128xf16>{%dim_2052} -> tensor<4x?x32x64xcomplex<f16>>{%dim_2052}
%1959 = torch_c.from_builtin_tensor %1958 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %1959, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%1960 = torch.aten.mul.Tensor %1959, %1956 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %1960, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%1961 = torch_c.to_builtin_tensor %1960 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_2053 = arith.constant 1 : index
%dim_2054 = tensor.dim %1961, %c1_2053 : tensor<4x?x32x64xcomplex<f32>>
%1962 = flow.tensor.bitcast %1961 : tensor<4x?x32x64xcomplex<f32>>{%dim_2054} -> tensor<4x?x32x128xf32>{%dim_2054}
%1963 = torch_c.from_builtin_tensor %1962 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %1963, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_2055 = torch.constant.int 5
%1964 = torch.prims.convert_element_type %1963, %int5_2055 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %1964, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_2056 = torch.constant.int 131072
%none_2057 = torch.constant.none
%none_2058 = torch.constant.none
%cpu_2059 = torch.constant.device "cpu"
%false_2060 = torch.constant.bool false
%1965 = torch.aten.arange %int131072_2056, %none_2057, %none_2058, %cpu_2059, %false_2060 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2061 = torch.constant.int 0
%int128_2062 = torch.constant.int 128
%int2_2063 = torch.constant.int 2
%none_2064 = torch.constant.none
%none_2065 = torch.constant.none
%cpu_2066 = torch.constant.device "cpu"
%false_2067 = torch.constant.bool false
%1966 = torch.aten.arange.start_step %int0_2061, %int128_2062, %int2_2063, %none_2064, %none_2065, %cpu_2066, %false_2067 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2068 = torch.constant.int 0
%int0_2069 = torch.constant.int 0
%int64_2070 = torch.constant.int 64
%int1_2071 = torch.constant.int 1
%1967 = torch.aten.slice.Tensor %1966, %int0_2068, %int0_2069, %int64_2070, %int1_2071 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2072 = torch.constant.int 6
%1968 = torch.prims.convert_element_type %1967, %int6_2072 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2073 = torch.constant.int 128
%1969 = torch.aten.div.Scalar %1968, %int128_2073 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2074 = torch.constant.float 5.000000e+05
%1970 = torch.aten.pow.Scalar %float5.000000e05_2074, %1969 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%1971 = torch.aten.reciprocal %1970 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2075 = torch.constant.float 1.000000e+00
%1972 = torch.aten.mul.Scalar %1971, %float1.000000e00_2075 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2076 = torch.constant.int 131072
%int1_2077 = torch.constant.int 1
%1973 = torch.prim.ListConstruct %int131072_2076, %int1_2077 : (!torch.int, !torch.int) -> !torch.list<int>
%1974 = torch.aten.view %1965, %1973 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%1975 = torch.aten.mul.Tensor %1974, %1972 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%1976 = torch.aten.cos %1975 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1977 = torch.aten.sin %1975 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%1978 = torch.aten.complex %1976, %1977 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2078 = torch.constant.int 1
%1979 = torch.aten.size.int %1922, %int1_2078 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_2079 = torch.constant.int 0
%1980 = torch.aten.add.int %int0_2079, %1979 : !torch.int, !torch.int -> !torch.int
%int0_2080 = torch.constant.int 0
%int0_2081 = torch.constant.int 0
%int1_2082 = torch.constant.int 1
%1981 = torch.aten.slice.Tensor %1978, %int0_2080, %int0_2081, %1980, %int1_2082 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1981, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2083 = torch.constant.int 1
%int0_2084 = torch.constant.int 0
%int9223372036854775807_2085 = torch.constant.int 9223372036854775807
%int1_2086 = torch.constant.int 1
%1982 = torch.aten.slice.Tensor %1981, %int1_2083, %int0_2084, %int9223372036854775807_2085, %int1_2086 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %1982, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2087 = torch.constant.int 0
%1983 = torch.aten.unsqueeze %1982, %int0_2087 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %1983, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2088 = torch.constant.int 2
%1984 = torch.aten.unsqueeze %1983, %int2_2088 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1984, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2089 = torch.constant.int 3
%int0_2090 = torch.constant.int 0
%int9223372036854775807_2091 = torch.constant.int 9223372036854775807
%int1_2092 = torch.constant.int 1
%1985 = torch.aten.slice.Tensor %1984, %int3_2089, %int0_2090, %int9223372036854775807_2091, %int1_2092 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %1985, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%1986 = torch_c.to_builtin_tensor %1933 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_2093 = arith.constant 1 : index
%dim_2094 = tensor.dim %1986, %c1_2093 : tensor<4x?x8x128xf16>
%1987 = flow.tensor.bitcast %1986 : tensor<4x?x8x128xf16>{%dim_2094} -> tensor<4x?x8x64xcomplex<f16>>{%dim_2094}
%1988 = torch_c.from_builtin_tensor %1987 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %1988, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%1989 = torch.aten.mul.Tensor %1988, %1985 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %1989, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%1990 = torch_c.to_builtin_tensor %1989 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_2095 = arith.constant 1 : index
%dim_2096 = tensor.dim %1990, %c1_2095 : tensor<4x?x8x64xcomplex<f32>>
%1991 = flow.tensor.bitcast %1990 : tensor<4x?x8x64xcomplex<f32>>{%dim_2096} -> tensor<4x?x8x128xf32>{%dim_2096}
%1992 = torch_c.from_builtin_tensor %1991 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %1992, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_2097 = torch.constant.int 5
%1993 = torch.prims.convert_element_type %1992, %int5_2097 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %1993, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_2098 = torch.constant.int 64
%1994 = torch.aten.mul.Scalar %arg2, %int64_2098 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1994, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int16 = torch.constant.int 16
%int1_2099 = torch.constant.int 1
%1995 = torch.aten.add.Scalar %1994, %int16, %int1_2099 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %1995, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2100 = torch.constant.int 4
%int32_2101 = torch.constant.int 32
%int8_2102 = torch.constant.int 8
%int128_2103 = torch.constant.int 128
%1996 = torch.prim.ListConstruct %int4_2100, %425, %int32_2101, %int8_2102, %int128_2103 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%1997 = torch.aten.view %1993, %1996 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %1997, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2104 = torch.constant.int 4
%1998 = torch.aten.mul.int %int4_2104, %425 : !torch.int, !torch.int -> !torch.int
%int32_2105 = torch.constant.int 32
%int8_2106 = torch.constant.int 8
%int128_2107 = torch.constant.int 128
%1999 = torch.prim.ListConstruct %1998, %int32_2105, %int8_2106, %int128_2107 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2000 = torch.aten.view %1997, %1999 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2000, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2108 = torch.constant.int 4
%2001 = torch.aten.mul.int %int4_2108, %425 : !torch.int, !torch.int -> !torch.int
%2002 = torch.prim.ListConstruct %2001 : (!torch.int) -> !torch.list<int>
%2003 = torch.aten.view %1995, %2002 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2003, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_2109 = torch.constant.int 32
%int2_2110 = torch.constant.int 2
%int32_2111 = torch.constant.int 32
%int8_2112 = torch.constant.int 8
%int128_2113 = torch.constant.int 128
%2004 = torch.prim.ListConstruct %416, %int32_2109, %int2_2110, %int32_2111, %int8_2112, %int128_2113 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2005 = torch.aten.view %1837, %2004 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2005, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2114 = torch.constant.int 32
%2006 = torch.aten.mul.int %416, %int32_2114 : !torch.int, !torch.int -> !torch.int
%int2_2115 = torch.constant.int 2
%2007 = torch.aten.mul.int %2006, %int2_2115 : !torch.int, !torch.int -> !torch.int
%int32_2116 = torch.constant.int 32
%int8_2117 = torch.constant.int 8
%int128_2118 = torch.constant.int 128
%2008 = torch.prim.ListConstruct %2007, %int32_2116, %int8_2117, %int128_2118 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2009 = torch.aten.view %2005, %2008 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2009, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2010 = torch.prim.ListConstruct %2003 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2119 = torch.constant.bool false
%2011 = torch.aten.index_put %2009, %2010, %2000, %false_2119 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2011, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2120 = torch.constant.int 32
%int2_2121 = torch.constant.int 2
%int32_2122 = torch.constant.int 32
%int8_2123 = torch.constant.int 8
%int128_2124 = torch.constant.int 128
%2012 = torch.prim.ListConstruct %416, %int32_2120, %int2_2121, %int32_2122, %int8_2123, %int128_2124 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2013 = torch.aten.view %2011, %2012 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2013, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2125 = torch.constant.int 2097152
%2014 = torch.prim.ListConstruct %416, %int2097152_2125 : (!torch.int, !torch.int) -> !torch.list<int>
%2015 = torch.aten.view %2013, %2014 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2015, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_2126 = torch.constant.int 32
%int2_2127 = torch.constant.int 2
%int32_2128 = torch.constant.int 32
%int8_2129 = torch.constant.int 8
%int128_2130 = torch.constant.int 128
%2016 = torch.prim.ListConstruct %416, %int32_2126, %int2_2127, %int32_2128, %int8_2129, %int128_2130 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2017 = torch.aten.view %2015, %2016 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2017, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2131 = torch.constant.int 32
%int8_2132 = torch.constant.int 8
%int128_2133 = torch.constant.int 128
%2018 = torch.prim.ListConstruct %2007, %int32_2131, %int8_2132, %int128_2133 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2019 = torch.aten.view %2017, %2018 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2019, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2134 = torch.constant.int 4
%int32_2135 = torch.constant.int 32
%int8_2136 = torch.constant.int 8
%int128_2137 = torch.constant.int 128
%2020 = torch.prim.ListConstruct %int4_2134, %425, %int32_2135, %int8_2136, %int128_2137 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2021 = torch.aten.view %1935, %2020 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2021, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2138 = torch.constant.int 4
%2022 = torch.aten.mul.int %int4_2138, %425 : !torch.int, !torch.int -> !torch.int
%int32_2139 = torch.constant.int 32
%int8_2140 = torch.constant.int 8
%int128_2141 = torch.constant.int 128
%2023 = torch.prim.ListConstruct %2022, %int32_2139, %int8_2140, %int128_2141 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2024 = torch.aten.view %2021, %2023 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2024, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_2142 = torch.constant.int 1
%int1_2143 = torch.constant.int 1
%2025 = torch.aten.add.Scalar %1995, %int1_2142, %int1_2143 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2025, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2144 = torch.constant.int 4
%2026 = torch.aten.mul.int %int4_2144, %425 : !torch.int, !torch.int -> !torch.int
%2027 = torch.prim.ListConstruct %2026 : (!torch.int) -> !torch.list<int>
%2028 = torch.aten.view %2025, %2027 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2028, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%2029 = torch.prim.ListConstruct %2028 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2145 = torch.constant.bool false
%2030 = torch.aten.index_put %2019, %2029, %2024, %false_2145 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2030, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2146 = torch.constant.int 32
%int2_2147 = torch.constant.int 2
%int32_2148 = torch.constant.int 32
%int8_2149 = torch.constant.int 8
%int128_2150 = torch.constant.int 128
%2031 = torch.prim.ListConstruct %416, %int32_2146, %int2_2147, %int32_2148, %int8_2149, %int128_2150 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2032 = torch.aten.view %2030, %2031 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2032, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2151 = torch.constant.int 2097152
%2033 = torch.prim.ListConstruct %416, %int2097152_2151 : (!torch.int, !torch.int) -> !torch.list<int>
%2034 = torch.aten.view %2032, %2033 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2034, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_2152 = torch.constant.int -2
%2035 = torch.aten.unsqueeze %1993, %int-2_2152 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2035, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_2153 = torch.constant.int 4
%int8_2154 = torch.constant.int 8
%int4_2155 = torch.constant.int 4
%int128_2156 = torch.constant.int 128
%2036 = torch.prim.ListConstruct %int4_2153, %1979, %int8_2154, %int4_2155, %int128_2156 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2157 = torch.constant.bool false
%2037 = torch.aten.expand %2035, %2036, %false_2157 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2037, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2158 = torch.constant.int 0
%2038 = torch.aten.clone %2037, %int0_2158 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2038, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2159 = torch.constant.int 4
%int32_2160 = torch.constant.int 32
%int128_2161 = torch.constant.int 128
%2039 = torch.prim.ListConstruct %int4_2159, %1979, %int32_2160, %int128_2161 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2040 = torch.aten._unsafe_view %2038, %2039 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2040, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_2162 = torch.constant.int -2
%2041 = torch.aten.unsqueeze %1935, %int-2_2162 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2041, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_2163 = torch.constant.int 1
%2042 = torch.aten.size.int %1929, %int1_2163 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_2164 = torch.constant.int 4
%int8_2165 = torch.constant.int 8
%int4_2166 = torch.constant.int 4
%int128_2167 = torch.constant.int 128
%2043 = torch.prim.ListConstruct %int4_2164, %2042, %int8_2165, %int4_2166, %int128_2167 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2168 = torch.constant.bool false
%2044 = torch.aten.expand %2041, %2043, %false_2168 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2044, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2169 = torch.constant.int 0
%2045 = torch.aten.clone %2044, %int0_2169 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2045, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2170 = torch.constant.int 4
%int32_2171 = torch.constant.int 32
%int128_2172 = torch.constant.int 128
%2046 = torch.prim.ListConstruct %int4_2170, %2042, %int32_2171, %int128_2172 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2047 = torch.aten._unsafe_view %2045, %2046 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2047, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_2173 = torch.constant.int 1
%int2_2174 = torch.constant.int 2
%2048 = torch.aten.transpose.int %1964, %int1_2173, %int2_2174 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2048, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2175 = torch.constant.int 1
%int2_2176 = torch.constant.int 2
%2049 = torch.aten.transpose.int %2040, %int1_2175, %int2_2176 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2049, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2177 = torch.constant.int 1
%int2_2178 = torch.constant.int 2
%2050 = torch.aten.transpose.int %2047, %int1_2177, %int2_2178 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2050, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_2179 = torch.constant.float 0.000000e+00
%false_2180 = torch.constant.bool false
%none_2181 = torch.constant.none
%2051:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%2048, %2049, %2050, %float0.000000e00_2179, %false_2180, %320, %none_2181) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %2051#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2182 = torch.constant.int 1
%int2_2183 = torch.constant.int 2
%2052 = torch.aten.transpose.int %2051#0, %int1_2182, %int2_2183 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2052, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2184 = torch.constant.int 4
%int4096_2185 = torch.constant.int 4096
%2053 = torch.prim.ListConstruct %int4_2184, %1950, %int4096_2185 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2054 = torch.aten.view %2052, %2053 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2054, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2186 = torch.constant.int -2
%int-1_2187 = torch.constant.int -1
%2055 = torch.aten.transpose.int %77, %int-2_2186, %int-1_2187 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2188 = torch.constant.int 4
%2056 = torch.aten.mul.int %int4_2188, %1950 : !torch.int, !torch.int -> !torch.int
%int4096_2189 = torch.constant.int 4096
%2057 = torch.prim.ListConstruct %2056, %int4096_2189 : (!torch.int, !torch.int) -> !torch.list<int>
%2058 = torch.aten.view %2054, %2057 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2058, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2059 = torch.aten.mm %2058, %2055 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2059, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2190 = torch.constant.int 4
%int4096_2191 = torch.constant.int 4096
%2060 = torch.prim.ListConstruct %int4_2190, %1950, %int4096_2191 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2061 = torch.aten.view %2059, %2060 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2061, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2192 = torch.constant.int 1
%2062 = torch.aten.add.Tensor %1899, %2061, %int1_2192 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2062, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2193 = torch.constant.int 6
%2063 = torch.prims.convert_element_type %2062, %int6_2193 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2063, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2194 = torch.constant.int 2
%2064 = torch.aten.pow.Tensor_Scalar %2063, %int2_2194 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2064, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2195 = torch.constant.int -1
%2065 = torch.prim.ListConstruct %int-1_2195 : (!torch.int) -> !torch.list<int>
%true_2196 = torch.constant.bool true
%none_2197 = torch.constant.none
%2066 = torch.aten.mean.dim %2064, %2065, %true_2196, %none_2197 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2066, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2198 = torch.constant.float 9.9999997473787516E-6
%int1_2199 = torch.constant.int 1
%2067 = torch.aten.add.Scalar %2066, %float9.999990e-06_2198, %int1_2199 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2067, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2068 = torch.aten.rsqrt %2067 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2068, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2069 = torch.aten.mul.Tensor %2063, %2068 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2069, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2070 = torch.aten.mul.Tensor %78, %2069 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2070, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2200 = torch.constant.int 5
%2071 = torch.prims.convert_element_type %2070, %int5_2200 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2071, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2201 = torch.constant.int -2
%int-1_2202 = torch.constant.int -1
%2072 = torch.aten.transpose.int %79, %int-2_2201, %int-1_2202 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2203 = torch.constant.int 4
%2073 = torch.aten.mul.int %int4_2203, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2204 = torch.constant.int 4096
%2074 = torch.prim.ListConstruct %2073, %int4096_2204 : (!torch.int, !torch.int) -> !torch.list<int>
%2075 = torch.aten.view %2071, %2074 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2075, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2076 = torch.aten.mm %2075, %2072 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2076, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2205 = torch.constant.int 4
%int14336_2206 = torch.constant.int 14336
%2077 = torch.prim.ListConstruct %int4_2205, %294, %int14336_2206 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2078 = torch.aten.view %2076, %2077 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2078, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2079 = torch.aten.silu %2078 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2079, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2207 = torch.constant.int -2
%int-1_2208 = torch.constant.int -1
%2080 = torch.aten.transpose.int %80, %int-2_2207, %int-1_2208 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2209 = torch.constant.int 4
%2081 = torch.aten.mul.int %int4_2209, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2210 = torch.constant.int 4096
%2082 = torch.prim.ListConstruct %2081, %int4096_2210 : (!torch.int, !torch.int) -> !torch.list<int>
%2083 = torch.aten.view %2071, %2082 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2083, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2084 = torch.aten.mm %2083, %2080 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2084, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2211 = torch.constant.int 4
%int14336_2212 = torch.constant.int 14336
%2085 = torch.prim.ListConstruct %int4_2211, %294, %int14336_2212 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2086 = torch.aten.view %2084, %2085 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2086, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2087 = torch.aten.mul.Tensor %2079, %2086 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2087, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2213 = torch.constant.int -2
%int-1_2214 = torch.constant.int -1
%2088 = torch.aten.transpose.int %81, %int-2_2213, %int-1_2214 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_2215 = torch.constant.int 1
%2089 = torch.aten.size.int %2078, %int1_2215 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_2216 = torch.constant.int 4
%2090 = torch.aten.mul.int %int4_2216, %2089 : !torch.int, !torch.int -> !torch.int
%int14336_2217 = torch.constant.int 14336
%2091 = torch.prim.ListConstruct %2090, %int14336_2217 : (!torch.int, !torch.int) -> !torch.list<int>
%2092 = torch.aten.view %2087, %2091 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2092, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%2093 = torch.aten.mm %2092, %2088 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2093, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2218 = torch.constant.int 4
%int4096_2219 = torch.constant.int 4096
%2094 = torch.prim.ListConstruct %int4_2218, %2089, %int4096_2219 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2095 = torch.aten.view %2093, %2094 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2095, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2220 = torch.constant.int 1
%2096 = torch.aten.add.Tensor %2062, %2095, %int1_2220 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2096, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2221 = torch.constant.int 6
%2097 = torch.prims.convert_element_type %2096, %int6_2221 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2097, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2222 = torch.constant.int 2
%2098 = torch.aten.pow.Tensor_Scalar %2097, %int2_2222 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2098, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2223 = torch.constant.int -1
%2099 = torch.prim.ListConstruct %int-1_2223 : (!torch.int) -> !torch.list<int>
%true_2224 = torch.constant.bool true
%none_2225 = torch.constant.none
%2100 = torch.aten.mean.dim %2098, %2099, %true_2224, %none_2225 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2100, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2226 = torch.constant.float 9.9999997473787516E-6
%int1_2227 = torch.constant.int 1
%2101 = torch.aten.add.Scalar %2100, %float9.999990e-06_2226, %int1_2227 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2101, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2102 = torch.aten.rsqrt %2101 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2102, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2103 = torch.aten.mul.Tensor %2097, %2102 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2103, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2104 = torch.aten.mul.Tensor %82, %2103 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2104, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2228 = torch.constant.int 5
%2105 = torch.prims.convert_element_type %2104, %int5_2228 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2105, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2229 = torch.constant.int -2
%int-1_2230 = torch.constant.int -1
%2106 = torch.aten.transpose.int %83, %int-2_2229, %int-1_2230 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2231 = torch.constant.int 4
%2107 = torch.aten.mul.int %int4_2231, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2232 = torch.constant.int 4096
%2108 = torch.prim.ListConstruct %2107, %int4096_2232 : (!torch.int, !torch.int) -> !torch.list<int>
%2109 = torch.aten.view %2105, %2108 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2109, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2110 = torch.aten.mm %2109, %2106 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2110, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2233 = torch.constant.int 4
%int4096_2234 = torch.constant.int 4096
%2111 = torch.prim.ListConstruct %int4_2233, %294, %int4096_2234 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2112 = torch.aten.view %2110, %2111 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2112, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2235 = torch.constant.int -2
%int-1_2236 = torch.constant.int -1
%2113 = torch.aten.transpose.int %84, %int-2_2235, %int-1_2236 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2237 = torch.constant.int 4
%2114 = torch.aten.mul.int %int4_2237, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2238 = torch.constant.int 4096
%2115 = torch.prim.ListConstruct %2114, %int4096_2238 : (!torch.int, !torch.int) -> !torch.list<int>
%2116 = torch.aten.view %2105, %2115 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2116, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2117 = torch.aten.mm %2116, %2113 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2117, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2239 = torch.constant.int 4
%int1024_2240 = torch.constant.int 1024
%2118 = torch.prim.ListConstruct %int4_2239, %294, %int1024_2240 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2119 = torch.aten.view %2117, %2118 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2119, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_2241 = torch.constant.int -2
%int-1_2242 = torch.constant.int -1
%2120 = torch.aten.transpose.int %85, %int-2_2241, %int-1_2242 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2243 = torch.constant.int 4
%2121 = torch.aten.mul.int %int4_2243, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2244 = torch.constant.int 4096
%2122 = torch.prim.ListConstruct %2121, %int4096_2244 : (!torch.int, !torch.int) -> !torch.list<int>
%2123 = torch.aten.view %2105, %2122 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2123, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2124 = torch.aten.mm %2123, %2120 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2124, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2245 = torch.constant.int 4
%int1024_2246 = torch.constant.int 1024
%2125 = torch.prim.ListConstruct %int4_2245, %294, %int1024_2246 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2126 = torch.aten.view %2124, %2125 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2126, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_2247 = torch.constant.int 4
%int32_2248 = torch.constant.int 32
%int128_2249 = torch.constant.int 128
%2127 = torch.prim.ListConstruct %int4_2247, %294, %int32_2248, %int128_2249 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2128 = torch.aten.view %2112, %2127 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2128, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2250 = torch.constant.int 4
%int8_2251 = torch.constant.int 8
%int128_2252 = torch.constant.int 128
%2129 = torch.prim.ListConstruct %int4_2250, %294, %int8_2251, %int128_2252 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2130 = torch.aten.view %2119, %2129 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2130, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_2253 = torch.constant.int 4
%int8_2254 = torch.constant.int 8
%int128_2255 = torch.constant.int 128
%2131 = torch.prim.ListConstruct %int4_2253, %294, %int8_2254, %int128_2255 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2132 = torch.aten.view %2126, %2131 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2132, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_2256 = torch.constant.int 131072
%none_2257 = torch.constant.none
%none_2258 = torch.constant.none
%cpu_2259 = torch.constant.device "cpu"
%false_2260 = torch.constant.bool false
%2133 = torch.aten.arange %int131072_2256, %none_2257, %none_2258, %cpu_2259, %false_2260 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2261 = torch.constant.int 0
%int128_2262 = torch.constant.int 128
%int2_2263 = torch.constant.int 2
%none_2264 = torch.constant.none
%none_2265 = torch.constant.none
%cpu_2266 = torch.constant.device "cpu"
%false_2267 = torch.constant.bool false
%2134 = torch.aten.arange.start_step %int0_2261, %int128_2262, %int2_2263, %none_2264, %none_2265, %cpu_2266, %false_2267 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2268 = torch.constant.int 0
%int0_2269 = torch.constant.int 0
%int64_2270 = torch.constant.int 64
%int1_2271 = torch.constant.int 1
%2135 = torch.aten.slice.Tensor %2134, %int0_2268, %int0_2269, %int64_2270, %int1_2271 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2272 = torch.constant.int 6
%2136 = torch.prims.convert_element_type %2135, %int6_2272 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2273 = torch.constant.int 128
%2137 = torch.aten.div.Scalar %2136, %int128_2273 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2274 = torch.constant.float 5.000000e+05
%2138 = torch.aten.pow.Scalar %float5.000000e05_2274, %2137 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2139 = torch.aten.reciprocal %2138 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2275 = torch.constant.float 1.000000e+00
%2140 = torch.aten.mul.Scalar %2139, %float1.000000e00_2275 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2276 = torch.constant.int 131072
%int1_2277 = torch.constant.int 1
%2141 = torch.prim.ListConstruct %int131072_2276, %int1_2277 : (!torch.int, !torch.int) -> !torch.list<int>
%2142 = torch.aten.view %2133, %2141 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2143 = torch.aten.mul.Tensor %2142, %2140 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2144 = torch.aten.cos %2143 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2145 = torch.aten.sin %2143 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2146 = torch.aten.complex %2144, %2145 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2278 = torch.constant.int 1
%2147 = torch.aten.size.int %2112, %int1_2278 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_2279 = torch.constant.int 0
%2148 = torch.aten.add.int %int0_2279, %2147 : !torch.int, !torch.int -> !torch.int
%int0_2280 = torch.constant.int 0
%int0_2281 = torch.constant.int 0
%int1_2282 = torch.constant.int 1
%2149 = torch.aten.slice.Tensor %2146, %int0_2280, %int0_2281, %2148, %int1_2282 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2149, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2283 = torch.constant.int 1
%int0_2284 = torch.constant.int 0
%int9223372036854775807_2285 = torch.constant.int 9223372036854775807
%int1_2286 = torch.constant.int 1
%2150 = torch.aten.slice.Tensor %2149, %int1_2283, %int0_2284, %int9223372036854775807_2285, %int1_2286 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2150, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2287 = torch.constant.int 0
%2151 = torch.aten.unsqueeze %2150, %int0_2287 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2151, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2288 = torch.constant.int 2
%2152 = torch.aten.unsqueeze %2151, %int2_2288 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2152, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2289 = torch.constant.int 3
%int0_2290 = torch.constant.int 0
%int9223372036854775807_2291 = torch.constant.int 9223372036854775807
%int1_2292 = torch.constant.int 1
%2153 = torch.aten.slice.Tensor %2152, %int3_2289, %int0_2290, %int9223372036854775807_2291, %int1_2292 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2153, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2154 = torch_c.to_builtin_tensor %2128 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_2293 = arith.constant 1 : index
%dim_2294 = tensor.dim %2154, %c1_2293 : tensor<4x?x32x128xf16>
%2155 = flow.tensor.bitcast %2154 : tensor<4x?x32x128xf16>{%dim_2294} -> tensor<4x?x32x64xcomplex<f16>>{%dim_2294}
%2156 = torch_c.from_builtin_tensor %2155 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %2156, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%2157 = torch.aten.mul.Tensor %2156, %2153 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %2157, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%2158 = torch_c.to_builtin_tensor %2157 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_2295 = arith.constant 1 : index
%dim_2296 = tensor.dim %2158, %c1_2295 : tensor<4x?x32x64xcomplex<f32>>
%2159 = flow.tensor.bitcast %2158 : tensor<4x?x32x64xcomplex<f32>>{%dim_2296} -> tensor<4x?x32x128xf32>{%dim_2296}
%2160 = torch_c.from_builtin_tensor %2159 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %2160, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_2297 = torch.constant.int 5
%2161 = torch.prims.convert_element_type %2160, %int5_2297 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2161, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_2298 = torch.constant.int 131072
%none_2299 = torch.constant.none
%none_2300 = torch.constant.none
%cpu_2301 = torch.constant.device "cpu"
%false_2302 = torch.constant.bool false
%2162 = torch.aten.arange %int131072_2298, %none_2299, %none_2300, %cpu_2301, %false_2302 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2303 = torch.constant.int 0
%int128_2304 = torch.constant.int 128
%int2_2305 = torch.constant.int 2
%none_2306 = torch.constant.none
%none_2307 = torch.constant.none
%cpu_2308 = torch.constant.device "cpu"
%false_2309 = torch.constant.bool false
%2163 = torch.aten.arange.start_step %int0_2303, %int128_2304, %int2_2305, %none_2306, %none_2307, %cpu_2308, %false_2309 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2310 = torch.constant.int 0
%int0_2311 = torch.constant.int 0
%int64_2312 = torch.constant.int 64
%int1_2313 = torch.constant.int 1
%2164 = torch.aten.slice.Tensor %2163, %int0_2310, %int0_2311, %int64_2312, %int1_2313 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2314 = torch.constant.int 6
%2165 = torch.prims.convert_element_type %2164, %int6_2314 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2315 = torch.constant.int 128
%2166 = torch.aten.div.Scalar %2165, %int128_2315 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2316 = torch.constant.float 5.000000e+05
%2167 = torch.aten.pow.Scalar %float5.000000e05_2316, %2166 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2168 = torch.aten.reciprocal %2167 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2317 = torch.constant.float 1.000000e+00
%2169 = torch.aten.mul.Scalar %2168, %float1.000000e00_2317 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2318 = torch.constant.int 131072
%int1_2319 = torch.constant.int 1
%2170 = torch.prim.ListConstruct %int131072_2318, %int1_2319 : (!torch.int, !torch.int) -> !torch.list<int>
%2171 = torch.aten.view %2162, %2170 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2172 = torch.aten.mul.Tensor %2171, %2169 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2173 = torch.aten.cos %2172 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2174 = torch.aten.sin %2172 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2175 = torch.aten.complex %2173, %2174 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2320 = torch.constant.int 1
%2176 = torch.aten.size.int %2119, %int1_2320 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_2321 = torch.constant.int 0
%2177 = torch.aten.add.int %int0_2321, %2176 : !torch.int, !torch.int -> !torch.int
%int0_2322 = torch.constant.int 0
%int0_2323 = torch.constant.int 0
%int1_2324 = torch.constant.int 1
%2178 = torch.aten.slice.Tensor %2175, %int0_2322, %int0_2323, %2177, %int1_2324 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2178, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2325 = torch.constant.int 1
%int0_2326 = torch.constant.int 0
%int9223372036854775807_2327 = torch.constant.int 9223372036854775807
%int1_2328 = torch.constant.int 1
%2179 = torch.aten.slice.Tensor %2178, %int1_2325, %int0_2326, %int9223372036854775807_2327, %int1_2328 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2179, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2329 = torch.constant.int 0
%2180 = torch.aten.unsqueeze %2179, %int0_2329 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2180, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2330 = torch.constant.int 2
%2181 = torch.aten.unsqueeze %2180, %int2_2330 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2181, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2331 = torch.constant.int 3
%int0_2332 = torch.constant.int 0
%int9223372036854775807_2333 = torch.constant.int 9223372036854775807
%int1_2334 = torch.constant.int 1
%2182 = torch.aten.slice.Tensor %2181, %int3_2331, %int0_2332, %int9223372036854775807_2333, %int1_2334 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2182, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2183 = torch_c.to_builtin_tensor %2130 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_2335 = arith.constant 1 : index
%dim_2336 = tensor.dim %2183, %c1_2335 : tensor<4x?x8x128xf16>
%2184 = flow.tensor.bitcast %2183 : tensor<4x?x8x128xf16>{%dim_2336} -> tensor<4x?x8x64xcomplex<f16>>{%dim_2336}
%2185 = torch_c.from_builtin_tensor %2184 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %2185, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%2186 = torch.aten.mul.Tensor %2185, %2182 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %2186, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%2187 = torch_c.to_builtin_tensor %2186 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_2337 = arith.constant 1 : index
%dim_2338 = tensor.dim %2187, %c1_2337 : tensor<4x?x8x64xcomplex<f32>>
%2188 = flow.tensor.bitcast %2187 : tensor<4x?x8x64xcomplex<f32>>{%dim_2338} -> tensor<4x?x8x128xf32>{%dim_2338}
%2189 = torch_c.from_builtin_tensor %2188 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %2189, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_2339 = torch.constant.int 5
%2190 = torch.prims.convert_element_type %2189, %int5_2339 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2190, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_2340 = torch.constant.int 64
%2191 = torch.aten.mul.Scalar %arg2, %int64_2340 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2191, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int18 = torch.constant.int 18
%int1_2341 = torch.constant.int 1
%2192 = torch.aten.add.Scalar %2191, %int18, %int1_2341 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2192, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2342 = torch.constant.int 4
%int32_2343 = torch.constant.int 32
%int8_2344 = torch.constant.int 8
%int128_2345 = torch.constant.int 128
%2193 = torch.prim.ListConstruct %int4_2342, %425, %int32_2343, %int8_2344, %int128_2345 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2194 = torch.aten.view %2190, %2193 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2194, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2346 = torch.constant.int 4
%2195 = torch.aten.mul.int %int4_2346, %425 : !torch.int, !torch.int -> !torch.int
%int32_2347 = torch.constant.int 32
%int8_2348 = torch.constant.int 8
%int128_2349 = torch.constant.int 128
%2196 = torch.prim.ListConstruct %2195, %int32_2347, %int8_2348, %int128_2349 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2197 = torch.aten.view %2194, %2196 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2197, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2350 = torch.constant.int 4
%2198 = torch.aten.mul.int %int4_2350, %425 : !torch.int, !torch.int -> !torch.int
%2199 = torch.prim.ListConstruct %2198 : (!torch.int) -> !torch.list<int>
%2200 = torch.aten.view %2192, %2199 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2200, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_2351 = torch.constant.int 32
%int2_2352 = torch.constant.int 2
%int32_2353 = torch.constant.int 32
%int8_2354 = torch.constant.int 8
%int128_2355 = torch.constant.int 128
%2201 = torch.prim.ListConstruct %416, %int32_2351, %int2_2352, %int32_2353, %int8_2354, %int128_2355 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2202 = torch.aten.view %2034, %2201 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2202, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2356 = torch.constant.int 32
%2203 = torch.aten.mul.int %416, %int32_2356 : !torch.int, !torch.int -> !torch.int
%int2_2357 = torch.constant.int 2
%2204 = torch.aten.mul.int %2203, %int2_2357 : !torch.int, !torch.int -> !torch.int
%int32_2358 = torch.constant.int 32
%int8_2359 = torch.constant.int 8
%int128_2360 = torch.constant.int 128
%2205 = torch.prim.ListConstruct %2204, %int32_2358, %int8_2359, %int128_2360 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2206 = torch.aten.view %2202, %2205 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2206, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2207 = torch.prim.ListConstruct %2200 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2361 = torch.constant.bool false
%2208 = torch.aten.index_put %2206, %2207, %2197, %false_2361 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2208, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2362 = torch.constant.int 32
%int2_2363 = torch.constant.int 2
%int32_2364 = torch.constant.int 32
%int8_2365 = torch.constant.int 8
%int128_2366 = torch.constant.int 128
%2209 = torch.prim.ListConstruct %416, %int32_2362, %int2_2363, %int32_2364, %int8_2365, %int128_2366 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2210 = torch.aten.view %2208, %2209 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2210, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2367 = torch.constant.int 2097152
%2211 = torch.prim.ListConstruct %416, %int2097152_2367 : (!torch.int, !torch.int) -> !torch.list<int>
%2212 = torch.aten.view %2210, %2211 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2212, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_2368 = torch.constant.int 32
%int2_2369 = torch.constant.int 2
%int32_2370 = torch.constant.int 32
%int8_2371 = torch.constant.int 8
%int128_2372 = torch.constant.int 128
%2213 = torch.prim.ListConstruct %416, %int32_2368, %int2_2369, %int32_2370, %int8_2371, %int128_2372 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2214 = torch.aten.view %2212, %2213 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2214, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2373 = torch.constant.int 32
%int8_2374 = torch.constant.int 8
%int128_2375 = torch.constant.int 128
%2215 = torch.prim.ListConstruct %2204, %int32_2373, %int8_2374, %int128_2375 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2216 = torch.aten.view %2214, %2215 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2216, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2376 = torch.constant.int 4
%int32_2377 = torch.constant.int 32
%int8_2378 = torch.constant.int 8
%int128_2379 = torch.constant.int 128
%2217 = torch.prim.ListConstruct %int4_2376, %425, %int32_2377, %int8_2378, %int128_2379 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2218 = torch.aten.view %2132, %2217 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2218, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2380 = torch.constant.int 4
%2219 = torch.aten.mul.int %int4_2380, %425 : !torch.int, !torch.int -> !torch.int
%int32_2381 = torch.constant.int 32
%int8_2382 = torch.constant.int 8
%int128_2383 = torch.constant.int 128
%2220 = torch.prim.ListConstruct %2219, %int32_2381, %int8_2382, %int128_2383 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2221 = torch.aten.view %2218, %2220 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2221, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_2384 = torch.constant.int 1
%int1_2385 = torch.constant.int 1
%2222 = torch.aten.add.Scalar %2192, %int1_2384, %int1_2385 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2222, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2386 = torch.constant.int 4
%2223 = torch.aten.mul.int %int4_2386, %425 : !torch.int, !torch.int -> !torch.int
%2224 = torch.prim.ListConstruct %2223 : (!torch.int) -> !torch.list<int>
%2225 = torch.aten.view %2222, %2224 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2225, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%2226 = torch.prim.ListConstruct %2225 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2387 = torch.constant.bool false
%2227 = torch.aten.index_put %2216, %2226, %2221, %false_2387 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2227, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2388 = torch.constant.int 32
%int2_2389 = torch.constant.int 2
%int32_2390 = torch.constant.int 32
%int8_2391 = torch.constant.int 8
%int128_2392 = torch.constant.int 128
%2228 = torch.prim.ListConstruct %416, %int32_2388, %int2_2389, %int32_2390, %int8_2391, %int128_2392 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2229 = torch.aten.view %2227, %2228 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2229, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2393 = torch.constant.int 2097152
%2230 = torch.prim.ListConstruct %416, %int2097152_2393 : (!torch.int, !torch.int) -> !torch.list<int>
%2231 = torch.aten.view %2229, %2230 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2231, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_2394 = torch.constant.int -2
%2232 = torch.aten.unsqueeze %2190, %int-2_2394 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2232, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_2395 = torch.constant.int 4
%int8_2396 = torch.constant.int 8
%int4_2397 = torch.constant.int 4
%int128_2398 = torch.constant.int 128
%2233 = torch.prim.ListConstruct %int4_2395, %2176, %int8_2396, %int4_2397, %int128_2398 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2399 = torch.constant.bool false
%2234 = torch.aten.expand %2232, %2233, %false_2399 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2234, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2400 = torch.constant.int 0
%2235 = torch.aten.clone %2234, %int0_2400 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2235, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2401 = torch.constant.int 4
%int32_2402 = torch.constant.int 32
%int128_2403 = torch.constant.int 128
%2236 = torch.prim.ListConstruct %int4_2401, %2176, %int32_2402, %int128_2403 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2237 = torch.aten._unsafe_view %2235, %2236 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2237, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_2404 = torch.constant.int -2
%2238 = torch.aten.unsqueeze %2132, %int-2_2404 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2238, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_2405 = torch.constant.int 1
%2239 = torch.aten.size.int %2126, %int1_2405 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_2406 = torch.constant.int 4
%int8_2407 = torch.constant.int 8
%int4_2408 = torch.constant.int 4
%int128_2409 = torch.constant.int 128
%2240 = torch.prim.ListConstruct %int4_2406, %2239, %int8_2407, %int4_2408, %int128_2409 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2410 = torch.constant.bool false
%2241 = torch.aten.expand %2238, %2240, %false_2410 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2241, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2411 = torch.constant.int 0
%2242 = torch.aten.clone %2241, %int0_2411 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2242, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2412 = torch.constant.int 4
%int32_2413 = torch.constant.int 32
%int128_2414 = torch.constant.int 128
%2243 = torch.prim.ListConstruct %int4_2412, %2239, %int32_2413, %int128_2414 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2244 = torch.aten._unsafe_view %2242, %2243 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2244, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_2415 = torch.constant.int 1
%int2_2416 = torch.constant.int 2
%2245 = torch.aten.transpose.int %2161, %int1_2415, %int2_2416 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2245, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2417 = torch.constant.int 1
%int2_2418 = torch.constant.int 2
%2246 = torch.aten.transpose.int %2237, %int1_2417, %int2_2418 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2246, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2419 = torch.constant.int 1
%int2_2420 = torch.constant.int 2
%2247 = torch.aten.transpose.int %2244, %int1_2419, %int2_2420 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2247, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_2421 = torch.constant.float 0.000000e+00
%false_2422 = torch.constant.bool false
%none_2423 = torch.constant.none
%2248:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%2245, %2246, %2247, %float0.000000e00_2421, %false_2422, %320, %none_2423) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %2248#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2424 = torch.constant.int 1
%int2_2425 = torch.constant.int 2
%2249 = torch.aten.transpose.int %2248#0, %int1_2424, %int2_2425 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2249, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2426 = torch.constant.int 4
%int4096_2427 = torch.constant.int 4096
%2250 = torch.prim.ListConstruct %int4_2426, %2147, %int4096_2427 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2251 = torch.aten.view %2249, %2250 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2251, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2428 = torch.constant.int -2
%int-1_2429 = torch.constant.int -1
%2252 = torch.aten.transpose.int %86, %int-2_2428, %int-1_2429 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2430 = torch.constant.int 4
%2253 = torch.aten.mul.int %int4_2430, %2147 : !torch.int, !torch.int -> !torch.int
%int4096_2431 = torch.constant.int 4096
%2254 = torch.prim.ListConstruct %2253, %int4096_2431 : (!torch.int, !torch.int) -> !torch.list<int>
%2255 = torch.aten.view %2251, %2254 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2255, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2256 = torch.aten.mm %2255, %2252 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2256, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2432 = torch.constant.int 4
%int4096_2433 = torch.constant.int 4096
%2257 = torch.prim.ListConstruct %int4_2432, %2147, %int4096_2433 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2258 = torch.aten.view %2256, %2257 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2258, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2434 = torch.constant.int 1
%2259 = torch.aten.add.Tensor %2096, %2258, %int1_2434 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2259, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2435 = torch.constant.int 6
%2260 = torch.prims.convert_element_type %2259, %int6_2435 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2260, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2436 = torch.constant.int 2
%2261 = torch.aten.pow.Tensor_Scalar %2260, %int2_2436 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2261, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2437 = torch.constant.int -1
%2262 = torch.prim.ListConstruct %int-1_2437 : (!torch.int) -> !torch.list<int>
%true_2438 = torch.constant.bool true
%none_2439 = torch.constant.none
%2263 = torch.aten.mean.dim %2261, %2262, %true_2438, %none_2439 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2263, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2440 = torch.constant.float 9.9999997473787516E-6
%int1_2441 = torch.constant.int 1
%2264 = torch.aten.add.Scalar %2263, %float9.999990e-06_2440, %int1_2441 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2264, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2265 = torch.aten.rsqrt %2264 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2265, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2266 = torch.aten.mul.Tensor %2260, %2265 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2266, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2267 = torch.aten.mul.Tensor %87, %2266 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2267, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2442 = torch.constant.int 5
%2268 = torch.prims.convert_element_type %2267, %int5_2442 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2268, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2443 = torch.constant.int -2
%int-1_2444 = torch.constant.int -1
%2269 = torch.aten.transpose.int %88, %int-2_2443, %int-1_2444 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2445 = torch.constant.int 4
%2270 = torch.aten.mul.int %int4_2445, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2446 = torch.constant.int 4096
%2271 = torch.prim.ListConstruct %2270, %int4096_2446 : (!torch.int, !torch.int) -> !torch.list<int>
%2272 = torch.aten.view %2268, %2271 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2272, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2273 = torch.aten.mm %2272, %2269 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2273, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2447 = torch.constant.int 4
%int14336_2448 = torch.constant.int 14336
%2274 = torch.prim.ListConstruct %int4_2447, %294, %int14336_2448 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2275 = torch.aten.view %2273, %2274 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2275, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2276 = torch.aten.silu %2275 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2276, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2449 = torch.constant.int -2
%int-1_2450 = torch.constant.int -1
%2277 = torch.aten.transpose.int %89, %int-2_2449, %int-1_2450 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2451 = torch.constant.int 4
%2278 = torch.aten.mul.int %int4_2451, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2452 = torch.constant.int 4096
%2279 = torch.prim.ListConstruct %2278, %int4096_2452 : (!torch.int, !torch.int) -> !torch.list<int>
%2280 = torch.aten.view %2268, %2279 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2280, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2281 = torch.aten.mm %2280, %2277 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2281, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2453 = torch.constant.int 4
%int14336_2454 = torch.constant.int 14336
%2282 = torch.prim.ListConstruct %int4_2453, %294, %int14336_2454 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2283 = torch.aten.view %2281, %2282 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2283, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2284 = torch.aten.mul.Tensor %2276, %2283 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2284, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2455 = torch.constant.int -2
%int-1_2456 = torch.constant.int -1
%2285 = torch.aten.transpose.int %90, %int-2_2455, %int-1_2456 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_2457 = torch.constant.int 1
%2286 = torch.aten.size.int %2275, %int1_2457 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_2458 = torch.constant.int 4
%2287 = torch.aten.mul.int %int4_2458, %2286 : !torch.int, !torch.int -> !torch.int
%int14336_2459 = torch.constant.int 14336
%2288 = torch.prim.ListConstruct %2287, %int14336_2459 : (!torch.int, !torch.int) -> !torch.list<int>
%2289 = torch.aten.view %2284, %2288 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2289, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%2290 = torch.aten.mm %2289, %2285 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2290, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2460 = torch.constant.int 4
%int4096_2461 = torch.constant.int 4096
%2291 = torch.prim.ListConstruct %int4_2460, %2286, %int4096_2461 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2292 = torch.aten.view %2290, %2291 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2292, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2462 = torch.constant.int 1
%2293 = torch.aten.add.Tensor %2259, %2292, %int1_2462 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2293, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2463 = torch.constant.int 6
%2294 = torch.prims.convert_element_type %2293, %int6_2463 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2294, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2464 = torch.constant.int 2
%2295 = torch.aten.pow.Tensor_Scalar %2294, %int2_2464 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2295, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2465 = torch.constant.int -1
%2296 = torch.prim.ListConstruct %int-1_2465 : (!torch.int) -> !torch.list<int>
%true_2466 = torch.constant.bool true
%none_2467 = torch.constant.none
%2297 = torch.aten.mean.dim %2295, %2296, %true_2466, %none_2467 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2297, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2468 = torch.constant.float 9.9999997473787516E-6
%int1_2469 = torch.constant.int 1
%2298 = torch.aten.add.Scalar %2297, %float9.999990e-06_2468, %int1_2469 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2298, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2299 = torch.aten.rsqrt %2298 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2299, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2300 = torch.aten.mul.Tensor %2294, %2299 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2300, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2301 = torch.aten.mul.Tensor %91, %2300 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2301, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2470 = torch.constant.int 5
%2302 = torch.prims.convert_element_type %2301, %int5_2470 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2302, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2471 = torch.constant.int -2
%int-1_2472 = torch.constant.int -1
%2303 = torch.aten.transpose.int %92, %int-2_2471, %int-1_2472 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2473 = torch.constant.int 4
%2304 = torch.aten.mul.int %int4_2473, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2474 = torch.constant.int 4096
%2305 = torch.prim.ListConstruct %2304, %int4096_2474 : (!torch.int, !torch.int) -> !torch.list<int>
%2306 = torch.aten.view %2302, %2305 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2306, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2307 = torch.aten.mm %2306, %2303 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2307, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2475 = torch.constant.int 4
%int4096_2476 = torch.constant.int 4096
%2308 = torch.prim.ListConstruct %int4_2475, %294, %int4096_2476 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2309 = torch.aten.view %2307, %2308 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2309, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2477 = torch.constant.int -2
%int-1_2478 = torch.constant.int -1
%2310 = torch.aten.transpose.int %93, %int-2_2477, %int-1_2478 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2479 = torch.constant.int 4
%2311 = torch.aten.mul.int %int4_2479, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2480 = torch.constant.int 4096
%2312 = torch.prim.ListConstruct %2311, %int4096_2480 : (!torch.int, !torch.int) -> !torch.list<int>
%2313 = torch.aten.view %2302, %2312 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2313, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2314 = torch.aten.mm %2313, %2310 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2314, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2481 = torch.constant.int 4
%int1024_2482 = torch.constant.int 1024
%2315 = torch.prim.ListConstruct %int4_2481, %294, %int1024_2482 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2316 = torch.aten.view %2314, %2315 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2316, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_2483 = torch.constant.int -2
%int-1_2484 = torch.constant.int -1
%2317 = torch.aten.transpose.int %94, %int-2_2483, %int-1_2484 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2485 = torch.constant.int 4
%2318 = torch.aten.mul.int %int4_2485, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2486 = torch.constant.int 4096
%2319 = torch.prim.ListConstruct %2318, %int4096_2486 : (!torch.int, !torch.int) -> !torch.list<int>
%2320 = torch.aten.view %2302, %2319 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2320, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2321 = torch.aten.mm %2320, %2317 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2321, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2487 = torch.constant.int 4
%int1024_2488 = torch.constant.int 1024
%2322 = torch.prim.ListConstruct %int4_2487, %294, %int1024_2488 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2323 = torch.aten.view %2321, %2322 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2323, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_2489 = torch.constant.int 4
%int32_2490 = torch.constant.int 32
%int128_2491 = torch.constant.int 128
%2324 = torch.prim.ListConstruct %int4_2489, %294, %int32_2490, %int128_2491 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2325 = torch.aten.view %2309, %2324 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2325, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2492 = torch.constant.int 4
%int8_2493 = torch.constant.int 8
%int128_2494 = torch.constant.int 128
%2326 = torch.prim.ListConstruct %int4_2492, %294, %int8_2493, %int128_2494 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2327 = torch.aten.view %2316, %2326 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2327, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_2495 = torch.constant.int 4
%int8_2496 = torch.constant.int 8
%int128_2497 = torch.constant.int 128
%2328 = torch.prim.ListConstruct %int4_2495, %294, %int8_2496, %int128_2497 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2329 = torch.aten.view %2323, %2328 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2329, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_2498 = torch.constant.int 131072
%none_2499 = torch.constant.none
%none_2500 = torch.constant.none
%cpu_2501 = torch.constant.device "cpu"
%false_2502 = torch.constant.bool false
%2330 = torch.aten.arange %int131072_2498, %none_2499, %none_2500, %cpu_2501, %false_2502 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2503 = torch.constant.int 0
%int128_2504 = torch.constant.int 128
%int2_2505 = torch.constant.int 2
%none_2506 = torch.constant.none
%none_2507 = torch.constant.none
%cpu_2508 = torch.constant.device "cpu"
%false_2509 = torch.constant.bool false
%2331 = torch.aten.arange.start_step %int0_2503, %int128_2504, %int2_2505, %none_2506, %none_2507, %cpu_2508, %false_2509 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2510 = torch.constant.int 0
%int0_2511 = torch.constant.int 0
%int64_2512 = torch.constant.int 64
%int1_2513 = torch.constant.int 1
%2332 = torch.aten.slice.Tensor %2331, %int0_2510, %int0_2511, %int64_2512, %int1_2513 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2514 = torch.constant.int 6
%2333 = torch.prims.convert_element_type %2332, %int6_2514 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2515 = torch.constant.int 128
%2334 = torch.aten.div.Scalar %2333, %int128_2515 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2516 = torch.constant.float 5.000000e+05
%2335 = torch.aten.pow.Scalar %float5.000000e05_2516, %2334 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2336 = torch.aten.reciprocal %2335 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2517 = torch.constant.float 1.000000e+00
%2337 = torch.aten.mul.Scalar %2336, %float1.000000e00_2517 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2518 = torch.constant.int 131072
%int1_2519 = torch.constant.int 1
%2338 = torch.prim.ListConstruct %int131072_2518, %int1_2519 : (!torch.int, !torch.int) -> !torch.list<int>
%2339 = torch.aten.view %2330, %2338 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2340 = torch.aten.mul.Tensor %2339, %2337 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2341 = torch.aten.cos %2340 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2342 = torch.aten.sin %2340 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2343 = torch.aten.complex %2341, %2342 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2520 = torch.constant.int 1
%2344 = torch.aten.size.int %2309, %int1_2520 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_2521 = torch.constant.int 0
%2345 = torch.aten.add.int %int0_2521, %2344 : !torch.int, !torch.int -> !torch.int
%int0_2522 = torch.constant.int 0
%int0_2523 = torch.constant.int 0
%int1_2524 = torch.constant.int 1
%2346 = torch.aten.slice.Tensor %2343, %int0_2522, %int0_2523, %2345, %int1_2524 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2346, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2525 = torch.constant.int 1
%int0_2526 = torch.constant.int 0
%int9223372036854775807_2527 = torch.constant.int 9223372036854775807
%int1_2528 = torch.constant.int 1
%2347 = torch.aten.slice.Tensor %2346, %int1_2525, %int0_2526, %int9223372036854775807_2527, %int1_2528 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2347, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2529 = torch.constant.int 0
%2348 = torch.aten.unsqueeze %2347, %int0_2529 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2348, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2530 = torch.constant.int 2
%2349 = torch.aten.unsqueeze %2348, %int2_2530 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2349, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2531 = torch.constant.int 3
%int0_2532 = torch.constant.int 0
%int9223372036854775807_2533 = torch.constant.int 9223372036854775807
%int1_2534 = torch.constant.int 1
%2350 = torch.aten.slice.Tensor %2349, %int3_2531, %int0_2532, %int9223372036854775807_2533, %int1_2534 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2350, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2351 = torch_c.to_builtin_tensor %2325 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_2535 = arith.constant 1 : index
%dim_2536 = tensor.dim %2351, %c1_2535 : tensor<4x?x32x128xf16>
%2352 = flow.tensor.bitcast %2351 : tensor<4x?x32x128xf16>{%dim_2536} -> tensor<4x?x32x64xcomplex<f16>>{%dim_2536}
%2353 = torch_c.from_builtin_tensor %2352 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %2353, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%2354 = torch.aten.mul.Tensor %2353, %2350 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %2354, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%2355 = torch_c.to_builtin_tensor %2354 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_2537 = arith.constant 1 : index
%dim_2538 = tensor.dim %2355, %c1_2537 : tensor<4x?x32x64xcomplex<f32>>
%2356 = flow.tensor.bitcast %2355 : tensor<4x?x32x64xcomplex<f32>>{%dim_2538} -> tensor<4x?x32x128xf32>{%dim_2538}
%2357 = torch_c.from_builtin_tensor %2356 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %2357, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_2539 = torch.constant.int 5
%2358 = torch.prims.convert_element_type %2357, %int5_2539 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2358, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_2540 = torch.constant.int 131072
%none_2541 = torch.constant.none
%none_2542 = torch.constant.none
%cpu_2543 = torch.constant.device "cpu"
%false_2544 = torch.constant.bool false
%2359 = torch.aten.arange %int131072_2540, %none_2541, %none_2542, %cpu_2543, %false_2544 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2545 = torch.constant.int 0
%int128_2546 = torch.constant.int 128
%int2_2547 = torch.constant.int 2
%none_2548 = torch.constant.none
%none_2549 = torch.constant.none
%cpu_2550 = torch.constant.device "cpu"
%false_2551 = torch.constant.bool false
%2360 = torch.aten.arange.start_step %int0_2545, %int128_2546, %int2_2547, %none_2548, %none_2549, %cpu_2550, %false_2551 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2552 = torch.constant.int 0
%int0_2553 = torch.constant.int 0
%int64_2554 = torch.constant.int 64
%int1_2555 = torch.constant.int 1
%2361 = torch.aten.slice.Tensor %2360, %int0_2552, %int0_2553, %int64_2554, %int1_2555 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2556 = torch.constant.int 6
%2362 = torch.prims.convert_element_type %2361, %int6_2556 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2557 = torch.constant.int 128
%2363 = torch.aten.div.Scalar %2362, %int128_2557 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2558 = torch.constant.float 5.000000e+05
%2364 = torch.aten.pow.Scalar %float5.000000e05_2558, %2363 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2365 = torch.aten.reciprocal %2364 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2559 = torch.constant.float 1.000000e+00
%2366 = torch.aten.mul.Scalar %2365, %float1.000000e00_2559 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2560 = torch.constant.int 131072
%int1_2561 = torch.constant.int 1
%2367 = torch.prim.ListConstruct %int131072_2560, %int1_2561 : (!torch.int, !torch.int) -> !torch.list<int>
%2368 = torch.aten.view %2359, %2367 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2369 = torch.aten.mul.Tensor %2368, %2366 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2370 = torch.aten.cos %2369 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2371 = torch.aten.sin %2369 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2372 = torch.aten.complex %2370, %2371 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2562 = torch.constant.int 1
%2373 = torch.aten.size.int %2316, %int1_2562 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_2563 = torch.constant.int 0
%2374 = torch.aten.add.int %int0_2563, %2373 : !torch.int, !torch.int -> !torch.int
%int0_2564 = torch.constant.int 0
%int0_2565 = torch.constant.int 0
%int1_2566 = torch.constant.int 1
%2375 = torch.aten.slice.Tensor %2372, %int0_2564, %int0_2565, %2374, %int1_2566 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2375, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2567 = torch.constant.int 1
%int0_2568 = torch.constant.int 0
%int9223372036854775807_2569 = torch.constant.int 9223372036854775807
%int1_2570 = torch.constant.int 1
%2376 = torch.aten.slice.Tensor %2375, %int1_2567, %int0_2568, %int9223372036854775807_2569, %int1_2570 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2376, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2571 = torch.constant.int 0
%2377 = torch.aten.unsqueeze %2376, %int0_2571 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2377, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2572 = torch.constant.int 2
%2378 = torch.aten.unsqueeze %2377, %int2_2572 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2378, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2573 = torch.constant.int 3
%int0_2574 = torch.constant.int 0
%int9223372036854775807_2575 = torch.constant.int 9223372036854775807
%int1_2576 = torch.constant.int 1
%2379 = torch.aten.slice.Tensor %2378, %int3_2573, %int0_2574, %int9223372036854775807_2575, %int1_2576 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2379, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2380 = torch_c.to_builtin_tensor %2327 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_2577 = arith.constant 1 : index
%dim_2578 = tensor.dim %2380, %c1_2577 : tensor<4x?x8x128xf16>
%2381 = flow.tensor.bitcast %2380 : tensor<4x?x8x128xf16>{%dim_2578} -> tensor<4x?x8x64xcomplex<f16>>{%dim_2578}
%2382 = torch_c.from_builtin_tensor %2381 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %2382, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%2383 = torch.aten.mul.Tensor %2382, %2379 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %2383, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%2384 = torch_c.to_builtin_tensor %2383 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_2579 = arith.constant 1 : index
%dim_2580 = tensor.dim %2384, %c1_2579 : tensor<4x?x8x64xcomplex<f32>>
%2385 = flow.tensor.bitcast %2384 : tensor<4x?x8x64xcomplex<f32>>{%dim_2580} -> tensor<4x?x8x128xf32>{%dim_2580}
%2386 = torch_c.from_builtin_tensor %2385 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %2386, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_2581 = torch.constant.int 5
%2387 = torch.prims.convert_element_type %2386, %int5_2581 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2387, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_2582 = torch.constant.int 64
%2388 = torch.aten.mul.Scalar %arg2, %int64_2582 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2388, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int20 = torch.constant.int 20
%int1_2583 = torch.constant.int 1
%2389 = torch.aten.add.Scalar %2388, %int20, %int1_2583 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2389, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2584 = torch.constant.int 4
%int32_2585 = torch.constant.int 32
%int8_2586 = torch.constant.int 8
%int128_2587 = torch.constant.int 128
%2390 = torch.prim.ListConstruct %int4_2584, %425, %int32_2585, %int8_2586, %int128_2587 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2391 = torch.aten.view %2387, %2390 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2391, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2588 = torch.constant.int 4
%2392 = torch.aten.mul.int %int4_2588, %425 : !torch.int, !torch.int -> !torch.int
%int32_2589 = torch.constant.int 32
%int8_2590 = torch.constant.int 8
%int128_2591 = torch.constant.int 128
%2393 = torch.prim.ListConstruct %2392, %int32_2589, %int8_2590, %int128_2591 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2394 = torch.aten.view %2391, %2393 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2394, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2592 = torch.constant.int 4
%2395 = torch.aten.mul.int %int4_2592, %425 : !torch.int, !torch.int -> !torch.int
%2396 = torch.prim.ListConstruct %2395 : (!torch.int) -> !torch.list<int>
%2397 = torch.aten.view %2389, %2396 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2397, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_2593 = torch.constant.int 32
%int2_2594 = torch.constant.int 2
%int32_2595 = torch.constant.int 32
%int8_2596 = torch.constant.int 8
%int128_2597 = torch.constant.int 128
%2398 = torch.prim.ListConstruct %416, %int32_2593, %int2_2594, %int32_2595, %int8_2596, %int128_2597 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2399 = torch.aten.view %2231, %2398 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2399, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2598 = torch.constant.int 32
%2400 = torch.aten.mul.int %416, %int32_2598 : !torch.int, !torch.int -> !torch.int
%int2_2599 = torch.constant.int 2
%2401 = torch.aten.mul.int %2400, %int2_2599 : !torch.int, !torch.int -> !torch.int
%int32_2600 = torch.constant.int 32
%int8_2601 = torch.constant.int 8
%int128_2602 = torch.constant.int 128
%2402 = torch.prim.ListConstruct %2401, %int32_2600, %int8_2601, %int128_2602 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2403 = torch.aten.view %2399, %2402 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2403, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2404 = torch.prim.ListConstruct %2397 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2603 = torch.constant.bool false
%2405 = torch.aten.index_put %2403, %2404, %2394, %false_2603 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2405, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2604 = torch.constant.int 32
%int2_2605 = torch.constant.int 2
%int32_2606 = torch.constant.int 32
%int8_2607 = torch.constant.int 8
%int128_2608 = torch.constant.int 128
%2406 = torch.prim.ListConstruct %416, %int32_2604, %int2_2605, %int32_2606, %int8_2607, %int128_2608 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2407 = torch.aten.view %2405, %2406 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2407, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2609 = torch.constant.int 2097152
%2408 = torch.prim.ListConstruct %416, %int2097152_2609 : (!torch.int, !torch.int) -> !torch.list<int>
%2409 = torch.aten.view %2407, %2408 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2409, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_2610 = torch.constant.int 32
%int2_2611 = torch.constant.int 2
%int32_2612 = torch.constant.int 32
%int8_2613 = torch.constant.int 8
%int128_2614 = torch.constant.int 128
%2410 = torch.prim.ListConstruct %416, %int32_2610, %int2_2611, %int32_2612, %int8_2613, %int128_2614 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2411 = torch.aten.view %2409, %2410 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2411, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2615 = torch.constant.int 32
%int8_2616 = torch.constant.int 8
%int128_2617 = torch.constant.int 128
%2412 = torch.prim.ListConstruct %2401, %int32_2615, %int8_2616, %int128_2617 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2413 = torch.aten.view %2411, %2412 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2413, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2618 = torch.constant.int 4
%int32_2619 = torch.constant.int 32
%int8_2620 = torch.constant.int 8
%int128_2621 = torch.constant.int 128
%2414 = torch.prim.ListConstruct %int4_2618, %425, %int32_2619, %int8_2620, %int128_2621 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2415 = torch.aten.view %2329, %2414 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2415, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2622 = torch.constant.int 4
%2416 = torch.aten.mul.int %int4_2622, %425 : !torch.int, !torch.int -> !torch.int
%int32_2623 = torch.constant.int 32
%int8_2624 = torch.constant.int 8
%int128_2625 = torch.constant.int 128
%2417 = torch.prim.ListConstruct %2416, %int32_2623, %int8_2624, %int128_2625 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2418 = torch.aten.view %2415, %2417 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2418, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_2626 = torch.constant.int 1
%int1_2627 = torch.constant.int 1
%2419 = torch.aten.add.Scalar %2389, %int1_2626, %int1_2627 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2419, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2628 = torch.constant.int 4
%2420 = torch.aten.mul.int %int4_2628, %425 : !torch.int, !torch.int -> !torch.int
%2421 = torch.prim.ListConstruct %2420 : (!torch.int) -> !torch.list<int>
%2422 = torch.aten.view %2419, %2421 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2422, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%2423 = torch.prim.ListConstruct %2422 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2629 = torch.constant.bool false
%2424 = torch.aten.index_put %2413, %2423, %2418, %false_2629 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2424, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2630 = torch.constant.int 32
%int2_2631 = torch.constant.int 2
%int32_2632 = torch.constant.int 32
%int8_2633 = torch.constant.int 8
%int128_2634 = torch.constant.int 128
%2425 = torch.prim.ListConstruct %416, %int32_2630, %int2_2631, %int32_2632, %int8_2633, %int128_2634 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2426 = torch.aten.view %2424, %2425 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2426, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2635 = torch.constant.int 2097152
%2427 = torch.prim.ListConstruct %416, %int2097152_2635 : (!torch.int, !torch.int) -> !torch.list<int>
%2428 = torch.aten.view %2426, %2427 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2428, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_2636 = torch.constant.int -2
%2429 = torch.aten.unsqueeze %2387, %int-2_2636 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2429, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_2637 = torch.constant.int 4
%int8_2638 = torch.constant.int 8
%int4_2639 = torch.constant.int 4
%int128_2640 = torch.constant.int 128
%2430 = torch.prim.ListConstruct %int4_2637, %2373, %int8_2638, %int4_2639, %int128_2640 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2641 = torch.constant.bool false
%2431 = torch.aten.expand %2429, %2430, %false_2641 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2431, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2642 = torch.constant.int 0
%2432 = torch.aten.clone %2431, %int0_2642 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2432, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2643 = torch.constant.int 4
%int32_2644 = torch.constant.int 32
%int128_2645 = torch.constant.int 128
%2433 = torch.prim.ListConstruct %int4_2643, %2373, %int32_2644, %int128_2645 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2434 = torch.aten._unsafe_view %2432, %2433 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2434, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_2646 = torch.constant.int -2
%2435 = torch.aten.unsqueeze %2329, %int-2_2646 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2435, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_2647 = torch.constant.int 1
%2436 = torch.aten.size.int %2323, %int1_2647 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_2648 = torch.constant.int 4
%int8_2649 = torch.constant.int 8
%int4_2650 = torch.constant.int 4
%int128_2651 = torch.constant.int 128
%2437 = torch.prim.ListConstruct %int4_2648, %2436, %int8_2649, %int4_2650, %int128_2651 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2652 = torch.constant.bool false
%2438 = torch.aten.expand %2435, %2437, %false_2652 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2438, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2653 = torch.constant.int 0
%2439 = torch.aten.clone %2438, %int0_2653 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2439, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2654 = torch.constant.int 4
%int32_2655 = torch.constant.int 32
%int128_2656 = torch.constant.int 128
%2440 = torch.prim.ListConstruct %int4_2654, %2436, %int32_2655, %int128_2656 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2441 = torch.aten._unsafe_view %2439, %2440 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2441, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_2657 = torch.constant.int 1
%int2_2658 = torch.constant.int 2
%2442 = torch.aten.transpose.int %2358, %int1_2657, %int2_2658 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2442, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2659 = torch.constant.int 1
%int2_2660 = torch.constant.int 2
%2443 = torch.aten.transpose.int %2434, %int1_2659, %int2_2660 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2443, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2661 = torch.constant.int 1
%int2_2662 = torch.constant.int 2
%2444 = torch.aten.transpose.int %2441, %int1_2661, %int2_2662 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2444, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_2663 = torch.constant.float 0.000000e+00
%false_2664 = torch.constant.bool false
%none_2665 = torch.constant.none
%2445:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%2442, %2443, %2444, %float0.000000e00_2663, %false_2664, %320, %none_2665) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %2445#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2666 = torch.constant.int 1
%int2_2667 = torch.constant.int 2
%2446 = torch.aten.transpose.int %2445#0, %int1_2666, %int2_2667 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2446, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2668 = torch.constant.int 4
%int4096_2669 = torch.constant.int 4096
%2447 = torch.prim.ListConstruct %int4_2668, %2344, %int4096_2669 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2448 = torch.aten.view %2446, %2447 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2448, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2670 = torch.constant.int -2
%int-1_2671 = torch.constant.int -1
%2449 = torch.aten.transpose.int %95, %int-2_2670, %int-1_2671 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2672 = torch.constant.int 4
%2450 = torch.aten.mul.int %int4_2672, %2344 : !torch.int, !torch.int -> !torch.int
%int4096_2673 = torch.constant.int 4096
%2451 = torch.prim.ListConstruct %2450, %int4096_2673 : (!torch.int, !torch.int) -> !torch.list<int>
%2452 = torch.aten.view %2448, %2451 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2452, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2453 = torch.aten.mm %2452, %2449 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2453, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2674 = torch.constant.int 4
%int4096_2675 = torch.constant.int 4096
%2454 = torch.prim.ListConstruct %int4_2674, %2344, %int4096_2675 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2455 = torch.aten.view %2453, %2454 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2455, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2676 = torch.constant.int 1
%2456 = torch.aten.add.Tensor %2293, %2455, %int1_2676 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2456, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2677 = torch.constant.int 6
%2457 = torch.prims.convert_element_type %2456, %int6_2677 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2457, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2678 = torch.constant.int 2
%2458 = torch.aten.pow.Tensor_Scalar %2457, %int2_2678 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2458, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2679 = torch.constant.int -1
%2459 = torch.prim.ListConstruct %int-1_2679 : (!torch.int) -> !torch.list<int>
%true_2680 = torch.constant.bool true
%none_2681 = torch.constant.none
%2460 = torch.aten.mean.dim %2458, %2459, %true_2680, %none_2681 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2460, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2682 = torch.constant.float 9.9999997473787516E-6
%int1_2683 = torch.constant.int 1
%2461 = torch.aten.add.Scalar %2460, %float9.999990e-06_2682, %int1_2683 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2461, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2462 = torch.aten.rsqrt %2461 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2462, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2463 = torch.aten.mul.Tensor %2457, %2462 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2463, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2464 = torch.aten.mul.Tensor %96, %2463 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2464, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2684 = torch.constant.int 5
%2465 = torch.prims.convert_element_type %2464, %int5_2684 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2465, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2685 = torch.constant.int -2
%int-1_2686 = torch.constant.int -1
%2466 = torch.aten.transpose.int %97, %int-2_2685, %int-1_2686 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2687 = torch.constant.int 4
%2467 = torch.aten.mul.int %int4_2687, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2688 = torch.constant.int 4096
%2468 = torch.prim.ListConstruct %2467, %int4096_2688 : (!torch.int, !torch.int) -> !torch.list<int>
%2469 = torch.aten.view %2465, %2468 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2469, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2470 = torch.aten.mm %2469, %2466 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2470, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2689 = torch.constant.int 4
%int14336_2690 = torch.constant.int 14336
%2471 = torch.prim.ListConstruct %int4_2689, %294, %int14336_2690 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2472 = torch.aten.view %2470, %2471 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2472, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2473 = torch.aten.silu %2472 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2473, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2691 = torch.constant.int -2
%int-1_2692 = torch.constant.int -1
%2474 = torch.aten.transpose.int %98, %int-2_2691, %int-1_2692 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2693 = torch.constant.int 4
%2475 = torch.aten.mul.int %int4_2693, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2694 = torch.constant.int 4096
%2476 = torch.prim.ListConstruct %2475, %int4096_2694 : (!torch.int, !torch.int) -> !torch.list<int>
%2477 = torch.aten.view %2465, %2476 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2477, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2478 = torch.aten.mm %2477, %2474 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2478, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2695 = torch.constant.int 4
%int14336_2696 = torch.constant.int 14336
%2479 = torch.prim.ListConstruct %int4_2695, %294, %int14336_2696 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2480 = torch.aten.view %2478, %2479 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2480, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2481 = torch.aten.mul.Tensor %2473, %2480 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2481, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2697 = torch.constant.int -2
%int-1_2698 = torch.constant.int -1
%2482 = torch.aten.transpose.int %99, %int-2_2697, %int-1_2698 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_2699 = torch.constant.int 1
%2483 = torch.aten.size.int %2472, %int1_2699 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_2700 = torch.constant.int 4
%2484 = torch.aten.mul.int %int4_2700, %2483 : !torch.int, !torch.int -> !torch.int
%int14336_2701 = torch.constant.int 14336
%2485 = torch.prim.ListConstruct %2484, %int14336_2701 : (!torch.int, !torch.int) -> !torch.list<int>
%2486 = torch.aten.view %2481, %2485 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2486, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%2487 = torch.aten.mm %2486, %2482 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2487, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2702 = torch.constant.int 4
%int4096_2703 = torch.constant.int 4096
%2488 = torch.prim.ListConstruct %int4_2702, %2483, %int4096_2703 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2489 = torch.aten.view %2487, %2488 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2489, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2704 = torch.constant.int 1
%2490 = torch.aten.add.Tensor %2456, %2489, %int1_2704 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2490, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2705 = torch.constant.int 6
%2491 = torch.prims.convert_element_type %2490, %int6_2705 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2491, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2706 = torch.constant.int 2
%2492 = torch.aten.pow.Tensor_Scalar %2491, %int2_2706 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2492, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2707 = torch.constant.int -1
%2493 = torch.prim.ListConstruct %int-1_2707 : (!torch.int) -> !torch.list<int>
%true_2708 = torch.constant.bool true
%none_2709 = torch.constant.none
%2494 = torch.aten.mean.dim %2492, %2493, %true_2708, %none_2709 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2494, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2710 = torch.constant.float 9.9999997473787516E-6
%int1_2711 = torch.constant.int 1
%2495 = torch.aten.add.Scalar %2494, %float9.999990e-06_2710, %int1_2711 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2495, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2496 = torch.aten.rsqrt %2495 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2496, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2497 = torch.aten.mul.Tensor %2491, %2496 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2497, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2498 = torch.aten.mul.Tensor %100, %2497 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2498, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2712 = torch.constant.int 5
%2499 = torch.prims.convert_element_type %2498, %int5_2712 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2499, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2713 = torch.constant.int -2
%int-1_2714 = torch.constant.int -1
%2500 = torch.aten.transpose.int %101, %int-2_2713, %int-1_2714 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2715 = torch.constant.int 4
%2501 = torch.aten.mul.int %int4_2715, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2716 = torch.constant.int 4096
%2502 = torch.prim.ListConstruct %2501, %int4096_2716 : (!torch.int, !torch.int) -> !torch.list<int>
%2503 = torch.aten.view %2499, %2502 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2503, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2504 = torch.aten.mm %2503, %2500 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2504, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2717 = torch.constant.int 4
%int4096_2718 = torch.constant.int 4096
%2505 = torch.prim.ListConstruct %int4_2717, %294, %int4096_2718 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2506 = torch.aten.view %2504, %2505 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2506, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2719 = torch.constant.int -2
%int-1_2720 = torch.constant.int -1
%2507 = torch.aten.transpose.int %102, %int-2_2719, %int-1_2720 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2721 = torch.constant.int 4
%2508 = torch.aten.mul.int %int4_2721, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2722 = torch.constant.int 4096
%2509 = torch.prim.ListConstruct %2508, %int4096_2722 : (!torch.int, !torch.int) -> !torch.list<int>
%2510 = torch.aten.view %2499, %2509 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2510, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2511 = torch.aten.mm %2510, %2507 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2511, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2723 = torch.constant.int 4
%int1024_2724 = torch.constant.int 1024
%2512 = torch.prim.ListConstruct %int4_2723, %294, %int1024_2724 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2513 = torch.aten.view %2511, %2512 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2513, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_2725 = torch.constant.int -2
%int-1_2726 = torch.constant.int -1
%2514 = torch.aten.transpose.int %103, %int-2_2725, %int-1_2726 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2727 = torch.constant.int 4
%2515 = torch.aten.mul.int %int4_2727, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2728 = torch.constant.int 4096
%2516 = torch.prim.ListConstruct %2515, %int4096_2728 : (!torch.int, !torch.int) -> !torch.list<int>
%2517 = torch.aten.view %2499, %2516 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2517, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2518 = torch.aten.mm %2517, %2514 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2518, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2729 = torch.constant.int 4
%int1024_2730 = torch.constant.int 1024
%2519 = torch.prim.ListConstruct %int4_2729, %294, %int1024_2730 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2520 = torch.aten.view %2518, %2519 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2520, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_2731 = torch.constant.int 4
%int32_2732 = torch.constant.int 32
%int128_2733 = torch.constant.int 128
%2521 = torch.prim.ListConstruct %int4_2731, %294, %int32_2732, %int128_2733 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2522 = torch.aten.view %2506, %2521 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2522, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2734 = torch.constant.int 4
%int8_2735 = torch.constant.int 8
%int128_2736 = torch.constant.int 128
%2523 = torch.prim.ListConstruct %int4_2734, %294, %int8_2735, %int128_2736 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2524 = torch.aten.view %2513, %2523 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2524, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_2737 = torch.constant.int 4
%int8_2738 = torch.constant.int 8
%int128_2739 = torch.constant.int 128
%2525 = torch.prim.ListConstruct %int4_2737, %294, %int8_2738, %int128_2739 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2526 = torch.aten.view %2520, %2525 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2526, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_2740 = torch.constant.int 131072
%none_2741 = torch.constant.none
%none_2742 = torch.constant.none
%cpu_2743 = torch.constant.device "cpu"
%false_2744 = torch.constant.bool false
%2527 = torch.aten.arange %int131072_2740, %none_2741, %none_2742, %cpu_2743, %false_2744 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2745 = torch.constant.int 0
%int128_2746 = torch.constant.int 128
%int2_2747 = torch.constant.int 2
%none_2748 = torch.constant.none
%none_2749 = torch.constant.none
%cpu_2750 = torch.constant.device "cpu"
%false_2751 = torch.constant.bool false
%2528 = torch.aten.arange.start_step %int0_2745, %int128_2746, %int2_2747, %none_2748, %none_2749, %cpu_2750, %false_2751 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2752 = torch.constant.int 0
%int0_2753 = torch.constant.int 0
%int64_2754 = torch.constant.int 64
%int1_2755 = torch.constant.int 1
%2529 = torch.aten.slice.Tensor %2528, %int0_2752, %int0_2753, %int64_2754, %int1_2755 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2756 = torch.constant.int 6
%2530 = torch.prims.convert_element_type %2529, %int6_2756 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2757 = torch.constant.int 128
%2531 = torch.aten.div.Scalar %2530, %int128_2757 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2758 = torch.constant.float 5.000000e+05
%2532 = torch.aten.pow.Scalar %float5.000000e05_2758, %2531 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2533 = torch.aten.reciprocal %2532 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2759 = torch.constant.float 1.000000e+00
%2534 = torch.aten.mul.Scalar %2533, %float1.000000e00_2759 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2760 = torch.constant.int 131072
%int1_2761 = torch.constant.int 1
%2535 = torch.prim.ListConstruct %int131072_2760, %int1_2761 : (!torch.int, !torch.int) -> !torch.list<int>
%2536 = torch.aten.view %2527, %2535 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2537 = torch.aten.mul.Tensor %2536, %2534 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2538 = torch.aten.cos %2537 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2539 = torch.aten.sin %2537 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2540 = torch.aten.complex %2538, %2539 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2762 = torch.constant.int 1
%2541 = torch.aten.size.int %2506, %int1_2762 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_2763 = torch.constant.int 0
%2542 = torch.aten.add.int %int0_2763, %2541 : !torch.int, !torch.int -> !torch.int
%int0_2764 = torch.constant.int 0
%int0_2765 = torch.constant.int 0
%int1_2766 = torch.constant.int 1
%2543 = torch.aten.slice.Tensor %2540, %int0_2764, %int0_2765, %2542, %int1_2766 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2543, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2767 = torch.constant.int 1
%int0_2768 = torch.constant.int 0
%int9223372036854775807_2769 = torch.constant.int 9223372036854775807
%int1_2770 = torch.constant.int 1
%2544 = torch.aten.slice.Tensor %2543, %int1_2767, %int0_2768, %int9223372036854775807_2769, %int1_2770 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2544, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2771 = torch.constant.int 0
%2545 = torch.aten.unsqueeze %2544, %int0_2771 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2545, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2772 = torch.constant.int 2
%2546 = torch.aten.unsqueeze %2545, %int2_2772 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2546, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2773 = torch.constant.int 3
%int0_2774 = torch.constant.int 0
%int9223372036854775807_2775 = torch.constant.int 9223372036854775807
%int1_2776 = torch.constant.int 1
%2547 = torch.aten.slice.Tensor %2546, %int3_2773, %int0_2774, %int9223372036854775807_2775, %int1_2776 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2547, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2548 = torch_c.to_builtin_tensor %2522 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_2777 = arith.constant 1 : index
%dim_2778 = tensor.dim %2548, %c1_2777 : tensor<4x?x32x128xf16>
%2549 = flow.tensor.bitcast %2548 : tensor<4x?x32x128xf16>{%dim_2778} -> tensor<4x?x32x64xcomplex<f16>>{%dim_2778}
%2550 = torch_c.from_builtin_tensor %2549 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %2550, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%2551 = torch.aten.mul.Tensor %2550, %2547 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %2551, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%2552 = torch_c.to_builtin_tensor %2551 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_2779 = arith.constant 1 : index
%dim_2780 = tensor.dim %2552, %c1_2779 : tensor<4x?x32x64xcomplex<f32>>
%2553 = flow.tensor.bitcast %2552 : tensor<4x?x32x64xcomplex<f32>>{%dim_2780} -> tensor<4x?x32x128xf32>{%dim_2780}
%2554 = torch_c.from_builtin_tensor %2553 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %2554, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_2781 = torch.constant.int 5
%2555 = torch.prims.convert_element_type %2554, %int5_2781 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2555, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_2782 = torch.constant.int 131072
%none_2783 = torch.constant.none
%none_2784 = torch.constant.none
%cpu_2785 = torch.constant.device "cpu"
%false_2786 = torch.constant.bool false
%2556 = torch.aten.arange %int131072_2782, %none_2783, %none_2784, %cpu_2785, %false_2786 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2787 = torch.constant.int 0
%int128_2788 = torch.constant.int 128
%int2_2789 = torch.constant.int 2
%none_2790 = torch.constant.none
%none_2791 = torch.constant.none
%cpu_2792 = torch.constant.device "cpu"
%false_2793 = torch.constant.bool false
%2557 = torch.aten.arange.start_step %int0_2787, %int128_2788, %int2_2789, %none_2790, %none_2791, %cpu_2792, %false_2793 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2794 = torch.constant.int 0
%int0_2795 = torch.constant.int 0
%int64_2796 = torch.constant.int 64
%int1_2797 = torch.constant.int 1
%2558 = torch.aten.slice.Tensor %2557, %int0_2794, %int0_2795, %int64_2796, %int1_2797 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2798 = torch.constant.int 6
%2559 = torch.prims.convert_element_type %2558, %int6_2798 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2799 = torch.constant.int 128
%2560 = torch.aten.div.Scalar %2559, %int128_2799 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_2800 = torch.constant.float 5.000000e+05
%2561 = torch.aten.pow.Scalar %float5.000000e05_2800, %2560 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2562 = torch.aten.reciprocal %2561 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_2801 = torch.constant.float 1.000000e+00
%2563 = torch.aten.mul.Scalar %2562, %float1.000000e00_2801 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_2802 = torch.constant.int 131072
%int1_2803 = torch.constant.int 1
%2564 = torch.prim.ListConstruct %int131072_2802, %int1_2803 : (!torch.int, !torch.int) -> !torch.list<int>
%2565 = torch.aten.view %2556, %2564 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2566 = torch.aten.mul.Tensor %2565, %2563 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2567 = torch.aten.cos %2566 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2568 = torch.aten.sin %2566 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2569 = torch.aten.complex %2567, %2568 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_2804 = torch.constant.int 1
%2570 = torch.aten.size.int %2513, %int1_2804 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_2805 = torch.constant.int 0
%2571 = torch.aten.add.int %int0_2805, %2570 : !torch.int, !torch.int -> !torch.int
%int0_2806 = torch.constant.int 0
%int0_2807 = torch.constant.int 0
%int1_2808 = torch.constant.int 1
%2572 = torch.aten.slice.Tensor %2569, %int0_2806, %int0_2807, %2571, %int1_2808 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2572, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_2809 = torch.constant.int 1
%int0_2810 = torch.constant.int 0
%int9223372036854775807_2811 = torch.constant.int 9223372036854775807
%int1_2812 = torch.constant.int 1
%2573 = torch.aten.slice.Tensor %2572, %int1_2809, %int0_2810, %int9223372036854775807_2811, %int1_2812 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2573, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_2813 = torch.constant.int 0
%2574 = torch.aten.unsqueeze %2573, %int0_2813 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2574, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_2814 = torch.constant.int 2
%2575 = torch.aten.unsqueeze %2574, %int2_2814 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2575, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_2815 = torch.constant.int 3
%int0_2816 = torch.constant.int 0
%int9223372036854775807_2817 = torch.constant.int 9223372036854775807
%int1_2818 = torch.constant.int 1
%2576 = torch.aten.slice.Tensor %2575, %int3_2815, %int0_2816, %int9223372036854775807_2817, %int1_2818 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2576, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2577 = torch_c.to_builtin_tensor %2524 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_2819 = arith.constant 1 : index
%dim_2820 = tensor.dim %2577, %c1_2819 : tensor<4x?x8x128xf16>
%2578 = flow.tensor.bitcast %2577 : tensor<4x?x8x128xf16>{%dim_2820} -> tensor<4x?x8x64xcomplex<f16>>{%dim_2820}
%2579 = torch_c.from_builtin_tensor %2578 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %2579, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%2580 = torch.aten.mul.Tensor %2579, %2576 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %2580, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%2581 = torch_c.to_builtin_tensor %2580 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_2821 = arith.constant 1 : index
%dim_2822 = tensor.dim %2581, %c1_2821 : tensor<4x?x8x64xcomplex<f32>>
%2582 = flow.tensor.bitcast %2581 : tensor<4x?x8x64xcomplex<f32>>{%dim_2822} -> tensor<4x?x8x128xf32>{%dim_2822}
%2583 = torch_c.from_builtin_tensor %2582 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %2583, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_2823 = torch.constant.int 5
%2584 = torch.prims.convert_element_type %2583, %int5_2823 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2584, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_2824 = torch.constant.int 64
%2585 = torch.aten.mul.Scalar %arg2, %int64_2824 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2585, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int22 = torch.constant.int 22
%int1_2825 = torch.constant.int 1
%2586 = torch.aten.add.Scalar %2585, %int22, %int1_2825 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2586, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2826 = torch.constant.int 4
%int32_2827 = torch.constant.int 32
%int8_2828 = torch.constant.int 8
%int128_2829 = torch.constant.int 128
%2587 = torch.prim.ListConstruct %int4_2826, %425, %int32_2827, %int8_2828, %int128_2829 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2588 = torch.aten.view %2584, %2587 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2588, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2830 = torch.constant.int 4
%2589 = torch.aten.mul.int %int4_2830, %425 : !torch.int, !torch.int -> !torch.int
%int32_2831 = torch.constant.int 32
%int8_2832 = torch.constant.int 8
%int128_2833 = torch.constant.int 128
%2590 = torch.prim.ListConstruct %2589, %int32_2831, %int8_2832, %int128_2833 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2591 = torch.aten.view %2588, %2590 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2591, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2834 = torch.constant.int 4
%2592 = torch.aten.mul.int %int4_2834, %425 : !torch.int, !torch.int -> !torch.int
%2593 = torch.prim.ListConstruct %2592 : (!torch.int) -> !torch.list<int>
%2594 = torch.aten.view %2586, %2593 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2594, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_2835 = torch.constant.int 32
%int2_2836 = torch.constant.int 2
%int32_2837 = torch.constant.int 32
%int8_2838 = torch.constant.int 8
%int128_2839 = torch.constant.int 128
%2595 = torch.prim.ListConstruct %416, %int32_2835, %int2_2836, %int32_2837, %int8_2838, %int128_2839 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2596 = torch.aten.view %2428, %2595 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2596, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2840 = torch.constant.int 32
%2597 = torch.aten.mul.int %416, %int32_2840 : !torch.int, !torch.int -> !torch.int
%int2_2841 = torch.constant.int 2
%2598 = torch.aten.mul.int %2597, %int2_2841 : !torch.int, !torch.int -> !torch.int
%int32_2842 = torch.constant.int 32
%int8_2843 = torch.constant.int 8
%int128_2844 = torch.constant.int 128
%2599 = torch.prim.ListConstruct %2598, %int32_2842, %int8_2843, %int128_2844 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2600 = torch.aten.view %2596, %2599 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2600, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2601 = torch.prim.ListConstruct %2594 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2845 = torch.constant.bool false
%2602 = torch.aten.index_put %2600, %2601, %2591, %false_2845 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2602, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2846 = torch.constant.int 32
%int2_2847 = torch.constant.int 2
%int32_2848 = torch.constant.int 32
%int8_2849 = torch.constant.int 8
%int128_2850 = torch.constant.int 128
%2603 = torch.prim.ListConstruct %416, %int32_2846, %int2_2847, %int32_2848, %int8_2849, %int128_2850 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2604 = torch.aten.view %2602, %2603 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2604, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2851 = torch.constant.int 2097152
%2605 = torch.prim.ListConstruct %416, %int2097152_2851 : (!torch.int, !torch.int) -> !torch.list<int>
%2606 = torch.aten.view %2604, %2605 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2606, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_2852 = torch.constant.int 32
%int2_2853 = torch.constant.int 2
%int32_2854 = torch.constant.int 32
%int8_2855 = torch.constant.int 8
%int128_2856 = torch.constant.int 128
%2607 = torch.prim.ListConstruct %416, %int32_2852, %int2_2853, %int32_2854, %int8_2855, %int128_2856 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2608 = torch.aten.view %2606, %2607 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2608, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_2857 = torch.constant.int 32
%int8_2858 = torch.constant.int 8
%int128_2859 = torch.constant.int 128
%2609 = torch.prim.ListConstruct %2598, %int32_2857, %int8_2858, %int128_2859 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2610 = torch.aten.view %2608, %2609 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2610, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_2860 = torch.constant.int 4
%int32_2861 = torch.constant.int 32
%int8_2862 = torch.constant.int 8
%int128_2863 = torch.constant.int 128
%2611 = torch.prim.ListConstruct %int4_2860, %425, %int32_2861, %int8_2862, %int128_2863 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2612 = torch.aten.view %2526, %2611 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2612, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_2864 = torch.constant.int 4
%2613 = torch.aten.mul.int %int4_2864, %425 : !torch.int, !torch.int -> !torch.int
%int32_2865 = torch.constant.int 32
%int8_2866 = torch.constant.int 8
%int128_2867 = torch.constant.int 128
%2614 = torch.prim.ListConstruct %2613, %int32_2865, %int8_2866, %int128_2867 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2615 = torch.aten.view %2612, %2614 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2615, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_2868 = torch.constant.int 1
%int1_2869 = torch.constant.int 1
%2616 = torch.aten.add.Scalar %2586, %int1_2868, %int1_2869 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2616, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_2870 = torch.constant.int 4
%2617 = torch.aten.mul.int %int4_2870, %425 : !torch.int, !torch.int -> !torch.int
%2618 = torch.prim.ListConstruct %2617 : (!torch.int) -> !torch.list<int>
%2619 = torch.aten.view %2616, %2618 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2619, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%2620 = torch.prim.ListConstruct %2619 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_2871 = torch.constant.bool false
%2621 = torch.aten.index_put %2610, %2620, %2615, %false_2871 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2621, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_2872 = torch.constant.int 32
%int2_2873 = torch.constant.int 2
%int32_2874 = torch.constant.int 32
%int8_2875 = torch.constant.int 8
%int128_2876 = torch.constant.int 128
%2622 = torch.prim.ListConstruct %416, %int32_2872, %int2_2873, %int32_2874, %int8_2875, %int128_2876 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2623 = torch.aten.view %2621, %2622 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2623, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_2877 = torch.constant.int 2097152
%2624 = torch.prim.ListConstruct %416, %int2097152_2877 : (!torch.int, !torch.int) -> !torch.list<int>
%2625 = torch.aten.view %2623, %2624 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2625, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_2878 = torch.constant.int -2
%2626 = torch.aten.unsqueeze %2584, %int-2_2878 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2626, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_2879 = torch.constant.int 4
%int8_2880 = torch.constant.int 8
%int4_2881 = torch.constant.int 4
%int128_2882 = torch.constant.int 128
%2627 = torch.prim.ListConstruct %int4_2879, %2570, %int8_2880, %int4_2881, %int128_2882 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2883 = torch.constant.bool false
%2628 = torch.aten.expand %2626, %2627, %false_2883 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2628, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2884 = torch.constant.int 0
%2629 = torch.aten.clone %2628, %int0_2884 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2629, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2885 = torch.constant.int 4
%int32_2886 = torch.constant.int 32
%int128_2887 = torch.constant.int 128
%2630 = torch.prim.ListConstruct %int4_2885, %2570, %int32_2886, %int128_2887 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2631 = torch.aten._unsafe_view %2629, %2630 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2631, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_2888 = torch.constant.int -2
%2632 = torch.aten.unsqueeze %2526, %int-2_2888 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2632, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_2889 = torch.constant.int 1
%2633 = torch.aten.size.int %2520, %int1_2889 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_2890 = torch.constant.int 4
%int8_2891 = torch.constant.int 8
%int4_2892 = torch.constant.int 4
%int128_2893 = torch.constant.int 128
%2634 = torch.prim.ListConstruct %int4_2890, %2633, %int8_2891, %int4_2892, %int128_2893 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_2894 = torch.constant.bool false
%2635 = torch.aten.expand %2632, %2634, %false_2894 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2635, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_2895 = torch.constant.int 0
%2636 = torch.aten.clone %2635, %int0_2895 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2636, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_2896 = torch.constant.int 4
%int32_2897 = torch.constant.int 32
%int128_2898 = torch.constant.int 128
%2637 = torch.prim.ListConstruct %int4_2896, %2633, %int32_2897, %int128_2898 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2638 = torch.aten._unsafe_view %2636, %2637 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2638, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_2899 = torch.constant.int 1
%int2_2900 = torch.constant.int 2
%2639 = torch.aten.transpose.int %2555, %int1_2899, %int2_2900 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2639, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2901 = torch.constant.int 1
%int2_2902 = torch.constant.int 2
%2640 = torch.aten.transpose.int %2631, %int1_2901, %int2_2902 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2640, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2903 = torch.constant.int 1
%int2_2904 = torch.constant.int 2
%2641 = torch.aten.transpose.int %2638, %int1_2903, %int2_2904 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2641, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_2905 = torch.constant.float 0.000000e+00
%false_2906 = torch.constant.bool false
%none_2907 = torch.constant.none
%2642:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%2639, %2640, %2641, %float0.000000e00_2905, %false_2906, %320, %none_2907) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %2642#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_2908 = torch.constant.int 1
%int2_2909 = torch.constant.int 2
%2643 = torch.aten.transpose.int %2642#0, %int1_2908, %int2_2909 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2643, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2910 = torch.constant.int 4
%int4096_2911 = torch.constant.int 4096
%2644 = torch.prim.ListConstruct %int4_2910, %2541, %int4096_2911 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2645 = torch.aten.view %2643, %2644 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2645, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2912 = torch.constant.int -2
%int-1_2913 = torch.constant.int -1
%2646 = torch.aten.transpose.int %104, %int-2_2912, %int-1_2913 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2914 = torch.constant.int 4
%2647 = torch.aten.mul.int %int4_2914, %2541 : !torch.int, !torch.int -> !torch.int
%int4096_2915 = torch.constant.int 4096
%2648 = torch.prim.ListConstruct %2647, %int4096_2915 : (!torch.int, !torch.int) -> !torch.list<int>
%2649 = torch.aten.view %2645, %2648 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2649, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2650 = torch.aten.mm %2649, %2646 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2650, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2916 = torch.constant.int 4
%int4096_2917 = torch.constant.int 4096
%2651 = torch.prim.ListConstruct %int4_2916, %2541, %int4096_2917 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2652 = torch.aten.view %2650, %2651 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2652, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2918 = torch.constant.int 1
%2653 = torch.aten.add.Tensor %2490, %2652, %int1_2918 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2653, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2919 = torch.constant.int 6
%2654 = torch.prims.convert_element_type %2653, %int6_2919 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2654, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2920 = torch.constant.int 2
%2655 = torch.aten.pow.Tensor_Scalar %2654, %int2_2920 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2655, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2921 = torch.constant.int -1
%2656 = torch.prim.ListConstruct %int-1_2921 : (!torch.int) -> !torch.list<int>
%true_2922 = torch.constant.bool true
%none_2923 = torch.constant.none
%2657 = torch.aten.mean.dim %2655, %2656, %true_2922, %none_2923 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2657, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2924 = torch.constant.float 9.9999997473787516E-6
%int1_2925 = torch.constant.int 1
%2658 = torch.aten.add.Scalar %2657, %float9.999990e-06_2924, %int1_2925 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2658, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2659 = torch.aten.rsqrt %2658 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2659, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2660 = torch.aten.mul.Tensor %2654, %2659 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2660, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2661 = torch.aten.mul.Tensor %105, %2660 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2661, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2926 = torch.constant.int 5
%2662 = torch.prims.convert_element_type %2661, %int5_2926 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2662, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2927 = torch.constant.int -2
%int-1_2928 = torch.constant.int -1
%2663 = torch.aten.transpose.int %106, %int-2_2927, %int-1_2928 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2929 = torch.constant.int 4
%2664 = torch.aten.mul.int %int4_2929, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2930 = torch.constant.int 4096
%2665 = torch.prim.ListConstruct %2664, %int4096_2930 : (!torch.int, !torch.int) -> !torch.list<int>
%2666 = torch.aten.view %2662, %2665 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2666, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2667 = torch.aten.mm %2666, %2663 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2667, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2931 = torch.constant.int 4
%int14336_2932 = torch.constant.int 14336
%2668 = torch.prim.ListConstruct %int4_2931, %294, %int14336_2932 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2669 = torch.aten.view %2667, %2668 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2669, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2670 = torch.aten.silu %2669 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2670, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2933 = torch.constant.int -2
%int-1_2934 = torch.constant.int -1
%2671 = torch.aten.transpose.int %107, %int-2_2933, %int-1_2934 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_2935 = torch.constant.int 4
%2672 = torch.aten.mul.int %int4_2935, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2936 = torch.constant.int 4096
%2673 = torch.prim.ListConstruct %2672, %int4096_2936 : (!torch.int, !torch.int) -> !torch.list<int>
%2674 = torch.aten.view %2662, %2673 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2674, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2675 = torch.aten.mm %2674, %2671 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2675, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_2937 = torch.constant.int 4
%int14336_2938 = torch.constant.int 14336
%2676 = torch.prim.ListConstruct %int4_2937, %294, %int14336_2938 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2677 = torch.aten.view %2675, %2676 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2677, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2678 = torch.aten.mul.Tensor %2670, %2677 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2678, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_2939 = torch.constant.int -2
%int-1_2940 = torch.constant.int -1
%2679 = torch.aten.transpose.int %108, %int-2_2939, %int-1_2940 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_2941 = torch.constant.int 1
%2680 = torch.aten.size.int %2669, %int1_2941 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_2942 = torch.constant.int 4
%2681 = torch.aten.mul.int %int4_2942, %2680 : !torch.int, !torch.int -> !torch.int
%int14336_2943 = torch.constant.int 14336
%2682 = torch.prim.ListConstruct %2681, %int14336_2943 : (!torch.int, !torch.int) -> !torch.list<int>
%2683 = torch.aten.view %2678, %2682 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2683, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%2684 = torch.aten.mm %2683, %2679 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2684, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2944 = torch.constant.int 4
%int4096_2945 = torch.constant.int 4096
%2685 = torch.prim.ListConstruct %int4_2944, %2680, %int4096_2945 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2686 = torch.aten.view %2684, %2685 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2686, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_2946 = torch.constant.int 1
%2687 = torch.aten.add.Tensor %2653, %2686, %int1_2946 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2687, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_2947 = torch.constant.int 6
%2688 = torch.prims.convert_element_type %2687, %int6_2947 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2688, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_2948 = torch.constant.int 2
%2689 = torch.aten.pow.Tensor_Scalar %2688, %int2_2948 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2689, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_2949 = torch.constant.int -1
%2690 = torch.prim.ListConstruct %int-1_2949 : (!torch.int) -> !torch.list<int>
%true_2950 = torch.constant.bool true
%none_2951 = torch.constant.none
%2691 = torch.aten.mean.dim %2689, %2690, %true_2950, %none_2951 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2691, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_2952 = torch.constant.float 9.9999997473787516E-6
%int1_2953 = torch.constant.int 1
%2692 = torch.aten.add.Scalar %2691, %float9.999990e-06_2952, %int1_2953 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2692, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2693 = torch.aten.rsqrt %2692 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2693, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2694 = torch.aten.mul.Tensor %2688, %2693 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2694, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2695 = torch.aten.mul.Tensor %109, %2694 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2695, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_2954 = torch.constant.int 5
%2696 = torch.prims.convert_element_type %2695, %int5_2954 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2696, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2955 = torch.constant.int -2
%int-1_2956 = torch.constant.int -1
%2697 = torch.aten.transpose.int %110, %int-2_2955, %int-1_2956 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_2957 = torch.constant.int 4
%2698 = torch.aten.mul.int %int4_2957, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2958 = torch.constant.int 4096
%2699 = torch.prim.ListConstruct %2698, %int4096_2958 : (!torch.int, !torch.int) -> !torch.list<int>
%2700 = torch.aten.view %2696, %2699 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2700, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2701 = torch.aten.mm %2700, %2697 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2701, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_2959 = torch.constant.int 4
%int4096_2960 = torch.constant.int 4096
%2702 = torch.prim.ListConstruct %int4_2959, %294, %int4096_2960 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2703 = torch.aten.view %2701, %2702 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2703, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_2961 = torch.constant.int -2
%int-1_2962 = torch.constant.int -1
%2704 = torch.aten.transpose.int %111, %int-2_2961, %int-1_2962 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2963 = torch.constant.int 4
%2705 = torch.aten.mul.int %int4_2963, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2964 = torch.constant.int 4096
%2706 = torch.prim.ListConstruct %2705, %int4096_2964 : (!torch.int, !torch.int) -> !torch.list<int>
%2707 = torch.aten.view %2696, %2706 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2707, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2708 = torch.aten.mm %2707, %2704 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2708, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2965 = torch.constant.int 4
%int1024_2966 = torch.constant.int 1024
%2709 = torch.prim.ListConstruct %int4_2965, %294, %int1024_2966 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2710 = torch.aten.view %2708, %2709 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2710, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_2967 = torch.constant.int -2
%int-1_2968 = torch.constant.int -1
%2711 = torch.aten.transpose.int %112, %int-2_2967, %int-1_2968 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_2969 = torch.constant.int 4
%2712 = torch.aten.mul.int %int4_2969, %294 : !torch.int, !torch.int -> !torch.int
%int4096_2970 = torch.constant.int 4096
%2713 = torch.prim.ListConstruct %2712, %int4096_2970 : (!torch.int, !torch.int) -> !torch.list<int>
%2714 = torch.aten.view %2696, %2713 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2714, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2715 = torch.aten.mm %2714, %2711 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2715, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_2971 = torch.constant.int 4
%int1024_2972 = torch.constant.int 1024
%2716 = torch.prim.ListConstruct %int4_2971, %294, %int1024_2972 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2717 = torch.aten.view %2715, %2716 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2717, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_2973 = torch.constant.int 4
%int32_2974 = torch.constant.int 32
%int128_2975 = torch.constant.int 128
%2718 = torch.prim.ListConstruct %int4_2973, %294, %int32_2974, %int128_2975 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2719 = torch.aten.view %2703, %2718 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2719, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_2976 = torch.constant.int 4
%int8_2977 = torch.constant.int 8
%int128_2978 = torch.constant.int 128
%2720 = torch.prim.ListConstruct %int4_2976, %294, %int8_2977, %int128_2978 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2721 = torch.aten.view %2710, %2720 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2721, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_2979 = torch.constant.int 4
%int8_2980 = torch.constant.int 8
%int128_2981 = torch.constant.int 128
%2722 = torch.prim.ListConstruct %int4_2979, %294, %int8_2980, %int128_2981 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2723 = torch.aten.view %2717, %2722 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2723, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_2982 = torch.constant.int 131072
%none_2983 = torch.constant.none
%none_2984 = torch.constant.none
%cpu_2985 = torch.constant.device "cpu"
%false_2986 = torch.constant.bool false
%2724 = torch.aten.arange %int131072_2982, %none_2983, %none_2984, %cpu_2985, %false_2986 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_2987 = torch.constant.int 0
%int128_2988 = torch.constant.int 128
%int2_2989 = torch.constant.int 2
%none_2990 = torch.constant.none
%none_2991 = torch.constant.none
%cpu_2992 = torch.constant.device "cpu"
%false_2993 = torch.constant.bool false
%2725 = torch.aten.arange.start_step %int0_2987, %int128_2988, %int2_2989, %none_2990, %none_2991, %cpu_2992, %false_2993 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_2994 = torch.constant.int 0
%int0_2995 = torch.constant.int 0
%int64_2996 = torch.constant.int 64
%int1_2997 = torch.constant.int 1
%2726 = torch.aten.slice.Tensor %2725, %int0_2994, %int0_2995, %int64_2996, %int1_2997 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_2998 = torch.constant.int 6
%2727 = torch.prims.convert_element_type %2726, %int6_2998 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_2999 = torch.constant.int 128
%2728 = torch.aten.div.Scalar %2727, %int128_2999 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3000 = torch.constant.float 5.000000e+05
%2729 = torch.aten.pow.Scalar %float5.000000e05_3000, %2728 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2730 = torch.aten.reciprocal %2729 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3001 = torch.constant.float 1.000000e+00
%2731 = torch.aten.mul.Scalar %2730, %float1.000000e00_3001 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3002 = torch.constant.int 131072
%int1_3003 = torch.constant.int 1
%2732 = torch.prim.ListConstruct %int131072_3002, %int1_3003 : (!torch.int, !torch.int) -> !torch.list<int>
%2733 = torch.aten.view %2724, %2732 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2734 = torch.aten.mul.Tensor %2733, %2731 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2735 = torch.aten.cos %2734 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2736 = torch.aten.sin %2734 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2737 = torch.aten.complex %2735, %2736 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3004 = torch.constant.int 1
%2738 = torch.aten.size.int %2703, %int1_3004 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_3005 = torch.constant.int 0
%2739 = torch.aten.add.int %int0_3005, %2738 : !torch.int, !torch.int -> !torch.int
%int0_3006 = torch.constant.int 0
%int0_3007 = torch.constant.int 0
%int1_3008 = torch.constant.int 1
%2740 = torch.aten.slice.Tensor %2737, %int0_3006, %int0_3007, %2739, %int1_3008 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2740, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3009 = torch.constant.int 1
%int0_3010 = torch.constant.int 0
%int9223372036854775807_3011 = torch.constant.int 9223372036854775807
%int1_3012 = torch.constant.int 1
%2741 = torch.aten.slice.Tensor %2740, %int1_3009, %int0_3010, %int9223372036854775807_3011, %int1_3012 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2741, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3013 = torch.constant.int 0
%2742 = torch.aten.unsqueeze %2741, %int0_3013 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2742, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3014 = torch.constant.int 2
%2743 = torch.aten.unsqueeze %2742, %int2_3014 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2743, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3015 = torch.constant.int 3
%int0_3016 = torch.constant.int 0
%int9223372036854775807_3017 = torch.constant.int 9223372036854775807
%int1_3018 = torch.constant.int 1
%2744 = torch.aten.slice.Tensor %2743, %int3_3015, %int0_3016, %int9223372036854775807_3017, %int1_3018 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2744, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2745 = torch_c.to_builtin_tensor %2719 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_3019 = arith.constant 1 : index
%dim_3020 = tensor.dim %2745, %c1_3019 : tensor<4x?x32x128xf16>
%2746 = flow.tensor.bitcast %2745 : tensor<4x?x32x128xf16>{%dim_3020} -> tensor<4x?x32x64xcomplex<f16>>{%dim_3020}
%2747 = torch_c.from_builtin_tensor %2746 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %2747, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%2748 = torch.aten.mul.Tensor %2747, %2744 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %2748, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%2749 = torch_c.to_builtin_tensor %2748 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_3021 = arith.constant 1 : index
%dim_3022 = tensor.dim %2749, %c1_3021 : tensor<4x?x32x64xcomplex<f32>>
%2750 = flow.tensor.bitcast %2749 : tensor<4x?x32x64xcomplex<f32>>{%dim_3022} -> tensor<4x?x32x128xf32>{%dim_3022}
%2751 = torch_c.from_builtin_tensor %2750 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %2751, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_3023 = torch.constant.int 5
%2752 = torch.prims.convert_element_type %2751, %int5_3023 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2752, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_3024 = torch.constant.int 131072
%none_3025 = torch.constant.none
%none_3026 = torch.constant.none
%cpu_3027 = torch.constant.device "cpu"
%false_3028 = torch.constant.bool false
%2753 = torch.aten.arange %int131072_3024, %none_3025, %none_3026, %cpu_3027, %false_3028 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3029 = torch.constant.int 0
%int128_3030 = torch.constant.int 128
%int2_3031 = torch.constant.int 2
%none_3032 = torch.constant.none
%none_3033 = torch.constant.none
%cpu_3034 = torch.constant.device "cpu"
%false_3035 = torch.constant.bool false
%2754 = torch.aten.arange.start_step %int0_3029, %int128_3030, %int2_3031, %none_3032, %none_3033, %cpu_3034, %false_3035 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3036 = torch.constant.int 0
%int0_3037 = torch.constant.int 0
%int64_3038 = torch.constant.int 64
%int1_3039 = torch.constant.int 1
%2755 = torch.aten.slice.Tensor %2754, %int0_3036, %int0_3037, %int64_3038, %int1_3039 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3040 = torch.constant.int 6
%2756 = torch.prims.convert_element_type %2755, %int6_3040 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3041 = torch.constant.int 128
%2757 = torch.aten.div.Scalar %2756, %int128_3041 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3042 = torch.constant.float 5.000000e+05
%2758 = torch.aten.pow.Scalar %float5.000000e05_3042, %2757 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2759 = torch.aten.reciprocal %2758 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3043 = torch.constant.float 1.000000e+00
%2760 = torch.aten.mul.Scalar %2759, %float1.000000e00_3043 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3044 = torch.constant.int 131072
%int1_3045 = torch.constant.int 1
%2761 = torch.prim.ListConstruct %int131072_3044, %int1_3045 : (!torch.int, !torch.int) -> !torch.list<int>
%2762 = torch.aten.view %2753, %2761 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2763 = torch.aten.mul.Tensor %2762, %2760 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2764 = torch.aten.cos %2763 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2765 = torch.aten.sin %2763 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2766 = torch.aten.complex %2764, %2765 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3046 = torch.constant.int 1
%2767 = torch.aten.size.int %2710, %int1_3046 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_3047 = torch.constant.int 0
%2768 = torch.aten.add.int %int0_3047, %2767 : !torch.int, !torch.int -> !torch.int
%int0_3048 = torch.constant.int 0
%int0_3049 = torch.constant.int 0
%int1_3050 = torch.constant.int 1
%2769 = torch.aten.slice.Tensor %2766, %int0_3048, %int0_3049, %2768, %int1_3050 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2769, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3051 = torch.constant.int 1
%int0_3052 = torch.constant.int 0
%int9223372036854775807_3053 = torch.constant.int 9223372036854775807
%int1_3054 = torch.constant.int 1
%2770 = torch.aten.slice.Tensor %2769, %int1_3051, %int0_3052, %int9223372036854775807_3053, %int1_3054 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2770, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3055 = torch.constant.int 0
%2771 = torch.aten.unsqueeze %2770, %int0_3055 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2771, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3056 = torch.constant.int 2
%2772 = torch.aten.unsqueeze %2771, %int2_3056 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2772, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3057 = torch.constant.int 3
%int0_3058 = torch.constant.int 0
%int9223372036854775807_3059 = torch.constant.int 9223372036854775807
%int1_3060 = torch.constant.int 1
%2773 = torch.aten.slice.Tensor %2772, %int3_3057, %int0_3058, %int9223372036854775807_3059, %int1_3060 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2773, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2774 = torch_c.to_builtin_tensor %2721 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_3061 = arith.constant 1 : index
%dim_3062 = tensor.dim %2774, %c1_3061 : tensor<4x?x8x128xf16>
%2775 = flow.tensor.bitcast %2774 : tensor<4x?x8x128xf16>{%dim_3062} -> tensor<4x?x8x64xcomplex<f16>>{%dim_3062}
%2776 = torch_c.from_builtin_tensor %2775 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %2776, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%2777 = torch.aten.mul.Tensor %2776, %2773 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %2777, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%2778 = torch_c.to_builtin_tensor %2777 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_3063 = arith.constant 1 : index
%dim_3064 = tensor.dim %2778, %c1_3063 : tensor<4x?x8x64xcomplex<f32>>
%2779 = flow.tensor.bitcast %2778 : tensor<4x?x8x64xcomplex<f32>>{%dim_3064} -> tensor<4x?x8x128xf32>{%dim_3064}
%2780 = torch_c.from_builtin_tensor %2779 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %2780, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_3065 = torch.constant.int 5
%2781 = torch.prims.convert_element_type %2780, %int5_3065 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2781, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_3066 = torch.constant.int 64
%2782 = torch.aten.mul.Scalar %arg2, %int64_3066 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2782, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int24 = torch.constant.int 24
%int1_3067 = torch.constant.int 1
%2783 = torch.aten.add.Scalar %2782, %int24, %int1_3067 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2783, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3068 = torch.constant.int 4
%int32_3069 = torch.constant.int 32
%int8_3070 = torch.constant.int 8
%int128_3071 = torch.constant.int 128
%2784 = torch.prim.ListConstruct %int4_3068, %425, %int32_3069, %int8_3070, %int128_3071 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2785 = torch.aten.view %2781, %2784 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2785, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3072 = torch.constant.int 4
%2786 = torch.aten.mul.int %int4_3072, %425 : !torch.int, !torch.int -> !torch.int
%int32_3073 = torch.constant.int 32
%int8_3074 = torch.constant.int 8
%int128_3075 = torch.constant.int 128
%2787 = torch.prim.ListConstruct %2786, %int32_3073, %int8_3074, %int128_3075 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2788 = torch.aten.view %2785, %2787 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2788, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3076 = torch.constant.int 4
%2789 = torch.aten.mul.int %int4_3076, %425 : !torch.int, !torch.int -> !torch.int
%2790 = torch.prim.ListConstruct %2789 : (!torch.int) -> !torch.list<int>
%2791 = torch.aten.view %2783, %2790 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2791, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_3077 = torch.constant.int 32
%int2_3078 = torch.constant.int 2
%int32_3079 = torch.constant.int 32
%int8_3080 = torch.constant.int 8
%int128_3081 = torch.constant.int 128
%2792 = torch.prim.ListConstruct %416, %int32_3077, %int2_3078, %int32_3079, %int8_3080, %int128_3081 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2793 = torch.aten.view %2625, %2792 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2793, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3082 = torch.constant.int 32
%2794 = torch.aten.mul.int %416, %int32_3082 : !torch.int, !torch.int -> !torch.int
%int2_3083 = torch.constant.int 2
%2795 = torch.aten.mul.int %2794, %int2_3083 : !torch.int, !torch.int -> !torch.int
%int32_3084 = torch.constant.int 32
%int8_3085 = torch.constant.int 8
%int128_3086 = torch.constant.int 128
%2796 = torch.prim.ListConstruct %2795, %int32_3084, %int8_3085, %int128_3086 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2797 = torch.aten.view %2793, %2796 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2797, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2798 = torch.prim.ListConstruct %2791 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3087 = torch.constant.bool false
%2799 = torch.aten.index_put %2797, %2798, %2788, %false_3087 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2799, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3088 = torch.constant.int 32
%int2_3089 = torch.constant.int 2
%int32_3090 = torch.constant.int 32
%int8_3091 = torch.constant.int 8
%int128_3092 = torch.constant.int 128
%2800 = torch.prim.ListConstruct %416, %int32_3088, %int2_3089, %int32_3090, %int8_3091, %int128_3092 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2801 = torch.aten.view %2799, %2800 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2801, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3093 = torch.constant.int 2097152
%2802 = torch.prim.ListConstruct %416, %int2097152_3093 : (!torch.int, !torch.int) -> !torch.list<int>
%2803 = torch.aten.view %2801, %2802 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2803, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_3094 = torch.constant.int 32
%int2_3095 = torch.constant.int 2
%int32_3096 = torch.constant.int 32
%int8_3097 = torch.constant.int 8
%int128_3098 = torch.constant.int 128
%2804 = torch.prim.ListConstruct %416, %int32_3094, %int2_3095, %int32_3096, %int8_3097, %int128_3098 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2805 = torch.aten.view %2803, %2804 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2805, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3099 = torch.constant.int 32
%int8_3100 = torch.constant.int 8
%int128_3101 = torch.constant.int 128
%2806 = torch.prim.ListConstruct %2795, %int32_3099, %int8_3100, %int128_3101 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2807 = torch.aten.view %2805, %2806 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2807, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3102 = torch.constant.int 4
%int32_3103 = torch.constant.int 32
%int8_3104 = torch.constant.int 8
%int128_3105 = torch.constant.int 128
%2808 = torch.prim.ListConstruct %int4_3102, %425, %int32_3103, %int8_3104, %int128_3105 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2809 = torch.aten.view %2723, %2808 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2809, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3106 = torch.constant.int 4
%2810 = torch.aten.mul.int %int4_3106, %425 : !torch.int, !torch.int -> !torch.int
%int32_3107 = torch.constant.int 32
%int8_3108 = torch.constant.int 8
%int128_3109 = torch.constant.int 128
%2811 = torch.prim.ListConstruct %2810, %int32_3107, %int8_3108, %int128_3109 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2812 = torch.aten.view %2809, %2811 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2812, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_3110 = torch.constant.int 1
%int1_3111 = torch.constant.int 1
%2813 = torch.aten.add.Scalar %2783, %int1_3110, %int1_3111 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2813, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3112 = torch.constant.int 4
%2814 = torch.aten.mul.int %int4_3112, %425 : !torch.int, !torch.int -> !torch.int
%2815 = torch.prim.ListConstruct %2814 : (!torch.int) -> !torch.list<int>
%2816 = torch.aten.view %2813, %2815 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2816, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%2817 = torch.prim.ListConstruct %2816 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3113 = torch.constant.bool false
%2818 = torch.aten.index_put %2807, %2817, %2812, %false_3113 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2818, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3114 = torch.constant.int 32
%int2_3115 = torch.constant.int 2
%int32_3116 = torch.constant.int 32
%int8_3117 = torch.constant.int 8
%int128_3118 = torch.constant.int 128
%2819 = torch.prim.ListConstruct %416, %int32_3114, %int2_3115, %int32_3116, %int8_3117, %int128_3118 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2820 = torch.aten.view %2818, %2819 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2820, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3119 = torch.constant.int 2097152
%2821 = torch.prim.ListConstruct %416, %int2097152_3119 : (!torch.int, !torch.int) -> !torch.list<int>
%2822 = torch.aten.view %2820, %2821 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %2822, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_3120 = torch.constant.int -2
%2823 = torch.aten.unsqueeze %2781, %int-2_3120 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2823, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_3121 = torch.constant.int 4
%int8_3122 = torch.constant.int 8
%int4_3123 = torch.constant.int 4
%int128_3124 = torch.constant.int 128
%2824 = torch.prim.ListConstruct %int4_3121, %2767, %int8_3122, %int4_3123, %int128_3124 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3125 = torch.constant.bool false
%2825 = torch.aten.expand %2823, %2824, %false_3125 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2825, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3126 = torch.constant.int 0
%2826 = torch.aten.clone %2825, %int0_3126 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2826, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3127 = torch.constant.int 4
%int32_3128 = torch.constant.int 32
%int128_3129 = torch.constant.int 128
%2827 = torch.prim.ListConstruct %int4_3127, %2767, %int32_3128, %int128_3129 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2828 = torch.aten._unsafe_view %2826, %2827 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2828, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_3130 = torch.constant.int -2
%2829 = torch.aten.unsqueeze %2723, %int-2_3130 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %2829, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_3131 = torch.constant.int 1
%2830 = torch.aten.size.int %2717, %int1_3131 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_3132 = torch.constant.int 4
%int8_3133 = torch.constant.int 8
%int4_3134 = torch.constant.int 4
%int128_3135 = torch.constant.int 128
%2831 = torch.prim.ListConstruct %int4_3132, %2830, %int8_3133, %int4_3134, %int128_3135 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3136 = torch.constant.bool false
%2832 = torch.aten.expand %2829, %2831, %false_3136 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2832, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3137 = torch.constant.int 0
%2833 = torch.aten.clone %2832, %int0_3137 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %2833, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3138 = torch.constant.int 4
%int32_3139 = torch.constant.int 32
%int128_3140 = torch.constant.int 128
%2834 = torch.prim.ListConstruct %int4_3138, %2830, %int32_3139, %int128_3140 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2835 = torch.aten._unsafe_view %2833, %2834 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2835, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_3141 = torch.constant.int 1
%int2_3142 = torch.constant.int 2
%2836 = torch.aten.transpose.int %2752, %int1_3141, %int2_3142 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2836, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3143 = torch.constant.int 1
%int2_3144 = torch.constant.int 2
%2837 = torch.aten.transpose.int %2828, %int1_3143, %int2_3144 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2837, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3145 = torch.constant.int 1
%int2_3146 = torch.constant.int 2
%2838 = torch.aten.transpose.int %2835, %int1_3145, %int2_3146 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %2838, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_3147 = torch.constant.float 0.000000e+00
%false_3148 = torch.constant.bool false
%none_3149 = torch.constant.none
%2839:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%2836, %2837, %2838, %float0.000000e00_3147, %false_3148, %320, %none_3149) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %2839#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3150 = torch.constant.int 1
%int2_3151 = torch.constant.int 2
%2840 = torch.aten.transpose.int %2839#0, %int1_3150, %int2_3151 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2840, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3152 = torch.constant.int 4
%int4096_3153 = torch.constant.int 4096
%2841 = torch.prim.ListConstruct %int4_3152, %2738, %int4096_3153 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2842 = torch.aten.view %2840, %2841 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2842, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3154 = torch.constant.int -2
%int-1_3155 = torch.constant.int -1
%2843 = torch.aten.transpose.int %113, %int-2_3154, %int-1_3155 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3156 = torch.constant.int 4
%2844 = torch.aten.mul.int %int4_3156, %2738 : !torch.int, !torch.int -> !torch.int
%int4096_3157 = torch.constant.int 4096
%2845 = torch.prim.ListConstruct %2844, %int4096_3157 : (!torch.int, !torch.int) -> !torch.list<int>
%2846 = torch.aten.view %2842, %2845 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2846, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2847 = torch.aten.mm %2846, %2843 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2847, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3158 = torch.constant.int 4
%int4096_3159 = torch.constant.int 4096
%2848 = torch.prim.ListConstruct %int4_3158, %2738, %int4096_3159 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2849 = torch.aten.view %2847, %2848 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2849, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3160 = torch.constant.int 1
%2850 = torch.aten.add.Tensor %2687, %2849, %int1_3160 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2850, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3161 = torch.constant.int 6
%2851 = torch.prims.convert_element_type %2850, %int6_3161 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2851, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3162 = torch.constant.int 2
%2852 = torch.aten.pow.Tensor_Scalar %2851, %int2_3162 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2852, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3163 = torch.constant.int -1
%2853 = torch.prim.ListConstruct %int-1_3163 : (!torch.int) -> !torch.list<int>
%true_3164 = torch.constant.bool true
%none_3165 = torch.constant.none
%2854 = torch.aten.mean.dim %2852, %2853, %true_3164, %none_3165 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2854, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3166 = torch.constant.float 9.9999997473787516E-6
%int1_3167 = torch.constant.int 1
%2855 = torch.aten.add.Scalar %2854, %float9.999990e-06_3166, %int1_3167 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2855, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2856 = torch.aten.rsqrt %2855 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2856, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2857 = torch.aten.mul.Tensor %2851, %2856 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2857, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2858 = torch.aten.mul.Tensor %114, %2857 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2858, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3168 = torch.constant.int 5
%2859 = torch.prims.convert_element_type %2858, %int5_3168 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2859, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3169 = torch.constant.int -2
%int-1_3170 = torch.constant.int -1
%2860 = torch.aten.transpose.int %115, %int-2_3169, %int-1_3170 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3171 = torch.constant.int 4
%2861 = torch.aten.mul.int %int4_3171, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3172 = torch.constant.int 4096
%2862 = torch.prim.ListConstruct %2861, %int4096_3172 : (!torch.int, !torch.int) -> !torch.list<int>
%2863 = torch.aten.view %2859, %2862 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2863, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2864 = torch.aten.mm %2863, %2860 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2864, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3173 = torch.constant.int 4
%int14336_3174 = torch.constant.int 14336
%2865 = torch.prim.ListConstruct %int4_3173, %294, %int14336_3174 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2866 = torch.aten.view %2864, %2865 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2866, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2867 = torch.aten.silu %2866 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2867, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3175 = torch.constant.int -2
%int-1_3176 = torch.constant.int -1
%2868 = torch.aten.transpose.int %116, %int-2_3175, %int-1_3176 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3177 = torch.constant.int 4
%2869 = torch.aten.mul.int %int4_3177, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3178 = torch.constant.int 4096
%2870 = torch.prim.ListConstruct %2869, %int4096_3178 : (!torch.int, !torch.int) -> !torch.list<int>
%2871 = torch.aten.view %2859, %2870 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2871, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2872 = torch.aten.mm %2871, %2868 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2872, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3179 = torch.constant.int 4
%int14336_3180 = torch.constant.int 14336
%2873 = torch.prim.ListConstruct %int4_3179, %294, %int14336_3180 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2874 = torch.aten.view %2872, %2873 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2874, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%2875 = torch.aten.mul.Tensor %2867, %2874 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %2875, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3181 = torch.constant.int -2
%int-1_3182 = torch.constant.int -1
%2876 = torch.aten.transpose.int %117, %int-2_3181, %int-1_3182 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_3183 = torch.constant.int 1
%2877 = torch.aten.size.int %2866, %int1_3183 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_3184 = torch.constant.int 4
%2878 = torch.aten.mul.int %int4_3184, %2877 : !torch.int, !torch.int -> !torch.int
%int14336_3185 = torch.constant.int 14336
%2879 = torch.prim.ListConstruct %2878, %int14336_3185 : (!torch.int, !torch.int) -> !torch.list<int>
%2880 = torch.aten.view %2875, %2879 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %2880, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%2881 = torch.aten.mm %2880, %2876 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2881, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3186 = torch.constant.int 4
%int4096_3187 = torch.constant.int 4096
%2882 = torch.prim.ListConstruct %int4_3186, %2877, %int4096_3187 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2883 = torch.aten.view %2881, %2882 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2883, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3188 = torch.constant.int 1
%2884 = torch.aten.add.Tensor %2850, %2883, %int1_3188 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2884, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3189 = torch.constant.int 6
%2885 = torch.prims.convert_element_type %2884, %int6_3189 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2885, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3190 = torch.constant.int 2
%2886 = torch.aten.pow.Tensor_Scalar %2885, %int2_3190 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2886, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3191 = torch.constant.int -1
%2887 = torch.prim.ListConstruct %int-1_3191 : (!torch.int) -> !torch.list<int>
%true_3192 = torch.constant.bool true
%none_3193 = torch.constant.none
%2888 = torch.aten.mean.dim %2886, %2887, %true_3192, %none_3193 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2888, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3194 = torch.constant.float 9.9999997473787516E-6
%int1_3195 = torch.constant.int 1
%2889 = torch.aten.add.Scalar %2888, %float9.999990e-06_3194, %int1_3195 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2889, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2890 = torch.aten.rsqrt %2889 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %2890, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%2891 = torch.aten.mul.Tensor %2885, %2890 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2891, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%2892 = torch.aten.mul.Tensor %118, %2891 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %2892, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3196 = torch.constant.int 5
%2893 = torch.prims.convert_element_type %2892, %int5_3196 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2893, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3197 = torch.constant.int -2
%int-1_3198 = torch.constant.int -1
%2894 = torch.aten.transpose.int %119, %int-2_3197, %int-1_3198 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3199 = torch.constant.int 4
%2895 = torch.aten.mul.int %int4_3199, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3200 = torch.constant.int 4096
%2896 = torch.prim.ListConstruct %2895, %int4096_3200 : (!torch.int, !torch.int) -> !torch.list<int>
%2897 = torch.aten.view %2893, %2896 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2897, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2898 = torch.aten.mm %2897, %2894 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2898, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3201 = torch.constant.int 4
%int4096_3202 = torch.constant.int 4096
%2899 = torch.prim.ListConstruct %int4_3201, %294, %int4096_3202 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2900 = torch.aten.view %2898, %2899 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %2900, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3203 = torch.constant.int -2
%int-1_3204 = torch.constant.int -1
%2901 = torch.aten.transpose.int %120, %int-2_3203, %int-1_3204 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3205 = torch.constant.int 4
%2902 = torch.aten.mul.int %int4_3205, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3206 = torch.constant.int 4096
%2903 = torch.prim.ListConstruct %2902, %int4096_3206 : (!torch.int, !torch.int) -> !torch.list<int>
%2904 = torch.aten.view %2893, %2903 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2904, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2905 = torch.aten.mm %2904, %2901 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2905, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3207 = torch.constant.int 4
%int1024_3208 = torch.constant.int 1024
%2906 = torch.prim.ListConstruct %int4_3207, %294, %int1024_3208 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2907 = torch.aten.view %2905, %2906 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2907, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_3209 = torch.constant.int -2
%int-1_3210 = torch.constant.int -1
%2908 = torch.aten.transpose.int %121, %int-2_3209, %int-1_3210 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3211 = torch.constant.int 4
%2909 = torch.aten.mul.int %int4_3211, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3212 = torch.constant.int 4096
%2910 = torch.prim.ListConstruct %2909, %int4096_3212 : (!torch.int, !torch.int) -> !torch.list<int>
%2911 = torch.aten.view %2893, %2910 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %2911, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%2912 = torch.aten.mm %2911, %2908 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %2912, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3213 = torch.constant.int 4
%int1024_3214 = torch.constant.int 1024
%2913 = torch.prim.ListConstruct %int4_3213, %294, %int1024_3214 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2914 = torch.aten.view %2912, %2913 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %2914, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_3215 = torch.constant.int 4
%int32_3216 = torch.constant.int 32
%int128_3217 = torch.constant.int 128
%2915 = torch.prim.ListConstruct %int4_3215, %294, %int32_3216, %int128_3217 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2916 = torch.aten.view %2900, %2915 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2916, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3218 = torch.constant.int 4
%int8_3219 = torch.constant.int 8
%int128_3220 = torch.constant.int 128
%2917 = torch.prim.ListConstruct %int4_3218, %294, %int8_3219, %int128_3220 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2918 = torch.aten.view %2907, %2917 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2918, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_3221 = torch.constant.int 4
%int8_3222 = torch.constant.int 8
%int128_3223 = torch.constant.int 128
%2919 = torch.prim.ListConstruct %int4_3221, %294, %int8_3222, %int128_3223 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2920 = torch.aten.view %2914, %2919 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2920, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_3224 = torch.constant.int 131072
%none_3225 = torch.constant.none
%none_3226 = torch.constant.none
%cpu_3227 = torch.constant.device "cpu"
%false_3228 = torch.constant.bool false
%2921 = torch.aten.arange %int131072_3224, %none_3225, %none_3226, %cpu_3227, %false_3228 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3229 = torch.constant.int 0
%int128_3230 = torch.constant.int 128
%int2_3231 = torch.constant.int 2
%none_3232 = torch.constant.none
%none_3233 = torch.constant.none
%cpu_3234 = torch.constant.device "cpu"
%false_3235 = torch.constant.bool false
%2922 = torch.aten.arange.start_step %int0_3229, %int128_3230, %int2_3231, %none_3232, %none_3233, %cpu_3234, %false_3235 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3236 = torch.constant.int 0
%int0_3237 = torch.constant.int 0
%int64_3238 = torch.constant.int 64
%int1_3239 = torch.constant.int 1
%2923 = torch.aten.slice.Tensor %2922, %int0_3236, %int0_3237, %int64_3238, %int1_3239 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3240 = torch.constant.int 6
%2924 = torch.prims.convert_element_type %2923, %int6_3240 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3241 = torch.constant.int 128
%2925 = torch.aten.div.Scalar %2924, %int128_3241 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3242 = torch.constant.float 5.000000e+05
%2926 = torch.aten.pow.Scalar %float5.000000e05_3242, %2925 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2927 = torch.aten.reciprocal %2926 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3243 = torch.constant.float 1.000000e+00
%2928 = torch.aten.mul.Scalar %2927, %float1.000000e00_3243 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3244 = torch.constant.int 131072
%int1_3245 = torch.constant.int 1
%2929 = torch.prim.ListConstruct %int131072_3244, %int1_3245 : (!torch.int, !torch.int) -> !torch.list<int>
%2930 = torch.aten.view %2921, %2929 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2931 = torch.aten.mul.Tensor %2930, %2928 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2932 = torch.aten.cos %2931 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2933 = torch.aten.sin %2931 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2934 = torch.aten.complex %2932, %2933 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3246 = torch.constant.int 1
%2935 = torch.aten.size.int %2900, %int1_3246 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_3247 = torch.constant.int 0
%2936 = torch.aten.add.int %int0_3247, %2935 : !torch.int, !torch.int -> !torch.int
%int0_3248 = torch.constant.int 0
%int0_3249 = torch.constant.int 0
%int1_3250 = torch.constant.int 1
%2937 = torch.aten.slice.Tensor %2934, %int0_3248, %int0_3249, %2936, %int1_3250 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2937, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3251 = torch.constant.int 1
%int0_3252 = torch.constant.int 0
%int9223372036854775807_3253 = torch.constant.int 9223372036854775807
%int1_3254 = torch.constant.int 1
%2938 = torch.aten.slice.Tensor %2937, %int1_3251, %int0_3252, %int9223372036854775807_3253, %int1_3254 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2938, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3255 = torch.constant.int 0
%2939 = torch.aten.unsqueeze %2938, %int0_3255 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2939, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3256 = torch.constant.int 2
%2940 = torch.aten.unsqueeze %2939, %int2_3256 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2940, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3257 = torch.constant.int 3
%int0_3258 = torch.constant.int 0
%int9223372036854775807_3259 = torch.constant.int 9223372036854775807
%int1_3260 = torch.constant.int 1
%2941 = torch.aten.slice.Tensor %2940, %int3_3257, %int0_3258, %int9223372036854775807_3259, %int1_3260 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2941, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2942 = torch_c.to_builtin_tensor %2916 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_3261 = arith.constant 1 : index
%dim_3262 = tensor.dim %2942, %c1_3261 : tensor<4x?x32x128xf16>
%2943 = flow.tensor.bitcast %2942 : tensor<4x?x32x128xf16>{%dim_3262} -> tensor<4x?x32x64xcomplex<f16>>{%dim_3262}
%2944 = torch_c.from_builtin_tensor %2943 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %2944, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%2945 = torch.aten.mul.Tensor %2944, %2941 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %2945, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%2946 = torch_c.to_builtin_tensor %2945 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_3263 = arith.constant 1 : index
%dim_3264 = tensor.dim %2946, %c1_3263 : tensor<4x?x32x64xcomplex<f32>>
%2947 = flow.tensor.bitcast %2946 : tensor<4x?x32x64xcomplex<f32>>{%dim_3264} -> tensor<4x?x32x128xf32>{%dim_3264}
%2948 = torch_c.from_builtin_tensor %2947 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %2948, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_3265 = torch.constant.int 5
%2949 = torch.prims.convert_element_type %2948, %int5_3265 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %2949, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_3266 = torch.constant.int 131072
%none_3267 = torch.constant.none
%none_3268 = torch.constant.none
%cpu_3269 = torch.constant.device "cpu"
%false_3270 = torch.constant.bool false
%2950 = torch.aten.arange %int131072_3266, %none_3267, %none_3268, %cpu_3269, %false_3270 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3271 = torch.constant.int 0
%int128_3272 = torch.constant.int 128
%int2_3273 = torch.constant.int 2
%none_3274 = torch.constant.none
%none_3275 = torch.constant.none
%cpu_3276 = torch.constant.device "cpu"
%false_3277 = torch.constant.bool false
%2951 = torch.aten.arange.start_step %int0_3271, %int128_3272, %int2_3273, %none_3274, %none_3275, %cpu_3276, %false_3277 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3278 = torch.constant.int 0
%int0_3279 = torch.constant.int 0
%int64_3280 = torch.constant.int 64
%int1_3281 = torch.constant.int 1
%2952 = torch.aten.slice.Tensor %2951, %int0_3278, %int0_3279, %int64_3280, %int1_3281 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3282 = torch.constant.int 6
%2953 = torch.prims.convert_element_type %2952, %int6_3282 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3283 = torch.constant.int 128
%2954 = torch.aten.div.Scalar %2953, %int128_3283 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3284 = torch.constant.float 5.000000e+05
%2955 = torch.aten.pow.Scalar %float5.000000e05_3284, %2954 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%2956 = torch.aten.reciprocal %2955 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3285 = torch.constant.float 1.000000e+00
%2957 = torch.aten.mul.Scalar %2956, %float1.000000e00_3285 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3286 = torch.constant.int 131072
%int1_3287 = torch.constant.int 1
%2958 = torch.prim.ListConstruct %int131072_3286, %int1_3287 : (!torch.int, !torch.int) -> !torch.list<int>
%2959 = torch.aten.view %2950, %2958 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%2960 = torch.aten.mul.Tensor %2959, %2957 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%2961 = torch.aten.cos %2960 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2962 = torch.aten.sin %2960 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%2963 = torch.aten.complex %2961, %2962 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3288 = torch.constant.int 1
%2964 = torch.aten.size.int %2907, %int1_3288 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_3289 = torch.constant.int 0
%2965 = torch.aten.add.int %int0_3289, %2964 : !torch.int, !torch.int -> !torch.int
%int0_3290 = torch.constant.int 0
%int0_3291 = torch.constant.int 0
%int1_3292 = torch.constant.int 1
%2966 = torch.aten.slice.Tensor %2963, %int0_3290, %int0_3291, %2965, %int1_3292 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2966, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3293 = torch.constant.int 1
%int0_3294 = torch.constant.int 0
%int9223372036854775807_3295 = torch.constant.int 9223372036854775807
%int1_3296 = torch.constant.int 1
%2967 = torch.aten.slice.Tensor %2966, %int1_3293, %int0_3294, %int9223372036854775807_3295, %int1_3296 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %2967, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3297 = torch.constant.int 0
%2968 = torch.aten.unsqueeze %2967, %int0_3297 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %2968, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3298 = torch.constant.int 2
%2969 = torch.aten.unsqueeze %2968, %int2_3298 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2969, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3299 = torch.constant.int 3
%int0_3300 = torch.constant.int 0
%int9223372036854775807_3301 = torch.constant.int 9223372036854775807
%int1_3302 = torch.constant.int 1
%2970 = torch.aten.slice.Tensor %2969, %int3_3299, %int0_3300, %int9223372036854775807_3301, %int1_3302 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %2970, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%2971 = torch_c.to_builtin_tensor %2918 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_3303 = arith.constant 1 : index
%dim_3304 = tensor.dim %2971, %c1_3303 : tensor<4x?x8x128xf16>
%2972 = flow.tensor.bitcast %2971 : tensor<4x?x8x128xf16>{%dim_3304} -> tensor<4x?x8x64xcomplex<f16>>{%dim_3304}
%2973 = torch_c.from_builtin_tensor %2972 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %2973, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%2974 = torch.aten.mul.Tensor %2973, %2970 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %2974, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%2975 = torch_c.to_builtin_tensor %2974 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_3305 = arith.constant 1 : index
%dim_3306 = tensor.dim %2975, %c1_3305 : tensor<4x?x8x64xcomplex<f32>>
%2976 = flow.tensor.bitcast %2975 : tensor<4x?x8x64xcomplex<f32>>{%dim_3306} -> tensor<4x?x8x128xf32>{%dim_3306}
%2977 = torch_c.from_builtin_tensor %2976 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %2977, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_3307 = torch.constant.int 5
%2978 = torch.prims.convert_element_type %2977, %int5_3307 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %2978, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_3308 = torch.constant.int 64
%2979 = torch.aten.mul.Scalar %arg2, %int64_3308 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2979, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int26 = torch.constant.int 26
%int1_3309 = torch.constant.int 1
%2980 = torch.aten.add.Scalar %2979, %int26, %int1_3309 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %2980, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3310 = torch.constant.int 4
%int32_3311 = torch.constant.int 32
%int8_3312 = torch.constant.int 8
%int128_3313 = torch.constant.int 128
%2981 = torch.prim.ListConstruct %int4_3310, %425, %int32_3311, %int8_3312, %int128_3313 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2982 = torch.aten.view %2978, %2981 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %2982, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3314 = torch.constant.int 4
%2983 = torch.aten.mul.int %int4_3314, %425 : !torch.int, !torch.int -> !torch.int
%int32_3315 = torch.constant.int 32
%int8_3316 = torch.constant.int 8
%int128_3317 = torch.constant.int 128
%2984 = torch.prim.ListConstruct %2983, %int32_3315, %int8_3316, %int128_3317 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2985 = torch.aten.view %2982, %2984 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2985, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3318 = torch.constant.int 4
%2986 = torch.aten.mul.int %int4_3318, %425 : !torch.int, !torch.int -> !torch.int
%2987 = torch.prim.ListConstruct %2986 : (!torch.int) -> !torch.list<int>
%2988 = torch.aten.view %2980, %2987 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %2988, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_3319 = torch.constant.int 32
%int2_3320 = torch.constant.int 2
%int32_3321 = torch.constant.int 32
%int8_3322 = torch.constant.int 8
%int128_3323 = torch.constant.int 128
%2989 = torch.prim.ListConstruct %416, %int32_3319, %int2_3320, %int32_3321, %int8_3322, %int128_3323 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2990 = torch.aten.view %2822, %2989 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2990, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3324 = torch.constant.int 32
%2991 = torch.aten.mul.int %416, %int32_3324 : !torch.int, !torch.int -> !torch.int
%int2_3325 = torch.constant.int 2
%2992 = torch.aten.mul.int %2991, %int2_3325 : !torch.int, !torch.int -> !torch.int
%int32_3326 = torch.constant.int 32
%int8_3327 = torch.constant.int 8
%int128_3328 = torch.constant.int 128
%2993 = torch.prim.ListConstruct %2992, %int32_3326, %int8_3327, %int128_3328 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2994 = torch.aten.view %2990, %2993 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2994, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%2995 = torch.prim.ListConstruct %2988 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3329 = torch.constant.bool false
%2996 = torch.aten.index_put %2994, %2995, %2985, %false_3329 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %2996, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3330 = torch.constant.int 32
%int2_3331 = torch.constant.int 2
%int32_3332 = torch.constant.int 32
%int8_3333 = torch.constant.int 8
%int128_3334 = torch.constant.int 128
%2997 = torch.prim.ListConstruct %416, %int32_3330, %int2_3331, %int32_3332, %int8_3333, %int128_3334 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%2998 = torch.aten.view %2996, %2997 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %2998, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3335 = torch.constant.int 2097152
%2999 = torch.prim.ListConstruct %416, %int2097152_3335 : (!torch.int, !torch.int) -> !torch.list<int>
%3000 = torch.aten.view %2998, %2999 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3000, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_3336 = torch.constant.int 32
%int2_3337 = torch.constant.int 2
%int32_3338 = torch.constant.int 32
%int8_3339 = torch.constant.int 8
%int128_3340 = torch.constant.int 128
%3001 = torch.prim.ListConstruct %416, %int32_3336, %int2_3337, %int32_3338, %int8_3339, %int128_3340 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3002 = torch.aten.view %3000, %3001 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3002, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3341 = torch.constant.int 32
%int8_3342 = torch.constant.int 8
%int128_3343 = torch.constant.int 128
%3003 = torch.prim.ListConstruct %2992, %int32_3341, %int8_3342, %int128_3343 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3004 = torch.aten.view %3002, %3003 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3004, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3344 = torch.constant.int 4
%int32_3345 = torch.constant.int 32
%int8_3346 = torch.constant.int 8
%int128_3347 = torch.constant.int 128
%3005 = torch.prim.ListConstruct %int4_3344, %425, %int32_3345, %int8_3346, %int128_3347 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3006 = torch.aten.view %2920, %3005 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3006, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3348 = torch.constant.int 4
%3007 = torch.aten.mul.int %int4_3348, %425 : !torch.int, !torch.int -> !torch.int
%int32_3349 = torch.constant.int 32
%int8_3350 = torch.constant.int 8
%int128_3351 = torch.constant.int 128
%3008 = torch.prim.ListConstruct %3007, %int32_3349, %int8_3350, %int128_3351 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3009 = torch.aten.view %3006, %3008 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3009, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_3352 = torch.constant.int 1
%int1_3353 = torch.constant.int 1
%3010 = torch.aten.add.Scalar %2980, %int1_3352, %int1_3353 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3010, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3354 = torch.constant.int 4
%3011 = torch.aten.mul.int %int4_3354, %425 : !torch.int, !torch.int -> !torch.int
%3012 = torch.prim.ListConstruct %3011 : (!torch.int) -> !torch.list<int>
%3013 = torch.aten.view %3010, %3012 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3013, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3014 = torch.prim.ListConstruct %3013 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3355 = torch.constant.bool false
%3015 = torch.aten.index_put %3004, %3014, %3009, %false_3355 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3015, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3356 = torch.constant.int 32
%int2_3357 = torch.constant.int 2
%int32_3358 = torch.constant.int 32
%int8_3359 = torch.constant.int 8
%int128_3360 = torch.constant.int 128
%3016 = torch.prim.ListConstruct %416, %int32_3356, %int2_3357, %int32_3358, %int8_3359, %int128_3360 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3017 = torch.aten.view %3015, %3016 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3017, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3361 = torch.constant.int 2097152
%3018 = torch.prim.ListConstruct %416, %int2097152_3361 : (!torch.int, !torch.int) -> !torch.list<int>
%3019 = torch.aten.view %3017, %3018 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3019, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_3362 = torch.constant.int -2
%3020 = torch.aten.unsqueeze %2978, %int-2_3362 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3020, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_3363 = torch.constant.int 4
%int8_3364 = torch.constant.int 8
%int4_3365 = torch.constant.int 4
%int128_3366 = torch.constant.int 128
%3021 = torch.prim.ListConstruct %int4_3363, %2964, %int8_3364, %int4_3365, %int128_3366 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3367 = torch.constant.bool false
%3022 = torch.aten.expand %3020, %3021, %false_3367 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3022, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3368 = torch.constant.int 0
%3023 = torch.aten.clone %3022, %int0_3368 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3023, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3369 = torch.constant.int 4
%int32_3370 = torch.constant.int 32
%int128_3371 = torch.constant.int 128
%3024 = torch.prim.ListConstruct %int4_3369, %2964, %int32_3370, %int128_3371 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3025 = torch.aten._unsafe_view %3023, %3024 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3025, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_3372 = torch.constant.int -2
%3026 = torch.aten.unsqueeze %2920, %int-2_3372 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3026, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_3373 = torch.constant.int 1
%3027 = torch.aten.size.int %2914, %int1_3373 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_3374 = torch.constant.int 4
%int8_3375 = torch.constant.int 8
%int4_3376 = torch.constant.int 4
%int128_3377 = torch.constant.int 128
%3028 = torch.prim.ListConstruct %int4_3374, %3027, %int8_3375, %int4_3376, %int128_3377 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3378 = torch.constant.bool false
%3029 = torch.aten.expand %3026, %3028, %false_3378 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3029, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3379 = torch.constant.int 0
%3030 = torch.aten.clone %3029, %int0_3379 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3030, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3380 = torch.constant.int 4
%int32_3381 = torch.constant.int 32
%int128_3382 = torch.constant.int 128
%3031 = torch.prim.ListConstruct %int4_3380, %3027, %int32_3381, %int128_3382 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3032 = torch.aten._unsafe_view %3030, %3031 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3032, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_3383 = torch.constant.int 1
%int2_3384 = torch.constant.int 2
%3033 = torch.aten.transpose.int %2949, %int1_3383, %int2_3384 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3033, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3385 = torch.constant.int 1
%int2_3386 = torch.constant.int 2
%3034 = torch.aten.transpose.int %3025, %int1_3385, %int2_3386 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3034, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3387 = torch.constant.int 1
%int2_3388 = torch.constant.int 2
%3035 = torch.aten.transpose.int %3032, %int1_3387, %int2_3388 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3035, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_3389 = torch.constant.float 0.000000e+00
%false_3390 = torch.constant.bool false
%none_3391 = torch.constant.none
%3036:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%3033, %3034, %3035, %float0.000000e00_3389, %false_3390, %320, %none_3391) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %3036#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3392 = torch.constant.int 1
%int2_3393 = torch.constant.int 2
%3037 = torch.aten.transpose.int %3036#0, %int1_3392, %int2_3393 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3037, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3394 = torch.constant.int 4
%int4096_3395 = torch.constant.int 4096
%3038 = torch.prim.ListConstruct %int4_3394, %2935, %int4096_3395 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3039 = torch.aten.view %3037, %3038 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3039, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3396 = torch.constant.int -2
%int-1_3397 = torch.constant.int -1
%3040 = torch.aten.transpose.int %122, %int-2_3396, %int-1_3397 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3398 = torch.constant.int 4
%3041 = torch.aten.mul.int %int4_3398, %2935 : !torch.int, !torch.int -> !torch.int
%int4096_3399 = torch.constant.int 4096
%3042 = torch.prim.ListConstruct %3041, %int4096_3399 : (!torch.int, !torch.int) -> !torch.list<int>
%3043 = torch.aten.view %3039, %3042 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3043, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3044 = torch.aten.mm %3043, %3040 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3044, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3400 = torch.constant.int 4
%int4096_3401 = torch.constant.int 4096
%3045 = torch.prim.ListConstruct %int4_3400, %2935, %int4096_3401 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3046 = torch.aten.view %3044, %3045 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3046, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3402 = torch.constant.int 1
%3047 = torch.aten.add.Tensor %2884, %3046, %int1_3402 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3047, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3403 = torch.constant.int 6
%3048 = torch.prims.convert_element_type %3047, %int6_3403 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3048, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3404 = torch.constant.int 2
%3049 = torch.aten.pow.Tensor_Scalar %3048, %int2_3404 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3049, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3405 = torch.constant.int -1
%3050 = torch.prim.ListConstruct %int-1_3405 : (!torch.int) -> !torch.list<int>
%true_3406 = torch.constant.bool true
%none_3407 = torch.constant.none
%3051 = torch.aten.mean.dim %3049, %3050, %true_3406, %none_3407 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3051, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3408 = torch.constant.float 9.9999997473787516E-6
%int1_3409 = torch.constant.int 1
%3052 = torch.aten.add.Scalar %3051, %float9.999990e-06_3408, %int1_3409 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3052, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3053 = torch.aten.rsqrt %3052 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3053, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3054 = torch.aten.mul.Tensor %3048, %3053 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3054, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3055 = torch.aten.mul.Tensor %123, %3054 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3055, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3410 = torch.constant.int 5
%3056 = torch.prims.convert_element_type %3055, %int5_3410 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3056, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3411 = torch.constant.int -2
%int-1_3412 = torch.constant.int -1
%3057 = torch.aten.transpose.int %124, %int-2_3411, %int-1_3412 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3413 = torch.constant.int 4
%3058 = torch.aten.mul.int %int4_3413, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3414 = torch.constant.int 4096
%3059 = torch.prim.ListConstruct %3058, %int4096_3414 : (!torch.int, !torch.int) -> !torch.list<int>
%3060 = torch.aten.view %3056, %3059 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3060, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3061 = torch.aten.mm %3060, %3057 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3061, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3415 = torch.constant.int 4
%int14336_3416 = torch.constant.int 14336
%3062 = torch.prim.ListConstruct %int4_3415, %294, %int14336_3416 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3063 = torch.aten.view %3061, %3062 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3063, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3064 = torch.aten.silu %3063 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3064, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3417 = torch.constant.int -2
%int-1_3418 = torch.constant.int -1
%3065 = torch.aten.transpose.int %125, %int-2_3417, %int-1_3418 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3419 = torch.constant.int 4
%3066 = torch.aten.mul.int %int4_3419, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3420 = torch.constant.int 4096
%3067 = torch.prim.ListConstruct %3066, %int4096_3420 : (!torch.int, !torch.int) -> !torch.list<int>
%3068 = torch.aten.view %3056, %3067 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3068, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3069 = torch.aten.mm %3068, %3065 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3069, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3421 = torch.constant.int 4
%int14336_3422 = torch.constant.int 14336
%3070 = torch.prim.ListConstruct %int4_3421, %294, %int14336_3422 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3071 = torch.aten.view %3069, %3070 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3071, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3072 = torch.aten.mul.Tensor %3064, %3071 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3072, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3423 = torch.constant.int -2
%int-1_3424 = torch.constant.int -1
%3073 = torch.aten.transpose.int %126, %int-2_3423, %int-1_3424 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_3425 = torch.constant.int 1
%3074 = torch.aten.size.int %3063, %int1_3425 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_3426 = torch.constant.int 4
%3075 = torch.aten.mul.int %int4_3426, %3074 : !torch.int, !torch.int -> !torch.int
%int14336_3427 = torch.constant.int 14336
%3076 = torch.prim.ListConstruct %3075, %int14336_3427 : (!torch.int, !torch.int) -> !torch.list<int>
%3077 = torch.aten.view %3072, %3076 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3077, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%3078 = torch.aten.mm %3077, %3073 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3078, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3428 = torch.constant.int 4
%int4096_3429 = torch.constant.int 4096
%3079 = torch.prim.ListConstruct %int4_3428, %3074, %int4096_3429 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3080 = torch.aten.view %3078, %3079 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3080, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3430 = torch.constant.int 1
%3081 = torch.aten.add.Tensor %3047, %3080, %int1_3430 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3081, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3431 = torch.constant.int 6
%3082 = torch.prims.convert_element_type %3081, %int6_3431 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3082, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3432 = torch.constant.int 2
%3083 = torch.aten.pow.Tensor_Scalar %3082, %int2_3432 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3083, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3433 = torch.constant.int -1
%3084 = torch.prim.ListConstruct %int-1_3433 : (!torch.int) -> !torch.list<int>
%true_3434 = torch.constant.bool true
%none_3435 = torch.constant.none
%3085 = torch.aten.mean.dim %3083, %3084, %true_3434, %none_3435 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3085, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3436 = torch.constant.float 9.9999997473787516E-6
%int1_3437 = torch.constant.int 1
%3086 = torch.aten.add.Scalar %3085, %float9.999990e-06_3436, %int1_3437 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3086, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3087 = torch.aten.rsqrt %3086 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3087, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3088 = torch.aten.mul.Tensor %3082, %3087 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3088, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3089 = torch.aten.mul.Tensor %127, %3088 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3089, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3438 = torch.constant.int 5
%3090 = torch.prims.convert_element_type %3089, %int5_3438 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3090, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3439 = torch.constant.int -2
%int-1_3440 = torch.constant.int -1
%3091 = torch.aten.transpose.int %128, %int-2_3439, %int-1_3440 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3441 = torch.constant.int 4
%3092 = torch.aten.mul.int %int4_3441, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3442 = torch.constant.int 4096
%3093 = torch.prim.ListConstruct %3092, %int4096_3442 : (!torch.int, !torch.int) -> !torch.list<int>
%3094 = torch.aten.view %3090, %3093 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3094, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3095 = torch.aten.mm %3094, %3091 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3095, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3443 = torch.constant.int 4
%int4096_3444 = torch.constant.int 4096
%3096 = torch.prim.ListConstruct %int4_3443, %294, %int4096_3444 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3097 = torch.aten.view %3095, %3096 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3097, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3445 = torch.constant.int -2
%int-1_3446 = torch.constant.int -1
%3098 = torch.aten.transpose.int %129, %int-2_3445, %int-1_3446 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3447 = torch.constant.int 4
%3099 = torch.aten.mul.int %int4_3447, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3448 = torch.constant.int 4096
%3100 = torch.prim.ListConstruct %3099, %int4096_3448 : (!torch.int, !torch.int) -> !torch.list<int>
%3101 = torch.aten.view %3090, %3100 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3101, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3102 = torch.aten.mm %3101, %3098 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3102, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3449 = torch.constant.int 4
%int1024_3450 = torch.constant.int 1024
%3103 = torch.prim.ListConstruct %int4_3449, %294, %int1024_3450 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3104 = torch.aten.view %3102, %3103 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3104, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_3451 = torch.constant.int -2
%int-1_3452 = torch.constant.int -1
%3105 = torch.aten.transpose.int %130, %int-2_3451, %int-1_3452 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3453 = torch.constant.int 4
%3106 = torch.aten.mul.int %int4_3453, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3454 = torch.constant.int 4096
%3107 = torch.prim.ListConstruct %3106, %int4096_3454 : (!torch.int, !torch.int) -> !torch.list<int>
%3108 = torch.aten.view %3090, %3107 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3108, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3109 = torch.aten.mm %3108, %3105 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3109, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3455 = torch.constant.int 4
%int1024_3456 = torch.constant.int 1024
%3110 = torch.prim.ListConstruct %int4_3455, %294, %int1024_3456 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3111 = torch.aten.view %3109, %3110 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3111, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_3457 = torch.constant.int 4
%int32_3458 = torch.constant.int 32
%int128_3459 = torch.constant.int 128
%3112 = torch.prim.ListConstruct %int4_3457, %294, %int32_3458, %int128_3459 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3113 = torch.aten.view %3097, %3112 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3113, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3460 = torch.constant.int 4
%int8_3461 = torch.constant.int 8
%int128_3462 = torch.constant.int 128
%3114 = torch.prim.ListConstruct %int4_3460, %294, %int8_3461, %int128_3462 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3115 = torch.aten.view %3104, %3114 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3115, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_3463 = torch.constant.int 4
%int8_3464 = torch.constant.int 8
%int128_3465 = torch.constant.int 128
%3116 = torch.prim.ListConstruct %int4_3463, %294, %int8_3464, %int128_3465 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3117 = torch.aten.view %3111, %3116 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3117, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_3466 = torch.constant.int 131072
%none_3467 = torch.constant.none
%none_3468 = torch.constant.none
%cpu_3469 = torch.constant.device "cpu"
%false_3470 = torch.constant.bool false
%3118 = torch.aten.arange %int131072_3466, %none_3467, %none_3468, %cpu_3469, %false_3470 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3471 = torch.constant.int 0
%int128_3472 = torch.constant.int 128
%int2_3473 = torch.constant.int 2
%none_3474 = torch.constant.none
%none_3475 = torch.constant.none
%cpu_3476 = torch.constant.device "cpu"
%false_3477 = torch.constant.bool false
%3119 = torch.aten.arange.start_step %int0_3471, %int128_3472, %int2_3473, %none_3474, %none_3475, %cpu_3476, %false_3477 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3478 = torch.constant.int 0
%int0_3479 = torch.constant.int 0
%int64_3480 = torch.constant.int 64
%int1_3481 = torch.constant.int 1
%3120 = torch.aten.slice.Tensor %3119, %int0_3478, %int0_3479, %int64_3480, %int1_3481 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3482 = torch.constant.int 6
%3121 = torch.prims.convert_element_type %3120, %int6_3482 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3483 = torch.constant.int 128
%3122 = torch.aten.div.Scalar %3121, %int128_3483 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3484 = torch.constant.float 5.000000e+05
%3123 = torch.aten.pow.Scalar %float5.000000e05_3484, %3122 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3124 = torch.aten.reciprocal %3123 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3485 = torch.constant.float 1.000000e+00
%3125 = torch.aten.mul.Scalar %3124, %float1.000000e00_3485 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3486 = torch.constant.int 131072
%int1_3487 = torch.constant.int 1
%3126 = torch.prim.ListConstruct %int131072_3486, %int1_3487 : (!torch.int, !torch.int) -> !torch.list<int>
%3127 = torch.aten.view %3118, %3126 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3128 = torch.aten.mul.Tensor %3127, %3125 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3129 = torch.aten.cos %3128 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3130 = torch.aten.sin %3128 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3131 = torch.aten.complex %3129, %3130 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3488 = torch.constant.int 1
%3132 = torch.aten.size.int %3097, %int1_3488 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_3489 = torch.constant.int 0
%3133 = torch.aten.add.int %int0_3489, %3132 : !torch.int, !torch.int -> !torch.int
%int0_3490 = torch.constant.int 0
%int0_3491 = torch.constant.int 0
%int1_3492 = torch.constant.int 1
%3134 = torch.aten.slice.Tensor %3131, %int0_3490, %int0_3491, %3133, %int1_3492 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3134, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3493 = torch.constant.int 1
%int0_3494 = torch.constant.int 0
%int9223372036854775807_3495 = torch.constant.int 9223372036854775807
%int1_3496 = torch.constant.int 1
%3135 = torch.aten.slice.Tensor %3134, %int1_3493, %int0_3494, %int9223372036854775807_3495, %int1_3496 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3135, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3497 = torch.constant.int 0
%3136 = torch.aten.unsqueeze %3135, %int0_3497 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3136, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3498 = torch.constant.int 2
%3137 = torch.aten.unsqueeze %3136, %int2_3498 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3137, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3499 = torch.constant.int 3
%int0_3500 = torch.constant.int 0
%int9223372036854775807_3501 = torch.constant.int 9223372036854775807
%int1_3502 = torch.constant.int 1
%3138 = torch.aten.slice.Tensor %3137, %int3_3499, %int0_3500, %int9223372036854775807_3501, %int1_3502 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3138, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3139 = torch_c.to_builtin_tensor %3113 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_3503 = arith.constant 1 : index
%dim_3504 = tensor.dim %3139, %c1_3503 : tensor<4x?x32x128xf16>
%3140 = flow.tensor.bitcast %3139 : tensor<4x?x32x128xf16>{%dim_3504} -> tensor<4x?x32x64xcomplex<f16>>{%dim_3504}
%3141 = torch_c.from_builtin_tensor %3140 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %3141, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%3142 = torch.aten.mul.Tensor %3141, %3138 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %3142, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%3143 = torch_c.to_builtin_tensor %3142 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_3505 = arith.constant 1 : index
%dim_3506 = tensor.dim %3143, %c1_3505 : tensor<4x?x32x64xcomplex<f32>>
%3144 = flow.tensor.bitcast %3143 : tensor<4x?x32x64xcomplex<f32>>{%dim_3506} -> tensor<4x?x32x128xf32>{%dim_3506}
%3145 = torch_c.from_builtin_tensor %3144 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %3145, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_3507 = torch.constant.int 5
%3146 = torch.prims.convert_element_type %3145, %int5_3507 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3146, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_3508 = torch.constant.int 131072
%none_3509 = torch.constant.none
%none_3510 = torch.constant.none
%cpu_3511 = torch.constant.device "cpu"
%false_3512 = torch.constant.bool false
%3147 = torch.aten.arange %int131072_3508, %none_3509, %none_3510, %cpu_3511, %false_3512 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3513 = torch.constant.int 0
%int128_3514 = torch.constant.int 128
%int2_3515 = torch.constant.int 2
%none_3516 = torch.constant.none
%none_3517 = torch.constant.none
%cpu_3518 = torch.constant.device "cpu"
%false_3519 = torch.constant.bool false
%3148 = torch.aten.arange.start_step %int0_3513, %int128_3514, %int2_3515, %none_3516, %none_3517, %cpu_3518, %false_3519 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3520 = torch.constant.int 0
%int0_3521 = torch.constant.int 0
%int64_3522 = torch.constant.int 64
%int1_3523 = torch.constant.int 1
%3149 = torch.aten.slice.Tensor %3148, %int0_3520, %int0_3521, %int64_3522, %int1_3523 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3524 = torch.constant.int 6
%3150 = torch.prims.convert_element_type %3149, %int6_3524 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3525 = torch.constant.int 128
%3151 = torch.aten.div.Scalar %3150, %int128_3525 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3526 = torch.constant.float 5.000000e+05
%3152 = torch.aten.pow.Scalar %float5.000000e05_3526, %3151 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3153 = torch.aten.reciprocal %3152 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3527 = torch.constant.float 1.000000e+00
%3154 = torch.aten.mul.Scalar %3153, %float1.000000e00_3527 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3528 = torch.constant.int 131072
%int1_3529 = torch.constant.int 1
%3155 = torch.prim.ListConstruct %int131072_3528, %int1_3529 : (!torch.int, !torch.int) -> !torch.list<int>
%3156 = torch.aten.view %3147, %3155 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3157 = torch.aten.mul.Tensor %3156, %3154 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3158 = torch.aten.cos %3157 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3159 = torch.aten.sin %3157 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3160 = torch.aten.complex %3158, %3159 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3530 = torch.constant.int 1
%3161 = torch.aten.size.int %3104, %int1_3530 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_3531 = torch.constant.int 0
%3162 = torch.aten.add.int %int0_3531, %3161 : !torch.int, !torch.int -> !torch.int
%int0_3532 = torch.constant.int 0
%int0_3533 = torch.constant.int 0
%int1_3534 = torch.constant.int 1
%3163 = torch.aten.slice.Tensor %3160, %int0_3532, %int0_3533, %3162, %int1_3534 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3163, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3535 = torch.constant.int 1
%int0_3536 = torch.constant.int 0
%int9223372036854775807_3537 = torch.constant.int 9223372036854775807
%int1_3538 = torch.constant.int 1
%3164 = torch.aten.slice.Tensor %3163, %int1_3535, %int0_3536, %int9223372036854775807_3537, %int1_3538 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3164, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3539 = torch.constant.int 0
%3165 = torch.aten.unsqueeze %3164, %int0_3539 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3165, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3540 = torch.constant.int 2
%3166 = torch.aten.unsqueeze %3165, %int2_3540 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3166, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3541 = torch.constant.int 3
%int0_3542 = torch.constant.int 0
%int9223372036854775807_3543 = torch.constant.int 9223372036854775807
%int1_3544 = torch.constant.int 1
%3167 = torch.aten.slice.Tensor %3166, %int3_3541, %int0_3542, %int9223372036854775807_3543, %int1_3544 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3167, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3168 = torch_c.to_builtin_tensor %3115 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_3545 = arith.constant 1 : index
%dim_3546 = tensor.dim %3168, %c1_3545 : tensor<4x?x8x128xf16>
%3169 = flow.tensor.bitcast %3168 : tensor<4x?x8x128xf16>{%dim_3546} -> tensor<4x?x8x64xcomplex<f16>>{%dim_3546}
%3170 = torch_c.from_builtin_tensor %3169 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %3170, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%3171 = torch.aten.mul.Tensor %3170, %3167 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %3171, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%3172 = torch_c.to_builtin_tensor %3171 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_3547 = arith.constant 1 : index
%dim_3548 = tensor.dim %3172, %c1_3547 : tensor<4x?x8x64xcomplex<f32>>
%3173 = flow.tensor.bitcast %3172 : tensor<4x?x8x64xcomplex<f32>>{%dim_3548} -> tensor<4x?x8x128xf32>{%dim_3548}
%3174 = torch_c.from_builtin_tensor %3173 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %3174, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_3549 = torch.constant.int 5
%3175 = torch.prims.convert_element_type %3174, %int5_3549 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3175, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_3550 = torch.constant.int 64
%3176 = torch.aten.mul.Scalar %arg2, %int64_3550 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3176, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int28 = torch.constant.int 28
%int1_3551 = torch.constant.int 1
%3177 = torch.aten.add.Scalar %3176, %int28, %int1_3551 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3177, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3552 = torch.constant.int 4
%int32_3553 = torch.constant.int 32
%int8_3554 = torch.constant.int 8
%int128_3555 = torch.constant.int 128
%3178 = torch.prim.ListConstruct %int4_3552, %425, %int32_3553, %int8_3554, %int128_3555 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3179 = torch.aten.view %3175, %3178 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3179, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3556 = torch.constant.int 4
%3180 = torch.aten.mul.int %int4_3556, %425 : !torch.int, !torch.int -> !torch.int
%int32_3557 = torch.constant.int 32
%int8_3558 = torch.constant.int 8
%int128_3559 = torch.constant.int 128
%3181 = torch.prim.ListConstruct %3180, %int32_3557, %int8_3558, %int128_3559 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3182 = torch.aten.view %3179, %3181 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3182, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3560 = torch.constant.int 4
%3183 = torch.aten.mul.int %int4_3560, %425 : !torch.int, !torch.int -> !torch.int
%3184 = torch.prim.ListConstruct %3183 : (!torch.int) -> !torch.list<int>
%3185 = torch.aten.view %3177, %3184 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3185, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_3561 = torch.constant.int 32
%int2_3562 = torch.constant.int 2
%int32_3563 = torch.constant.int 32
%int8_3564 = torch.constant.int 8
%int128_3565 = torch.constant.int 128
%3186 = torch.prim.ListConstruct %416, %int32_3561, %int2_3562, %int32_3563, %int8_3564, %int128_3565 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3187 = torch.aten.view %3019, %3186 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3187, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3566 = torch.constant.int 32
%3188 = torch.aten.mul.int %416, %int32_3566 : !torch.int, !torch.int -> !torch.int
%int2_3567 = torch.constant.int 2
%3189 = torch.aten.mul.int %3188, %int2_3567 : !torch.int, !torch.int -> !torch.int
%int32_3568 = torch.constant.int 32
%int8_3569 = torch.constant.int 8
%int128_3570 = torch.constant.int 128
%3190 = torch.prim.ListConstruct %3189, %int32_3568, %int8_3569, %int128_3570 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3191 = torch.aten.view %3187, %3190 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3191, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%3192 = torch.prim.ListConstruct %3185 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3571 = torch.constant.bool false
%3193 = torch.aten.index_put %3191, %3192, %3182, %false_3571 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3193, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3572 = torch.constant.int 32
%int2_3573 = torch.constant.int 2
%int32_3574 = torch.constant.int 32
%int8_3575 = torch.constant.int 8
%int128_3576 = torch.constant.int 128
%3194 = torch.prim.ListConstruct %416, %int32_3572, %int2_3573, %int32_3574, %int8_3575, %int128_3576 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3195 = torch.aten.view %3193, %3194 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3195, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3577 = torch.constant.int 2097152
%3196 = torch.prim.ListConstruct %416, %int2097152_3577 : (!torch.int, !torch.int) -> !torch.list<int>
%3197 = torch.aten.view %3195, %3196 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3197, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_3578 = torch.constant.int 32
%int2_3579 = torch.constant.int 2
%int32_3580 = torch.constant.int 32
%int8_3581 = torch.constant.int 8
%int128_3582 = torch.constant.int 128
%3198 = torch.prim.ListConstruct %416, %int32_3578, %int2_3579, %int32_3580, %int8_3581, %int128_3582 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3199 = torch.aten.view %3197, %3198 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3199, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3583 = torch.constant.int 32
%int8_3584 = torch.constant.int 8
%int128_3585 = torch.constant.int 128
%3200 = torch.prim.ListConstruct %3189, %int32_3583, %int8_3584, %int128_3585 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3201 = torch.aten.view %3199, %3200 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3201, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3586 = torch.constant.int 4
%int32_3587 = torch.constant.int 32
%int8_3588 = torch.constant.int 8
%int128_3589 = torch.constant.int 128
%3202 = torch.prim.ListConstruct %int4_3586, %425, %int32_3587, %int8_3588, %int128_3589 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3203 = torch.aten.view %3117, %3202 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3203, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3590 = torch.constant.int 4
%3204 = torch.aten.mul.int %int4_3590, %425 : !torch.int, !torch.int -> !torch.int
%int32_3591 = torch.constant.int 32
%int8_3592 = torch.constant.int 8
%int128_3593 = torch.constant.int 128
%3205 = torch.prim.ListConstruct %3204, %int32_3591, %int8_3592, %int128_3593 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3206 = torch.aten.view %3203, %3205 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3206, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_3594 = torch.constant.int 1
%int1_3595 = torch.constant.int 1
%3207 = torch.aten.add.Scalar %3177, %int1_3594, %int1_3595 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3207, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3596 = torch.constant.int 4
%3208 = torch.aten.mul.int %int4_3596, %425 : !torch.int, !torch.int -> !torch.int
%3209 = torch.prim.ListConstruct %3208 : (!torch.int) -> !torch.list<int>
%3210 = torch.aten.view %3207, %3209 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3210, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3211 = torch.prim.ListConstruct %3210 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3597 = torch.constant.bool false
%3212 = torch.aten.index_put %3201, %3211, %3206, %false_3597 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3212, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3598 = torch.constant.int 32
%int2_3599 = torch.constant.int 2
%int32_3600 = torch.constant.int 32
%int8_3601 = torch.constant.int 8
%int128_3602 = torch.constant.int 128
%3213 = torch.prim.ListConstruct %416, %int32_3598, %int2_3599, %int32_3600, %int8_3601, %int128_3602 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3214 = torch.aten.view %3212, %3213 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3214, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3603 = torch.constant.int 2097152
%3215 = torch.prim.ListConstruct %416, %int2097152_3603 : (!torch.int, !torch.int) -> !torch.list<int>
%3216 = torch.aten.view %3214, %3215 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3216, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_3604 = torch.constant.int -2
%3217 = torch.aten.unsqueeze %3175, %int-2_3604 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3217, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_3605 = torch.constant.int 4
%int8_3606 = torch.constant.int 8
%int4_3607 = torch.constant.int 4
%int128_3608 = torch.constant.int 128
%3218 = torch.prim.ListConstruct %int4_3605, %3161, %int8_3606, %int4_3607, %int128_3608 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3609 = torch.constant.bool false
%3219 = torch.aten.expand %3217, %3218, %false_3609 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3219, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3610 = torch.constant.int 0
%3220 = torch.aten.clone %3219, %int0_3610 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3220, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3611 = torch.constant.int 4
%int32_3612 = torch.constant.int 32
%int128_3613 = torch.constant.int 128
%3221 = torch.prim.ListConstruct %int4_3611, %3161, %int32_3612, %int128_3613 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3222 = torch.aten._unsafe_view %3220, %3221 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3222, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_3614 = torch.constant.int -2
%3223 = torch.aten.unsqueeze %3117, %int-2_3614 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3223, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_3615 = torch.constant.int 1
%3224 = torch.aten.size.int %3111, %int1_3615 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_3616 = torch.constant.int 4
%int8_3617 = torch.constant.int 8
%int4_3618 = torch.constant.int 4
%int128_3619 = torch.constant.int 128
%3225 = torch.prim.ListConstruct %int4_3616, %3224, %int8_3617, %int4_3618, %int128_3619 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3620 = torch.constant.bool false
%3226 = torch.aten.expand %3223, %3225, %false_3620 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3226, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3621 = torch.constant.int 0
%3227 = torch.aten.clone %3226, %int0_3621 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3227, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3622 = torch.constant.int 4
%int32_3623 = torch.constant.int 32
%int128_3624 = torch.constant.int 128
%3228 = torch.prim.ListConstruct %int4_3622, %3224, %int32_3623, %int128_3624 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3229 = torch.aten._unsafe_view %3227, %3228 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3229, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_3625 = torch.constant.int 1
%int2_3626 = torch.constant.int 2
%3230 = torch.aten.transpose.int %3146, %int1_3625, %int2_3626 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3230, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3627 = torch.constant.int 1
%int2_3628 = torch.constant.int 2
%3231 = torch.aten.transpose.int %3222, %int1_3627, %int2_3628 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3231, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3629 = torch.constant.int 1
%int2_3630 = torch.constant.int 2
%3232 = torch.aten.transpose.int %3229, %int1_3629, %int2_3630 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3232, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_3631 = torch.constant.float 0.000000e+00
%false_3632 = torch.constant.bool false
%none_3633 = torch.constant.none
%3233:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%3230, %3231, %3232, %float0.000000e00_3631, %false_3632, %320, %none_3633) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %3233#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3634 = torch.constant.int 1
%int2_3635 = torch.constant.int 2
%3234 = torch.aten.transpose.int %3233#0, %int1_3634, %int2_3635 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3234, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3636 = torch.constant.int 4
%int4096_3637 = torch.constant.int 4096
%3235 = torch.prim.ListConstruct %int4_3636, %3132, %int4096_3637 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3236 = torch.aten.view %3234, %3235 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3236, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3638 = torch.constant.int -2
%int-1_3639 = torch.constant.int -1
%3237 = torch.aten.transpose.int %131, %int-2_3638, %int-1_3639 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3640 = torch.constant.int 4
%3238 = torch.aten.mul.int %int4_3640, %3132 : !torch.int, !torch.int -> !torch.int
%int4096_3641 = torch.constant.int 4096
%3239 = torch.prim.ListConstruct %3238, %int4096_3641 : (!torch.int, !torch.int) -> !torch.list<int>
%3240 = torch.aten.view %3236, %3239 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3240, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3241 = torch.aten.mm %3240, %3237 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3241, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3642 = torch.constant.int 4
%int4096_3643 = torch.constant.int 4096
%3242 = torch.prim.ListConstruct %int4_3642, %3132, %int4096_3643 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3243 = torch.aten.view %3241, %3242 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3243, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3644 = torch.constant.int 1
%3244 = torch.aten.add.Tensor %3081, %3243, %int1_3644 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3244, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3645 = torch.constant.int 6
%3245 = torch.prims.convert_element_type %3244, %int6_3645 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3245, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3646 = torch.constant.int 2
%3246 = torch.aten.pow.Tensor_Scalar %3245, %int2_3646 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3246, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3647 = torch.constant.int -1
%3247 = torch.prim.ListConstruct %int-1_3647 : (!torch.int) -> !torch.list<int>
%true_3648 = torch.constant.bool true
%none_3649 = torch.constant.none
%3248 = torch.aten.mean.dim %3246, %3247, %true_3648, %none_3649 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3248, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3650 = torch.constant.float 9.9999997473787516E-6
%int1_3651 = torch.constant.int 1
%3249 = torch.aten.add.Scalar %3248, %float9.999990e-06_3650, %int1_3651 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3249, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3250 = torch.aten.rsqrt %3249 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3250, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3251 = torch.aten.mul.Tensor %3245, %3250 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3251, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3252 = torch.aten.mul.Tensor %132, %3251 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3252, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3652 = torch.constant.int 5
%3253 = torch.prims.convert_element_type %3252, %int5_3652 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3253, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3653 = torch.constant.int -2
%int-1_3654 = torch.constant.int -1
%3254 = torch.aten.transpose.int %133, %int-2_3653, %int-1_3654 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3655 = torch.constant.int 4
%3255 = torch.aten.mul.int %int4_3655, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3656 = torch.constant.int 4096
%3256 = torch.prim.ListConstruct %3255, %int4096_3656 : (!torch.int, !torch.int) -> !torch.list<int>
%3257 = torch.aten.view %3253, %3256 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3257, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3258 = torch.aten.mm %3257, %3254 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3258, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3657 = torch.constant.int 4
%int14336_3658 = torch.constant.int 14336
%3259 = torch.prim.ListConstruct %int4_3657, %294, %int14336_3658 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3260 = torch.aten.view %3258, %3259 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3260, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3261 = torch.aten.silu %3260 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3261, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3659 = torch.constant.int -2
%int-1_3660 = torch.constant.int -1
%3262 = torch.aten.transpose.int %134, %int-2_3659, %int-1_3660 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3661 = torch.constant.int 4
%3263 = torch.aten.mul.int %int4_3661, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3662 = torch.constant.int 4096
%3264 = torch.prim.ListConstruct %3263, %int4096_3662 : (!torch.int, !torch.int) -> !torch.list<int>
%3265 = torch.aten.view %3253, %3264 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3265, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3266 = torch.aten.mm %3265, %3262 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3266, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3663 = torch.constant.int 4
%int14336_3664 = torch.constant.int 14336
%3267 = torch.prim.ListConstruct %int4_3663, %294, %int14336_3664 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3268 = torch.aten.view %3266, %3267 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3268, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3269 = torch.aten.mul.Tensor %3261, %3268 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3269, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3665 = torch.constant.int -2
%int-1_3666 = torch.constant.int -1
%3270 = torch.aten.transpose.int %135, %int-2_3665, %int-1_3666 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_3667 = torch.constant.int 1
%3271 = torch.aten.size.int %3260, %int1_3667 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_3668 = torch.constant.int 4
%3272 = torch.aten.mul.int %int4_3668, %3271 : !torch.int, !torch.int -> !torch.int
%int14336_3669 = torch.constant.int 14336
%3273 = torch.prim.ListConstruct %3272, %int14336_3669 : (!torch.int, !torch.int) -> !torch.list<int>
%3274 = torch.aten.view %3269, %3273 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3274, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%3275 = torch.aten.mm %3274, %3270 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3275, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3670 = torch.constant.int 4
%int4096_3671 = torch.constant.int 4096
%3276 = torch.prim.ListConstruct %int4_3670, %3271, %int4096_3671 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3277 = torch.aten.view %3275, %3276 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3277, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3672 = torch.constant.int 1
%3278 = torch.aten.add.Tensor %3244, %3277, %int1_3672 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3278, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3673 = torch.constant.int 6
%3279 = torch.prims.convert_element_type %3278, %int6_3673 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3279, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3674 = torch.constant.int 2
%3280 = torch.aten.pow.Tensor_Scalar %3279, %int2_3674 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3280, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3675 = torch.constant.int -1
%3281 = torch.prim.ListConstruct %int-1_3675 : (!torch.int) -> !torch.list<int>
%true_3676 = torch.constant.bool true
%none_3677 = torch.constant.none
%3282 = torch.aten.mean.dim %3280, %3281, %true_3676, %none_3677 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3282, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3678 = torch.constant.float 9.9999997473787516E-6
%int1_3679 = torch.constant.int 1
%3283 = torch.aten.add.Scalar %3282, %float9.999990e-06_3678, %int1_3679 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3283, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3284 = torch.aten.rsqrt %3283 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3284, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3285 = torch.aten.mul.Tensor %3279, %3284 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3285, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3286 = torch.aten.mul.Tensor %136, %3285 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3286, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3680 = torch.constant.int 5
%3287 = torch.prims.convert_element_type %3286, %int5_3680 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3287, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3681 = torch.constant.int -2
%int-1_3682 = torch.constant.int -1
%3288 = torch.aten.transpose.int %137, %int-2_3681, %int-1_3682 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3683 = torch.constant.int 4
%3289 = torch.aten.mul.int %int4_3683, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3684 = torch.constant.int 4096
%3290 = torch.prim.ListConstruct %3289, %int4096_3684 : (!torch.int, !torch.int) -> !torch.list<int>
%3291 = torch.aten.view %3287, %3290 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3291, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3292 = torch.aten.mm %3291, %3288 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3292, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3685 = torch.constant.int 4
%int4096_3686 = torch.constant.int 4096
%3293 = torch.prim.ListConstruct %int4_3685, %294, %int4096_3686 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3294 = torch.aten.view %3292, %3293 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3294, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3687 = torch.constant.int -2
%int-1_3688 = torch.constant.int -1
%3295 = torch.aten.transpose.int %138, %int-2_3687, %int-1_3688 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3689 = torch.constant.int 4
%3296 = torch.aten.mul.int %int4_3689, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3690 = torch.constant.int 4096
%3297 = torch.prim.ListConstruct %3296, %int4096_3690 : (!torch.int, !torch.int) -> !torch.list<int>
%3298 = torch.aten.view %3287, %3297 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3298, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3299 = torch.aten.mm %3298, %3295 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3299, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3691 = torch.constant.int 4
%int1024_3692 = torch.constant.int 1024
%3300 = torch.prim.ListConstruct %int4_3691, %294, %int1024_3692 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3301 = torch.aten.view %3299, %3300 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3301, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_3693 = torch.constant.int -2
%int-1_3694 = torch.constant.int -1
%3302 = torch.aten.transpose.int %139, %int-2_3693, %int-1_3694 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3695 = torch.constant.int 4
%3303 = torch.aten.mul.int %int4_3695, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3696 = torch.constant.int 4096
%3304 = torch.prim.ListConstruct %3303, %int4096_3696 : (!torch.int, !torch.int) -> !torch.list<int>
%3305 = torch.aten.view %3287, %3304 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3305, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3306 = torch.aten.mm %3305, %3302 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3306, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3697 = torch.constant.int 4
%int1024_3698 = torch.constant.int 1024
%3307 = torch.prim.ListConstruct %int4_3697, %294, %int1024_3698 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3308 = torch.aten.view %3306, %3307 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3308, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_3699 = torch.constant.int 4
%int32_3700 = torch.constant.int 32
%int128_3701 = torch.constant.int 128
%3309 = torch.prim.ListConstruct %int4_3699, %294, %int32_3700, %int128_3701 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3310 = torch.aten.view %3294, %3309 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3310, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3702 = torch.constant.int 4
%int8_3703 = torch.constant.int 8
%int128_3704 = torch.constant.int 128
%3311 = torch.prim.ListConstruct %int4_3702, %294, %int8_3703, %int128_3704 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3312 = torch.aten.view %3301, %3311 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3312, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_3705 = torch.constant.int 4
%int8_3706 = torch.constant.int 8
%int128_3707 = torch.constant.int 128
%3313 = torch.prim.ListConstruct %int4_3705, %294, %int8_3706, %int128_3707 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3314 = torch.aten.view %3308, %3313 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3314, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_3708 = torch.constant.int 131072
%none_3709 = torch.constant.none
%none_3710 = torch.constant.none
%cpu_3711 = torch.constant.device "cpu"
%false_3712 = torch.constant.bool false
%3315 = torch.aten.arange %int131072_3708, %none_3709, %none_3710, %cpu_3711, %false_3712 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3713 = torch.constant.int 0
%int128_3714 = torch.constant.int 128
%int2_3715 = torch.constant.int 2
%none_3716 = torch.constant.none
%none_3717 = torch.constant.none
%cpu_3718 = torch.constant.device "cpu"
%false_3719 = torch.constant.bool false
%3316 = torch.aten.arange.start_step %int0_3713, %int128_3714, %int2_3715, %none_3716, %none_3717, %cpu_3718, %false_3719 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3720 = torch.constant.int 0
%int0_3721 = torch.constant.int 0
%int64_3722 = torch.constant.int 64
%int1_3723 = torch.constant.int 1
%3317 = torch.aten.slice.Tensor %3316, %int0_3720, %int0_3721, %int64_3722, %int1_3723 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3724 = torch.constant.int 6
%3318 = torch.prims.convert_element_type %3317, %int6_3724 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3725 = torch.constant.int 128
%3319 = torch.aten.div.Scalar %3318, %int128_3725 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3726 = torch.constant.float 5.000000e+05
%3320 = torch.aten.pow.Scalar %float5.000000e05_3726, %3319 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3321 = torch.aten.reciprocal %3320 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3727 = torch.constant.float 1.000000e+00
%3322 = torch.aten.mul.Scalar %3321, %float1.000000e00_3727 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3728 = torch.constant.int 131072
%int1_3729 = torch.constant.int 1
%3323 = torch.prim.ListConstruct %int131072_3728, %int1_3729 : (!torch.int, !torch.int) -> !torch.list<int>
%3324 = torch.aten.view %3315, %3323 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3325 = torch.aten.mul.Tensor %3324, %3322 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3326 = torch.aten.cos %3325 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3327 = torch.aten.sin %3325 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3328 = torch.aten.complex %3326, %3327 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3730 = torch.constant.int 1
%3329 = torch.aten.size.int %3294, %int1_3730 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_3731 = torch.constant.int 0
%3330 = torch.aten.add.int %int0_3731, %3329 : !torch.int, !torch.int -> !torch.int
%int0_3732 = torch.constant.int 0
%int0_3733 = torch.constant.int 0
%int1_3734 = torch.constant.int 1
%3331 = torch.aten.slice.Tensor %3328, %int0_3732, %int0_3733, %3330, %int1_3734 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3331, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3735 = torch.constant.int 1
%int0_3736 = torch.constant.int 0
%int9223372036854775807_3737 = torch.constant.int 9223372036854775807
%int1_3738 = torch.constant.int 1
%3332 = torch.aten.slice.Tensor %3331, %int1_3735, %int0_3736, %int9223372036854775807_3737, %int1_3738 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3332, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3739 = torch.constant.int 0
%3333 = torch.aten.unsqueeze %3332, %int0_3739 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3333, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3740 = torch.constant.int 2
%3334 = torch.aten.unsqueeze %3333, %int2_3740 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3334, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3741 = torch.constant.int 3
%int0_3742 = torch.constant.int 0
%int9223372036854775807_3743 = torch.constant.int 9223372036854775807
%int1_3744 = torch.constant.int 1
%3335 = torch.aten.slice.Tensor %3334, %int3_3741, %int0_3742, %int9223372036854775807_3743, %int1_3744 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3335, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3336 = torch_c.to_builtin_tensor %3310 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_3745 = arith.constant 1 : index
%dim_3746 = tensor.dim %3336, %c1_3745 : tensor<4x?x32x128xf16>
%3337 = flow.tensor.bitcast %3336 : tensor<4x?x32x128xf16>{%dim_3746} -> tensor<4x?x32x64xcomplex<f16>>{%dim_3746}
%3338 = torch_c.from_builtin_tensor %3337 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %3338, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%3339 = torch.aten.mul.Tensor %3338, %3335 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %3339, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%3340 = torch_c.to_builtin_tensor %3339 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_3747 = arith.constant 1 : index
%dim_3748 = tensor.dim %3340, %c1_3747 : tensor<4x?x32x64xcomplex<f32>>
%3341 = flow.tensor.bitcast %3340 : tensor<4x?x32x64xcomplex<f32>>{%dim_3748} -> tensor<4x?x32x128xf32>{%dim_3748}
%3342 = torch_c.from_builtin_tensor %3341 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %3342, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_3749 = torch.constant.int 5
%3343 = torch.prims.convert_element_type %3342, %int5_3749 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3343, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_3750 = torch.constant.int 131072
%none_3751 = torch.constant.none
%none_3752 = torch.constant.none
%cpu_3753 = torch.constant.device "cpu"
%false_3754 = torch.constant.bool false
%3344 = torch.aten.arange %int131072_3750, %none_3751, %none_3752, %cpu_3753, %false_3754 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3755 = torch.constant.int 0
%int128_3756 = torch.constant.int 128
%int2_3757 = torch.constant.int 2
%none_3758 = torch.constant.none
%none_3759 = torch.constant.none
%cpu_3760 = torch.constant.device "cpu"
%false_3761 = torch.constant.bool false
%3345 = torch.aten.arange.start_step %int0_3755, %int128_3756, %int2_3757, %none_3758, %none_3759, %cpu_3760, %false_3761 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3762 = torch.constant.int 0
%int0_3763 = torch.constant.int 0
%int64_3764 = torch.constant.int 64
%int1_3765 = torch.constant.int 1
%3346 = torch.aten.slice.Tensor %3345, %int0_3762, %int0_3763, %int64_3764, %int1_3765 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3766 = torch.constant.int 6
%3347 = torch.prims.convert_element_type %3346, %int6_3766 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3767 = torch.constant.int 128
%3348 = torch.aten.div.Scalar %3347, %int128_3767 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3768 = torch.constant.float 5.000000e+05
%3349 = torch.aten.pow.Scalar %float5.000000e05_3768, %3348 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3350 = torch.aten.reciprocal %3349 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3769 = torch.constant.float 1.000000e+00
%3351 = torch.aten.mul.Scalar %3350, %float1.000000e00_3769 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3770 = torch.constant.int 131072
%int1_3771 = torch.constant.int 1
%3352 = torch.prim.ListConstruct %int131072_3770, %int1_3771 : (!torch.int, !torch.int) -> !torch.list<int>
%3353 = torch.aten.view %3344, %3352 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3354 = torch.aten.mul.Tensor %3353, %3351 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3355 = torch.aten.cos %3354 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3356 = torch.aten.sin %3354 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3357 = torch.aten.complex %3355, %3356 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3772 = torch.constant.int 1
%3358 = torch.aten.size.int %3301, %int1_3772 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_3773 = torch.constant.int 0
%3359 = torch.aten.add.int %int0_3773, %3358 : !torch.int, !torch.int -> !torch.int
%int0_3774 = torch.constant.int 0
%int0_3775 = torch.constant.int 0
%int1_3776 = torch.constant.int 1
%3360 = torch.aten.slice.Tensor %3357, %int0_3774, %int0_3775, %3359, %int1_3776 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3360, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3777 = torch.constant.int 1
%int0_3778 = torch.constant.int 0
%int9223372036854775807_3779 = torch.constant.int 9223372036854775807
%int1_3780 = torch.constant.int 1
%3361 = torch.aten.slice.Tensor %3360, %int1_3777, %int0_3778, %int9223372036854775807_3779, %int1_3780 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3361, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3781 = torch.constant.int 0
%3362 = torch.aten.unsqueeze %3361, %int0_3781 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3362, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3782 = torch.constant.int 2
%3363 = torch.aten.unsqueeze %3362, %int2_3782 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3363, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3783 = torch.constant.int 3
%int0_3784 = torch.constant.int 0
%int9223372036854775807_3785 = torch.constant.int 9223372036854775807
%int1_3786 = torch.constant.int 1
%3364 = torch.aten.slice.Tensor %3363, %int3_3783, %int0_3784, %int9223372036854775807_3785, %int1_3786 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3364, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3365 = torch_c.to_builtin_tensor %3312 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_3787 = arith.constant 1 : index
%dim_3788 = tensor.dim %3365, %c1_3787 : tensor<4x?x8x128xf16>
%3366 = flow.tensor.bitcast %3365 : tensor<4x?x8x128xf16>{%dim_3788} -> tensor<4x?x8x64xcomplex<f16>>{%dim_3788}
%3367 = torch_c.from_builtin_tensor %3366 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %3367, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%3368 = torch.aten.mul.Tensor %3367, %3364 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %3368, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%3369 = torch_c.to_builtin_tensor %3368 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_3789 = arith.constant 1 : index
%dim_3790 = tensor.dim %3369, %c1_3789 : tensor<4x?x8x64xcomplex<f32>>
%3370 = flow.tensor.bitcast %3369 : tensor<4x?x8x64xcomplex<f32>>{%dim_3790} -> tensor<4x?x8x128xf32>{%dim_3790}
%3371 = torch_c.from_builtin_tensor %3370 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %3371, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_3791 = torch.constant.int 5
%3372 = torch.prims.convert_element_type %3371, %int5_3791 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3372, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_3792 = torch.constant.int 64
%3373 = torch.aten.mul.Scalar %arg2, %int64_3792 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3373, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int30 = torch.constant.int 30
%int1_3793 = torch.constant.int 1
%3374 = torch.aten.add.Scalar %3373, %int30, %int1_3793 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3374, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3794 = torch.constant.int 4
%int32_3795 = torch.constant.int 32
%int8_3796 = torch.constant.int 8
%int128_3797 = torch.constant.int 128
%3375 = torch.prim.ListConstruct %int4_3794, %425, %int32_3795, %int8_3796, %int128_3797 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3376 = torch.aten.view %3372, %3375 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3376, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3798 = torch.constant.int 4
%3377 = torch.aten.mul.int %int4_3798, %425 : !torch.int, !torch.int -> !torch.int
%int32_3799 = torch.constant.int 32
%int8_3800 = torch.constant.int 8
%int128_3801 = torch.constant.int 128
%3378 = torch.prim.ListConstruct %3377, %int32_3799, %int8_3800, %int128_3801 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3379 = torch.aten.view %3376, %3378 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3379, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3802 = torch.constant.int 4
%3380 = torch.aten.mul.int %int4_3802, %425 : !torch.int, !torch.int -> !torch.int
%3381 = torch.prim.ListConstruct %3380 : (!torch.int) -> !torch.list<int>
%3382 = torch.aten.view %3374, %3381 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3382, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_3803 = torch.constant.int 32
%int2_3804 = torch.constant.int 2
%int32_3805 = torch.constant.int 32
%int8_3806 = torch.constant.int 8
%int128_3807 = torch.constant.int 128
%3383 = torch.prim.ListConstruct %416, %int32_3803, %int2_3804, %int32_3805, %int8_3806, %int128_3807 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3384 = torch.aten.view %3216, %3383 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3384, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3808 = torch.constant.int 32
%3385 = torch.aten.mul.int %416, %int32_3808 : !torch.int, !torch.int -> !torch.int
%int2_3809 = torch.constant.int 2
%3386 = torch.aten.mul.int %3385, %int2_3809 : !torch.int, !torch.int -> !torch.int
%int32_3810 = torch.constant.int 32
%int8_3811 = torch.constant.int 8
%int128_3812 = torch.constant.int 128
%3387 = torch.prim.ListConstruct %3386, %int32_3810, %int8_3811, %int128_3812 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3388 = torch.aten.view %3384, %3387 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3388, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%3389 = torch.prim.ListConstruct %3382 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3813 = torch.constant.bool false
%3390 = torch.aten.index_put %3388, %3389, %3379, %false_3813 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3390, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3814 = torch.constant.int 32
%int2_3815 = torch.constant.int 2
%int32_3816 = torch.constant.int 32
%int8_3817 = torch.constant.int 8
%int128_3818 = torch.constant.int 128
%3391 = torch.prim.ListConstruct %416, %int32_3814, %int2_3815, %int32_3816, %int8_3817, %int128_3818 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3392 = torch.aten.view %3390, %3391 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3392, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3819 = torch.constant.int 2097152
%3393 = torch.prim.ListConstruct %416, %int2097152_3819 : (!torch.int, !torch.int) -> !torch.list<int>
%3394 = torch.aten.view %3392, %3393 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3394, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_3820 = torch.constant.int 32
%int2_3821 = torch.constant.int 2
%int32_3822 = torch.constant.int 32
%int8_3823 = torch.constant.int 8
%int128_3824 = torch.constant.int 128
%3395 = torch.prim.ListConstruct %416, %int32_3820, %int2_3821, %int32_3822, %int8_3823, %int128_3824 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3396 = torch.aten.view %3394, %3395 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3396, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_3825 = torch.constant.int 32
%int8_3826 = torch.constant.int 8
%int128_3827 = torch.constant.int 128
%3397 = torch.prim.ListConstruct %3386, %int32_3825, %int8_3826, %int128_3827 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3398 = torch.aten.view %3396, %3397 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3398, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_3828 = torch.constant.int 4
%int32_3829 = torch.constant.int 32
%int8_3830 = torch.constant.int 8
%int128_3831 = torch.constant.int 128
%3399 = torch.prim.ListConstruct %int4_3828, %425, %int32_3829, %int8_3830, %int128_3831 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3400 = torch.aten.view %3314, %3399 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3400, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_3832 = torch.constant.int 4
%3401 = torch.aten.mul.int %int4_3832, %425 : !torch.int, !torch.int -> !torch.int
%int32_3833 = torch.constant.int 32
%int8_3834 = torch.constant.int 8
%int128_3835 = torch.constant.int 128
%3402 = torch.prim.ListConstruct %3401, %int32_3833, %int8_3834, %int128_3835 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3403 = torch.aten.view %3400, %3402 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3403, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_3836 = torch.constant.int 1
%int1_3837 = torch.constant.int 1
%3404 = torch.aten.add.Scalar %3374, %int1_3836, %int1_3837 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3404, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_3838 = torch.constant.int 4
%3405 = torch.aten.mul.int %int4_3838, %425 : !torch.int, !torch.int -> !torch.int
%3406 = torch.prim.ListConstruct %3405 : (!torch.int) -> !torch.list<int>
%3407 = torch.aten.view %3404, %3406 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3407, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3408 = torch.prim.ListConstruct %3407 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_3839 = torch.constant.bool false
%3409 = torch.aten.index_put %3398, %3408, %3403, %false_3839 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3409, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_3840 = torch.constant.int 32
%int2_3841 = torch.constant.int 2
%int32_3842 = torch.constant.int 32
%int8_3843 = torch.constant.int 8
%int128_3844 = torch.constant.int 128
%3410 = torch.prim.ListConstruct %416, %int32_3840, %int2_3841, %int32_3842, %int8_3843, %int128_3844 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3411 = torch.aten.view %3409, %3410 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3411, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_3845 = torch.constant.int 2097152
%3412 = torch.prim.ListConstruct %416, %int2097152_3845 : (!torch.int, !torch.int) -> !torch.list<int>
%3413 = torch.aten.view %3411, %3412 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3413, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_3846 = torch.constant.int -2
%3414 = torch.aten.unsqueeze %3372, %int-2_3846 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3414, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_3847 = torch.constant.int 4
%int8_3848 = torch.constant.int 8
%int4_3849 = torch.constant.int 4
%int128_3850 = torch.constant.int 128
%3415 = torch.prim.ListConstruct %int4_3847, %3358, %int8_3848, %int4_3849, %int128_3850 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3851 = torch.constant.bool false
%3416 = torch.aten.expand %3414, %3415, %false_3851 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3416, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3852 = torch.constant.int 0
%3417 = torch.aten.clone %3416, %int0_3852 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3417, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3853 = torch.constant.int 4
%int32_3854 = torch.constant.int 32
%int128_3855 = torch.constant.int 128
%3418 = torch.prim.ListConstruct %int4_3853, %3358, %int32_3854, %int128_3855 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3419 = torch.aten._unsafe_view %3417, %3418 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3419, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_3856 = torch.constant.int -2
%3420 = torch.aten.unsqueeze %3314, %int-2_3856 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3420, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_3857 = torch.constant.int 1
%3421 = torch.aten.size.int %3308, %int1_3857 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_3858 = torch.constant.int 4
%int8_3859 = torch.constant.int 8
%int4_3860 = torch.constant.int 4
%int128_3861 = torch.constant.int 128
%3422 = torch.prim.ListConstruct %int4_3858, %3421, %int8_3859, %int4_3860, %int128_3861 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_3862 = torch.constant.bool false
%3423 = torch.aten.expand %3420, %3422, %false_3862 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3423, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_3863 = torch.constant.int 0
%3424 = torch.aten.clone %3423, %int0_3863 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3424, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_3864 = torch.constant.int 4
%int32_3865 = torch.constant.int 32
%int128_3866 = torch.constant.int 128
%3425 = torch.prim.ListConstruct %int4_3864, %3421, %int32_3865, %int128_3866 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3426 = torch.aten._unsafe_view %3424, %3425 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3426, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_3867 = torch.constant.int 1
%int2_3868 = torch.constant.int 2
%3427 = torch.aten.transpose.int %3343, %int1_3867, %int2_3868 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3427, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3869 = torch.constant.int 1
%int2_3870 = torch.constant.int 2
%3428 = torch.aten.transpose.int %3419, %int1_3869, %int2_3870 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3428, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3871 = torch.constant.int 1
%int2_3872 = torch.constant.int 2
%3429 = torch.aten.transpose.int %3426, %int1_3871, %int2_3872 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3429, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_3873 = torch.constant.float 0.000000e+00
%false_3874 = torch.constant.bool false
%none_3875 = torch.constant.none
%3430:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%3427, %3428, %3429, %float0.000000e00_3873, %false_3874, %320, %none_3875) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %3430#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_3876 = torch.constant.int 1
%int2_3877 = torch.constant.int 2
%3431 = torch.aten.transpose.int %3430#0, %int1_3876, %int2_3877 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3431, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3878 = torch.constant.int 4
%int4096_3879 = torch.constant.int 4096
%3432 = torch.prim.ListConstruct %int4_3878, %3329, %int4096_3879 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3433 = torch.aten.view %3431, %3432 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3433, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3880 = torch.constant.int -2
%int-1_3881 = torch.constant.int -1
%3434 = torch.aten.transpose.int %140, %int-2_3880, %int-1_3881 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3882 = torch.constant.int 4
%3435 = torch.aten.mul.int %int4_3882, %3329 : !torch.int, !torch.int -> !torch.int
%int4096_3883 = torch.constant.int 4096
%3436 = torch.prim.ListConstruct %3435, %int4096_3883 : (!torch.int, !torch.int) -> !torch.list<int>
%3437 = torch.aten.view %3433, %3436 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3437, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3438 = torch.aten.mm %3437, %3434 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3438, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3884 = torch.constant.int 4
%int4096_3885 = torch.constant.int 4096
%3439 = torch.prim.ListConstruct %int4_3884, %3329, %int4096_3885 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3440 = torch.aten.view %3438, %3439 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3440, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3886 = torch.constant.int 1
%3441 = torch.aten.add.Tensor %3278, %3440, %int1_3886 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3441, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3887 = torch.constant.int 6
%3442 = torch.prims.convert_element_type %3441, %int6_3887 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3442, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3888 = torch.constant.int 2
%3443 = torch.aten.pow.Tensor_Scalar %3442, %int2_3888 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3443, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3889 = torch.constant.int -1
%3444 = torch.prim.ListConstruct %int-1_3889 : (!torch.int) -> !torch.list<int>
%true_3890 = torch.constant.bool true
%none_3891 = torch.constant.none
%3445 = torch.aten.mean.dim %3443, %3444, %true_3890, %none_3891 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3445, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3892 = torch.constant.float 9.9999997473787516E-6
%int1_3893 = torch.constant.int 1
%3446 = torch.aten.add.Scalar %3445, %float9.999990e-06_3892, %int1_3893 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3446, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3447 = torch.aten.rsqrt %3446 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3447, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3448 = torch.aten.mul.Tensor %3442, %3447 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3448, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3449 = torch.aten.mul.Tensor %141, %3448 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3449, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3894 = torch.constant.int 5
%3450 = torch.prims.convert_element_type %3449, %int5_3894 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3450, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3895 = torch.constant.int -2
%int-1_3896 = torch.constant.int -1
%3451 = torch.aten.transpose.int %142, %int-2_3895, %int-1_3896 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3897 = torch.constant.int 4
%3452 = torch.aten.mul.int %int4_3897, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3898 = torch.constant.int 4096
%3453 = torch.prim.ListConstruct %3452, %int4096_3898 : (!torch.int, !torch.int) -> !torch.list<int>
%3454 = torch.aten.view %3450, %3453 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3454, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3455 = torch.aten.mm %3454, %3451 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3455, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3899 = torch.constant.int 4
%int14336_3900 = torch.constant.int 14336
%3456 = torch.prim.ListConstruct %int4_3899, %294, %int14336_3900 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3457 = torch.aten.view %3455, %3456 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3457, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3458 = torch.aten.silu %3457 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3458, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3901 = torch.constant.int -2
%int-1_3902 = torch.constant.int -1
%3459 = torch.aten.transpose.int %143, %int-2_3901, %int-1_3902 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_3903 = torch.constant.int 4
%3460 = torch.aten.mul.int %int4_3903, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3904 = torch.constant.int 4096
%3461 = torch.prim.ListConstruct %3460, %int4096_3904 : (!torch.int, !torch.int) -> !torch.list<int>
%3462 = torch.aten.view %3450, %3461 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3462, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3463 = torch.aten.mm %3462, %3459 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3463, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_3905 = torch.constant.int 4
%int14336_3906 = torch.constant.int 14336
%3464 = torch.prim.ListConstruct %int4_3905, %294, %int14336_3906 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3465 = torch.aten.view %3463, %3464 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3465, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3466 = torch.aten.mul.Tensor %3458, %3465 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3466, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_3907 = torch.constant.int -2
%int-1_3908 = torch.constant.int -1
%3467 = torch.aten.transpose.int %144, %int-2_3907, %int-1_3908 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_3909 = torch.constant.int 1
%3468 = torch.aten.size.int %3457, %int1_3909 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_3910 = torch.constant.int 4
%3469 = torch.aten.mul.int %int4_3910, %3468 : !torch.int, !torch.int -> !torch.int
%int14336_3911 = torch.constant.int 14336
%3470 = torch.prim.ListConstruct %3469, %int14336_3911 : (!torch.int, !torch.int) -> !torch.list<int>
%3471 = torch.aten.view %3466, %3470 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3471, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%3472 = torch.aten.mm %3471, %3467 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3472, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3912 = torch.constant.int 4
%int4096_3913 = torch.constant.int 4096
%3473 = torch.prim.ListConstruct %int4_3912, %3468, %int4096_3913 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3474 = torch.aten.view %3472, %3473 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3474, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_3914 = torch.constant.int 1
%3475 = torch.aten.add.Tensor %3441, %3474, %int1_3914 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3475, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_3915 = torch.constant.int 6
%3476 = torch.prims.convert_element_type %3475, %int6_3915 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3476, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_3916 = torch.constant.int 2
%3477 = torch.aten.pow.Tensor_Scalar %3476, %int2_3916 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3477, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_3917 = torch.constant.int -1
%3478 = torch.prim.ListConstruct %int-1_3917 : (!torch.int) -> !torch.list<int>
%true_3918 = torch.constant.bool true
%none_3919 = torch.constant.none
%3479 = torch.aten.mean.dim %3477, %3478, %true_3918, %none_3919 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3479, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_3920 = torch.constant.float 9.9999997473787516E-6
%int1_3921 = torch.constant.int 1
%3480 = torch.aten.add.Scalar %3479, %float9.999990e-06_3920, %int1_3921 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3480, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3481 = torch.aten.rsqrt %3480 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3481, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3482 = torch.aten.mul.Tensor %3476, %3481 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3482, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3483 = torch.aten.mul.Tensor %145, %3482 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3483, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_3922 = torch.constant.int 5
%3484 = torch.prims.convert_element_type %3483, %int5_3922 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3484, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3923 = torch.constant.int -2
%int-1_3924 = torch.constant.int -1
%3485 = torch.aten.transpose.int %146, %int-2_3923, %int-1_3924 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_3925 = torch.constant.int 4
%3486 = torch.aten.mul.int %int4_3925, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3926 = torch.constant.int 4096
%3487 = torch.prim.ListConstruct %3486, %int4096_3926 : (!torch.int, !torch.int) -> !torch.list<int>
%3488 = torch.aten.view %3484, %3487 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3488, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3489 = torch.aten.mm %3488, %3485 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3489, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_3927 = torch.constant.int 4
%int4096_3928 = torch.constant.int 4096
%3490 = torch.prim.ListConstruct %int4_3927, %294, %int4096_3928 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3491 = torch.aten.view %3489, %3490 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3491, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_3929 = torch.constant.int -2
%int-1_3930 = torch.constant.int -1
%3492 = torch.aten.transpose.int %147, %int-2_3929, %int-1_3930 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3931 = torch.constant.int 4
%3493 = torch.aten.mul.int %int4_3931, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3932 = torch.constant.int 4096
%3494 = torch.prim.ListConstruct %3493, %int4096_3932 : (!torch.int, !torch.int) -> !torch.list<int>
%3495 = torch.aten.view %3484, %3494 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3495, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3496 = torch.aten.mm %3495, %3492 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3496, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3933 = torch.constant.int 4
%int1024_3934 = torch.constant.int 1024
%3497 = torch.prim.ListConstruct %int4_3933, %294, %int1024_3934 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3498 = torch.aten.view %3496, %3497 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3498, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_3935 = torch.constant.int -2
%int-1_3936 = torch.constant.int -1
%3499 = torch.aten.transpose.int %148, %int-2_3935, %int-1_3936 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_3937 = torch.constant.int 4
%3500 = torch.aten.mul.int %int4_3937, %294 : !torch.int, !torch.int -> !torch.int
%int4096_3938 = torch.constant.int 4096
%3501 = torch.prim.ListConstruct %3500, %int4096_3938 : (!torch.int, !torch.int) -> !torch.list<int>
%3502 = torch.aten.view %3484, %3501 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3502, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3503 = torch.aten.mm %3502, %3499 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3503, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_3939 = torch.constant.int 4
%int1024_3940 = torch.constant.int 1024
%3504 = torch.prim.ListConstruct %int4_3939, %294, %int1024_3940 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3505 = torch.aten.view %3503, %3504 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3505, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_3941 = torch.constant.int 4
%int32_3942 = torch.constant.int 32
%int128_3943 = torch.constant.int 128
%3506 = torch.prim.ListConstruct %int4_3941, %294, %int32_3942, %int128_3943 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3507 = torch.aten.view %3491, %3506 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3507, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_3944 = torch.constant.int 4
%int8_3945 = torch.constant.int 8
%int128_3946 = torch.constant.int 128
%3508 = torch.prim.ListConstruct %int4_3944, %294, %int8_3945, %int128_3946 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3509 = torch.aten.view %3498, %3508 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3509, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_3947 = torch.constant.int 4
%int8_3948 = torch.constant.int 8
%int128_3949 = torch.constant.int 128
%3510 = torch.prim.ListConstruct %int4_3947, %294, %int8_3948, %int128_3949 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3511 = torch.aten.view %3505, %3510 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3511, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_3950 = torch.constant.int 131072
%none_3951 = torch.constant.none
%none_3952 = torch.constant.none
%cpu_3953 = torch.constant.device "cpu"
%false_3954 = torch.constant.bool false
%3512 = torch.aten.arange %int131072_3950, %none_3951, %none_3952, %cpu_3953, %false_3954 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3955 = torch.constant.int 0
%int128_3956 = torch.constant.int 128
%int2_3957 = torch.constant.int 2
%none_3958 = torch.constant.none
%none_3959 = torch.constant.none
%cpu_3960 = torch.constant.device "cpu"
%false_3961 = torch.constant.bool false
%3513 = torch.aten.arange.start_step %int0_3955, %int128_3956, %int2_3957, %none_3958, %none_3959, %cpu_3960, %false_3961 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_3962 = torch.constant.int 0
%int0_3963 = torch.constant.int 0
%int64_3964 = torch.constant.int 64
%int1_3965 = torch.constant.int 1
%3514 = torch.aten.slice.Tensor %3513, %int0_3962, %int0_3963, %int64_3964, %int1_3965 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_3966 = torch.constant.int 6
%3515 = torch.prims.convert_element_type %3514, %int6_3966 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_3967 = torch.constant.int 128
%3516 = torch.aten.div.Scalar %3515, %int128_3967 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_3968 = torch.constant.float 5.000000e+05
%3517 = torch.aten.pow.Scalar %float5.000000e05_3968, %3516 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3518 = torch.aten.reciprocal %3517 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_3969 = torch.constant.float 1.000000e+00
%3519 = torch.aten.mul.Scalar %3518, %float1.000000e00_3969 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_3970 = torch.constant.int 131072
%int1_3971 = torch.constant.int 1
%3520 = torch.prim.ListConstruct %int131072_3970, %int1_3971 : (!torch.int, !torch.int) -> !torch.list<int>
%3521 = torch.aten.view %3512, %3520 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3522 = torch.aten.mul.Tensor %3521, %3519 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3523 = torch.aten.cos %3522 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3524 = torch.aten.sin %3522 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3525 = torch.aten.complex %3523, %3524 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_3972 = torch.constant.int 1
%3526 = torch.aten.size.int %3491, %int1_3972 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_3973 = torch.constant.int 0
%3527 = torch.aten.add.int %int0_3973, %3526 : !torch.int, !torch.int -> !torch.int
%int0_3974 = torch.constant.int 0
%int0_3975 = torch.constant.int 0
%int1_3976 = torch.constant.int 1
%3528 = torch.aten.slice.Tensor %3525, %int0_3974, %int0_3975, %3527, %int1_3976 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3528, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_3977 = torch.constant.int 1
%int0_3978 = torch.constant.int 0
%int9223372036854775807_3979 = torch.constant.int 9223372036854775807
%int1_3980 = torch.constant.int 1
%3529 = torch.aten.slice.Tensor %3528, %int1_3977, %int0_3978, %int9223372036854775807_3979, %int1_3980 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3529, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_3981 = torch.constant.int 0
%3530 = torch.aten.unsqueeze %3529, %int0_3981 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3530, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_3982 = torch.constant.int 2
%3531 = torch.aten.unsqueeze %3530, %int2_3982 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3531, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_3983 = torch.constant.int 3
%int0_3984 = torch.constant.int 0
%int9223372036854775807_3985 = torch.constant.int 9223372036854775807
%int1_3986 = torch.constant.int 1
%3532 = torch.aten.slice.Tensor %3531, %int3_3983, %int0_3984, %int9223372036854775807_3985, %int1_3986 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3532, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3533 = torch_c.to_builtin_tensor %3507 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_3987 = arith.constant 1 : index
%dim_3988 = tensor.dim %3533, %c1_3987 : tensor<4x?x32x128xf16>
%3534 = flow.tensor.bitcast %3533 : tensor<4x?x32x128xf16>{%dim_3988} -> tensor<4x?x32x64xcomplex<f16>>{%dim_3988}
%3535 = torch_c.from_builtin_tensor %3534 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %3535, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%3536 = torch.aten.mul.Tensor %3535, %3532 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %3536, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%3537 = torch_c.to_builtin_tensor %3536 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_3989 = arith.constant 1 : index
%dim_3990 = tensor.dim %3537, %c1_3989 : tensor<4x?x32x64xcomplex<f32>>
%3538 = flow.tensor.bitcast %3537 : tensor<4x?x32x64xcomplex<f32>>{%dim_3990} -> tensor<4x?x32x128xf32>{%dim_3990}
%3539 = torch_c.from_builtin_tensor %3538 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %3539, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_3991 = torch.constant.int 5
%3540 = torch.prims.convert_element_type %3539, %int5_3991 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3540, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_3992 = torch.constant.int 131072
%none_3993 = torch.constant.none
%none_3994 = torch.constant.none
%cpu_3995 = torch.constant.device "cpu"
%false_3996 = torch.constant.bool false
%3541 = torch.aten.arange %int131072_3992, %none_3993, %none_3994, %cpu_3995, %false_3996 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_3997 = torch.constant.int 0
%int128_3998 = torch.constant.int 128
%int2_3999 = torch.constant.int 2
%none_4000 = torch.constant.none
%none_4001 = torch.constant.none
%cpu_4002 = torch.constant.device "cpu"
%false_4003 = torch.constant.bool false
%3542 = torch.aten.arange.start_step %int0_3997, %int128_3998, %int2_3999, %none_4000, %none_4001, %cpu_4002, %false_4003 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4004 = torch.constant.int 0
%int0_4005 = torch.constant.int 0
%int64_4006 = torch.constant.int 64
%int1_4007 = torch.constant.int 1
%3543 = torch.aten.slice.Tensor %3542, %int0_4004, %int0_4005, %int64_4006, %int1_4007 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4008 = torch.constant.int 6
%3544 = torch.prims.convert_element_type %3543, %int6_4008 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4009 = torch.constant.int 128
%3545 = torch.aten.div.Scalar %3544, %int128_4009 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4010 = torch.constant.float 5.000000e+05
%3546 = torch.aten.pow.Scalar %float5.000000e05_4010, %3545 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3547 = torch.aten.reciprocal %3546 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4011 = torch.constant.float 1.000000e+00
%3548 = torch.aten.mul.Scalar %3547, %float1.000000e00_4011 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4012 = torch.constant.int 131072
%int1_4013 = torch.constant.int 1
%3549 = torch.prim.ListConstruct %int131072_4012, %int1_4013 : (!torch.int, !torch.int) -> !torch.list<int>
%3550 = torch.aten.view %3541, %3549 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3551 = torch.aten.mul.Tensor %3550, %3548 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3552 = torch.aten.cos %3551 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3553 = torch.aten.sin %3551 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3554 = torch.aten.complex %3552, %3553 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4014 = torch.constant.int 1
%3555 = torch.aten.size.int %3498, %int1_4014 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_4015 = torch.constant.int 0
%3556 = torch.aten.add.int %int0_4015, %3555 : !torch.int, !torch.int -> !torch.int
%int0_4016 = torch.constant.int 0
%int0_4017 = torch.constant.int 0
%int1_4018 = torch.constant.int 1
%3557 = torch.aten.slice.Tensor %3554, %int0_4016, %int0_4017, %3556, %int1_4018 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3557, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4019 = torch.constant.int 1
%int0_4020 = torch.constant.int 0
%int9223372036854775807_4021 = torch.constant.int 9223372036854775807
%int1_4022 = torch.constant.int 1
%3558 = torch.aten.slice.Tensor %3557, %int1_4019, %int0_4020, %int9223372036854775807_4021, %int1_4022 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3558, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4023 = torch.constant.int 0
%3559 = torch.aten.unsqueeze %3558, %int0_4023 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3559, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4024 = torch.constant.int 2
%3560 = torch.aten.unsqueeze %3559, %int2_4024 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3560, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4025 = torch.constant.int 3
%int0_4026 = torch.constant.int 0
%int9223372036854775807_4027 = torch.constant.int 9223372036854775807
%int1_4028 = torch.constant.int 1
%3561 = torch.aten.slice.Tensor %3560, %int3_4025, %int0_4026, %int9223372036854775807_4027, %int1_4028 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3561, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3562 = torch_c.to_builtin_tensor %3509 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_4029 = arith.constant 1 : index
%dim_4030 = tensor.dim %3562, %c1_4029 : tensor<4x?x8x128xf16>
%3563 = flow.tensor.bitcast %3562 : tensor<4x?x8x128xf16>{%dim_4030} -> tensor<4x?x8x64xcomplex<f16>>{%dim_4030}
%3564 = torch_c.from_builtin_tensor %3563 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %3564, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%3565 = torch.aten.mul.Tensor %3564, %3561 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %3565, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%3566 = torch_c.to_builtin_tensor %3565 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_4031 = arith.constant 1 : index
%dim_4032 = tensor.dim %3566, %c1_4031 : tensor<4x?x8x64xcomplex<f32>>
%3567 = flow.tensor.bitcast %3566 : tensor<4x?x8x64xcomplex<f32>>{%dim_4032} -> tensor<4x?x8x128xf32>{%dim_4032}
%3568 = torch_c.from_builtin_tensor %3567 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %3568, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_4033 = torch.constant.int 5
%3569 = torch.prims.convert_element_type %3568, %int5_4033 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3569, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_4034 = torch.constant.int 64
%3570 = torch.aten.mul.Scalar %arg2, %int64_4034 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3570, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int32_4035 = torch.constant.int 32
%int1_4036 = torch.constant.int 1
%3571 = torch.aten.add.Scalar %3570, %int32_4035, %int1_4036 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3571, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4037 = torch.constant.int 4
%int32_4038 = torch.constant.int 32
%int8_4039 = torch.constant.int 8
%int128_4040 = torch.constant.int 128
%3572 = torch.prim.ListConstruct %int4_4037, %425, %int32_4038, %int8_4039, %int128_4040 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3573 = torch.aten.view %3569, %3572 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3573, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4041 = torch.constant.int 4
%3574 = torch.aten.mul.int %int4_4041, %425 : !torch.int, !torch.int -> !torch.int
%int32_4042 = torch.constant.int 32
%int8_4043 = torch.constant.int 8
%int128_4044 = torch.constant.int 128
%3575 = torch.prim.ListConstruct %3574, %int32_4042, %int8_4043, %int128_4044 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3576 = torch.aten.view %3573, %3575 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3576, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4045 = torch.constant.int 4
%3577 = torch.aten.mul.int %int4_4045, %425 : !torch.int, !torch.int -> !torch.int
%3578 = torch.prim.ListConstruct %3577 : (!torch.int) -> !torch.list<int>
%3579 = torch.aten.view %3571, %3578 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3579, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_4046 = torch.constant.int 32
%int2_4047 = torch.constant.int 2
%int32_4048 = torch.constant.int 32
%int8_4049 = torch.constant.int 8
%int128_4050 = torch.constant.int 128
%3580 = torch.prim.ListConstruct %416, %int32_4046, %int2_4047, %int32_4048, %int8_4049, %int128_4050 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3581 = torch.aten.view %3413, %3580 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3581, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4051 = torch.constant.int 32
%3582 = torch.aten.mul.int %416, %int32_4051 : !torch.int, !torch.int -> !torch.int
%int2_4052 = torch.constant.int 2
%3583 = torch.aten.mul.int %3582, %int2_4052 : !torch.int, !torch.int -> !torch.int
%int32_4053 = torch.constant.int 32
%int8_4054 = torch.constant.int 8
%int128_4055 = torch.constant.int 128
%3584 = torch.prim.ListConstruct %3583, %int32_4053, %int8_4054, %int128_4055 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3585 = torch.aten.view %3581, %3584 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3585, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%3586 = torch.prim.ListConstruct %3579 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4056 = torch.constant.bool false
%3587 = torch.aten.index_put %3585, %3586, %3576, %false_4056 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3587, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4057 = torch.constant.int 32
%int2_4058 = torch.constant.int 2
%int32_4059 = torch.constant.int 32
%int8_4060 = torch.constant.int 8
%int128_4061 = torch.constant.int 128
%3588 = torch.prim.ListConstruct %416, %int32_4057, %int2_4058, %int32_4059, %int8_4060, %int128_4061 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3589 = torch.aten.view %3587, %3588 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3589, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4062 = torch.constant.int 2097152
%3590 = torch.prim.ListConstruct %416, %int2097152_4062 : (!torch.int, !torch.int) -> !torch.list<int>
%3591 = torch.aten.view %3589, %3590 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3591, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_4063 = torch.constant.int 32
%int2_4064 = torch.constant.int 2
%int32_4065 = torch.constant.int 32
%int8_4066 = torch.constant.int 8
%int128_4067 = torch.constant.int 128
%3592 = torch.prim.ListConstruct %416, %int32_4063, %int2_4064, %int32_4065, %int8_4066, %int128_4067 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3593 = torch.aten.view %3591, %3592 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3593, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4068 = torch.constant.int 32
%int8_4069 = torch.constant.int 8
%int128_4070 = torch.constant.int 128
%3594 = torch.prim.ListConstruct %3583, %int32_4068, %int8_4069, %int128_4070 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3595 = torch.aten.view %3593, %3594 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3595, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4071 = torch.constant.int 4
%int32_4072 = torch.constant.int 32
%int8_4073 = torch.constant.int 8
%int128_4074 = torch.constant.int 128
%3596 = torch.prim.ListConstruct %int4_4071, %425, %int32_4072, %int8_4073, %int128_4074 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3597 = torch.aten.view %3511, %3596 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3597, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4075 = torch.constant.int 4
%3598 = torch.aten.mul.int %int4_4075, %425 : !torch.int, !torch.int -> !torch.int
%int32_4076 = torch.constant.int 32
%int8_4077 = torch.constant.int 8
%int128_4078 = torch.constant.int 128
%3599 = torch.prim.ListConstruct %3598, %int32_4076, %int8_4077, %int128_4078 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3600 = torch.aten.view %3597, %3599 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3600, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_4079 = torch.constant.int 1
%int1_4080 = torch.constant.int 1
%3601 = torch.aten.add.Scalar %3571, %int1_4079, %int1_4080 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3601, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4081 = torch.constant.int 4
%3602 = torch.aten.mul.int %int4_4081, %425 : !torch.int, !torch.int -> !torch.int
%3603 = torch.prim.ListConstruct %3602 : (!torch.int) -> !torch.list<int>
%3604 = torch.aten.view %3601, %3603 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3604, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3605 = torch.prim.ListConstruct %3604 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4082 = torch.constant.bool false
%3606 = torch.aten.index_put %3595, %3605, %3600, %false_4082 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3606, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4083 = torch.constant.int 32
%int2_4084 = torch.constant.int 2
%int32_4085 = torch.constant.int 32
%int8_4086 = torch.constant.int 8
%int128_4087 = torch.constant.int 128
%3607 = torch.prim.ListConstruct %416, %int32_4083, %int2_4084, %int32_4085, %int8_4086, %int128_4087 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3608 = torch.aten.view %3606, %3607 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3608, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4088 = torch.constant.int 2097152
%3609 = torch.prim.ListConstruct %416, %int2097152_4088 : (!torch.int, !torch.int) -> !torch.list<int>
%3610 = torch.aten.view %3608, %3609 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3610, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_4089 = torch.constant.int -2
%3611 = torch.aten.unsqueeze %3569, %int-2_4089 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3611, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_4090 = torch.constant.int 4
%int8_4091 = torch.constant.int 8
%int4_4092 = torch.constant.int 4
%int128_4093 = torch.constant.int 128
%3612 = torch.prim.ListConstruct %int4_4090, %3555, %int8_4091, %int4_4092, %int128_4093 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4094 = torch.constant.bool false
%3613 = torch.aten.expand %3611, %3612, %false_4094 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3613, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4095 = torch.constant.int 0
%3614 = torch.aten.clone %3613, %int0_4095 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3614, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4096 = torch.constant.int 4
%int32_4097 = torch.constant.int 32
%int128_4098 = torch.constant.int 128
%3615 = torch.prim.ListConstruct %int4_4096, %3555, %int32_4097, %int128_4098 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3616 = torch.aten._unsafe_view %3614, %3615 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3616, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_4099 = torch.constant.int -2
%3617 = torch.aten.unsqueeze %3511, %int-2_4099 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3617, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_4100 = torch.constant.int 1
%3618 = torch.aten.size.int %3505, %int1_4100 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_4101 = torch.constant.int 4
%int8_4102 = torch.constant.int 8
%int4_4103 = torch.constant.int 4
%int128_4104 = torch.constant.int 128
%3619 = torch.prim.ListConstruct %int4_4101, %3618, %int8_4102, %int4_4103, %int128_4104 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4105 = torch.constant.bool false
%3620 = torch.aten.expand %3617, %3619, %false_4105 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3620, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4106 = torch.constant.int 0
%3621 = torch.aten.clone %3620, %int0_4106 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3621, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4107 = torch.constant.int 4
%int32_4108 = torch.constant.int 32
%int128_4109 = torch.constant.int 128
%3622 = torch.prim.ListConstruct %int4_4107, %3618, %int32_4108, %int128_4109 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3623 = torch.aten._unsafe_view %3621, %3622 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3623, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_4110 = torch.constant.int 1
%int2_4111 = torch.constant.int 2
%3624 = torch.aten.transpose.int %3540, %int1_4110, %int2_4111 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3624, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4112 = torch.constant.int 1
%int2_4113 = torch.constant.int 2
%3625 = torch.aten.transpose.int %3616, %int1_4112, %int2_4113 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3625, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4114 = torch.constant.int 1
%int2_4115 = torch.constant.int 2
%3626 = torch.aten.transpose.int %3623, %int1_4114, %int2_4115 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3626, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_4116 = torch.constant.float 0.000000e+00
%false_4117 = torch.constant.bool false
%none_4118 = torch.constant.none
%3627:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%3624, %3625, %3626, %float0.000000e00_4116, %false_4117, %320, %none_4118) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %3627#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4119 = torch.constant.int 1
%int2_4120 = torch.constant.int 2
%3628 = torch.aten.transpose.int %3627#0, %int1_4119, %int2_4120 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3628, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4121 = torch.constant.int 4
%int4096_4122 = torch.constant.int 4096
%3629 = torch.prim.ListConstruct %int4_4121, %3526, %int4096_4122 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3630 = torch.aten.view %3628, %3629 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3630, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4123 = torch.constant.int -2
%int-1_4124 = torch.constant.int -1
%3631 = torch.aten.transpose.int %149, %int-2_4123, %int-1_4124 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4125 = torch.constant.int 4
%3632 = torch.aten.mul.int %int4_4125, %3526 : !torch.int, !torch.int -> !torch.int
%int4096_4126 = torch.constant.int 4096
%3633 = torch.prim.ListConstruct %3632, %int4096_4126 : (!torch.int, !torch.int) -> !torch.list<int>
%3634 = torch.aten.view %3630, %3633 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3634, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3635 = torch.aten.mm %3634, %3631 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3635, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4127 = torch.constant.int 4
%int4096_4128 = torch.constant.int 4096
%3636 = torch.prim.ListConstruct %int4_4127, %3526, %int4096_4128 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3637 = torch.aten.view %3635, %3636 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3637, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4129 = torch.constant.int 1
%3638 = torch.aten.add.Tensor %3475, %3637, %int1_4129 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3638, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4130 = torch.constant.int 6
%3639 = torch.prims.convert_element_type %3638, %int6_4130 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3639, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4131 = torch.constant.int 2
%3640 = torch.aten.pow.Tensor_Scalar %3639, %int2_4131 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3640, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4132 = torch.constant.int -1
%3641 = torch.prim.ListConstruct %int-1_4132 : (!torch.int) -> !torch.list<int>
%true_4133 = torch.constant.bool true
%none_4134 = torch.constant.none
%3642 = torch.aten.mean.dim %3640, %3641, %true_4133, %none_4134 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3642, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4135 = torch.constant.float 9.9999997473787516E-6
%int1_4136 = torch.constant.int 1
%3643 = torch.aten.add.Scalar %3642, %float9.999990e-06_4135, %int1_4136 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3643, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3644 = torch.aten.rsqrt %3643 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3644, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3645 = torch.aten.mul.Tensor %3639, %3644 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3645, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3646 = torch.aten.mul.Tensor %150, %3645 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3646, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4137 = torch.constant.int 5
%3647 = torch.prims.convert_element_type %3646, %int5_4137 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3647, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4138 = torch.constant.int -2
%int-1_4139 = torch.constant.int -1
%3648 = torch.aten.transpose.int %151, %int-2_4138, %int-1_4139 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4140 = torch.constant.int 4
%3649 = torch.aten.mul.int %int4_4140, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4141 = torch.constant.int 4096
%3650 = torch.prim.ListConstruct %3649, %int4096_4141 : (!torch.int, !torch.int) -> !torch.list<int>
%3651 = torch.aten.view %3647, %3650 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3651, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3652 = torch.aten.mm %3651, %3648 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3652, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4142 = torch.constant.int 4
%int14336_4143 = torch.constant.int 14336
%3653 = torch.prim.ListConstruct %int4_4142, %294, %int14336_4143 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3654 = torch.aten.view %3652, %3653 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3654, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3655 = torch.aten.silu %3654 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3655, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4144 = torch.constant.int -2
%int-1_4145 = torch.constant.int -1
%3656 = torch.aten.transpose.int %152, %int-2_4144, %int-1_4145 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4146 = torch.constant.int 4
%3657 = torch.aten.mul.int %int4_4146, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4147 = torch.constant.int 4096
%3658 = torch.prim.ListConstruct %3657, %int4096_4147 : (!torch.int, !torch.int) -> !torch.list<int>
%3659 = torch.aten.view %3647, %3658 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3659, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3660 = torch.aten.mm %3659, %3656 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3660, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4148 = torch.constant.int 4
%int14336_4149 = torch.constant.int 14336
%3661 = torch.prim.ListConstruct %int4_4148, %294, %int14336_4149 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3662 = torch.aten.view %3660, %3661 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3662, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3663 = torch.aten.mul.Tensor %3655, %3662 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3663, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4150 = torch.constant.int -2
%int-1_4151 = torch.constant.int -1
%3664 = torch.aten.transpose.int %153, %int-2_4150, %int-1_4151 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_4152 = torch.constant.int 1
%3665 = torch.aten.size.int %3654, %int1_4152 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_4153 = torch.constant.int 4
%3666 = torch.aten.mul.int %int4_4153, %3665 : !torch.int, !torch.int -> !torch.int
%int14336_4154 = torch.constant.int 14336
%3667 = torch.prim.ListConstruct %3666, %int14336_4154 : (!torch.int, !torch.int) -> !torch.list<int>
%3668 = torch.aten.view %3663, %3667 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3668, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%3669 = torch.aten.mm %3668, %3664 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3669, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4155 = torch.constant.int 4
%int4096_4156 = torch.constant.int 4096
%3670 = torch.prim.ListConstruct %int4_4155, %3665, %int4096_4156 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3671 = torch.aten.view %3669, %3670 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3671, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4157 = torch.constant.int 1
%3672 = torch.aten.add.Tensor %3638, %3671, %int1_4157 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3672, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4158 = torch.constant.int 6
%3673 = torch.prims.convert_element_type %3672, %int6_4158 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3673, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4159 = torch.constant.int 2
%3674 = torch.aten.pow.Tensor_Scalar %3673, %int2_4159 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3674, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4160 = torch.constant.int -1
%3675 = torch.prim.ListConstruct %int-1_4160 : (!torch.int) -> !torch.list<int>
%true_4161 = torch.constant.bool true
%none_4162 = torch.constant.none
%3676 = torch.aten.mean.dim %3674, %3675, %true_4161, %none_4162 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3676, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4163 = torch.constant.float 9.9999997473787516E-6
%int1_4164 = torch.constant.int 1
%3677 = torch.aten.add.Scalar %3676, %float9.999990e-06_4163, %int1_4164 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3677, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3678 = torch.aten.rsqrt %3677 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3678, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3679 = torch.aten.mul.Tensor %3673, %3678 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3679, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3680 = torch.aten.mul.Tensor %154, %3679 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3680, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4165 = torch.constant.int 5
%3681 = torch.prims.convert_element_type %3680, %int5_4165 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3681, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4166 = torch.constant.int -2
%int-1_4167 = torch.constant.int -1
%3682 = torch.aten.transpose.int %155, %int-2_4166, %int-1_4167 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4168 = torch.constant.int 4
%3683 = torch.aten.mul.int %int4_4168, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4169 = torch.constant.int 4096
%3684 = torch.prim.ListConstruct %3683, %int4096_4169 : (!torch.int, !torch.int) -> !torch.list<int>
%3685 = torch.aten.view %3681, %3684 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3685, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3686 = torch.aten.mm %3685, %3682 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3686, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4170 = torch.constant.int 4
%int4096_4171 = torch.constant.int 4096
%3687 = torch.prim.ListConstruct %int4_4170, %294, %int4096_4171 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3688 = torch.aten.view %3686, %3687 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3688, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4172 = torch.constant.int -2
%int-1_4173 = torch.constant.int -1
%3689 = torch.aten.transpose.int %156, %int-2_4172, %int-1_4173 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4174 = torch.constant.int 4
%3690 = torch.aten.mul.int %int4_4174, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4175 = torch.constant.int 4096
%3691 = torch.prim.ListConstruct %3690, %int4096_4175 : (!torch.int, !torch.int) -> !torch.list<int>
%3692 = torch.aten.view %3681, %3691 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3692, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3693 = torch.aten.mm %3692, %3689 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3693, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4176 = torch.constant.int 4
%int1024_4177 = torch.constant.int 1024
%3694 = torch.prim.ListConstruct %int4_4176, %294, %int1024_4177 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3695 = torch.aten.view %3693, %3694 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3695, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_4178 = torch.constant.int -2
%int-1_4179 = torch.constant.int -1
%3696 = torch.aten.transpose.int %157, %int-2_4178, %int-1_4179 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4180 = torch.constant.int 4
%3697 = torch.aten.mul.int %int4_4180, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4181 = torch.constant.int 4096
%3698 = torch.prim.ListConstruct %3697, %int4096_4181 : (!torch.int, !torch.int) -> !torch.list<int>
%3699 = torch.aten.view %3681, %3698 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3699, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3700 = torch.aten.mm %3699, %3696 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3700, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4182 = torch.constant.int 4
%int1024_4183 = torch.constant.int 1024
%3701 = torch.prim.ListConstruct %int4_4182, %294, %int1024_4183 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3702 = torch.aten.view %3700, %3701 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3702, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_4184 = torch.constant.int 4
%int32_4185 = torch.constant.int 32
%int128_4186 = torch.constant.int 128
%3703 = torch.prim.ListConstruct %int4_4184, %294, %int32_4185, %int128_4186 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3704 = torch.aten.view %3688, %3703 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3704, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4187 = torch.constant.int 4
%int8_4188 = torch.constant.int 8
%int128_4189 = torch.constant.int 128
%3705 = torch.prim.ListConstruct %int4_4187, %294, %int8_4188, %int128_4189 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3706 = torch.aten.view %3695, %3705 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3706, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_4190 = torch.constant.int 4
%int8_4191 = torch.constant.int 8
%int128_4192 = torch.constant.int 128
%3707 = torch.prim.ListConstruct %int4_4190, %294, %int8_4191, %int128_4192 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3708 = torch.aten.view %3702, %3707 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3708, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_4193 = torch.constant.int 131072
%none_4194 = torch.constant.none
%none_4195 = torch.constant.none
%cpu_4196 = torch.constant.device "cpu"
%false_4197 = torch.constant.bool false
%3709 = torch.aten.arange %int131072_4193, %none_4194, %none_4195, %cpu_4196, %false_4197 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4198 = torch.constant.int 0
%int128_4199 = torch.constant.int 128
%int2_4200 = torch.constant.int 2
%none_4201 = torch.constant.none
%none_4202 = torch.constant.none
%cpu_4203 = torch.constant.device "cpu"
%false_4204 = torch.constant.bool false
%3710 = torch.aten.arange.start_step %int0_4198, %int128_4199, %int2_4200, %none_4201, %none_4202, %cpu_4203, %false_4204 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4205 = torch.constant.int 0
%int0_4206 = torch.constant.int 0
%int64_4207 = torch.constant.int 64
%int1_4208 = torch.constant.int 1
%3711 = torch.aten.slice.Tensor %3710, %int0_4205, %int0_4206, %int64_4207, %int1_4208 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4209 = torch.constant.int 6
%3712 = torch.prims.convert_element_type %3711, %int6_4209 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4210 = torch.constant.int 128
%3713 = torch.aten.div.Scalar %3712, %int128_4210 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4211 = torch.constant.float 5.000000e+05
%3714 = torch.aten.pow.Scalar %float5.000000e05_4211, %3713 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3715 = torch.aten.reciprocal %3714 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4212 = torch.constant.float 1.000000e+00
%3716 = torch.aten.mul.Scalar %3715, %float1.000000e00_4212 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4213 = torch.constant.int 131072
%int1_4214 = torch.constant.int 1
%3717 = torch.prim.ListConstruct %int131072_4213, %int1_4214 : (!torch.int, !torch.int) -> !torch.list<int>
%3718 = torch.aten.view %3709, %3717 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3719 = torch.aten.mul.Tensor %3718, %3716 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3720 = torch.aten.cos %3719 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3721 = torch.aten.sin %3719 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3722 = torch.aten.complex %3720, %3721 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4215 = torch.constant.int 1
%3723 = torch.aten.size.int %3688, %int1_4215 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_4216 = torch.constant.int 0
%3724 = torch.aten.add.int %int0_4216, %3723 : !torch.int, !torch.int -> !torch.int
%int0_4217 = torch.constant.int 0
%int0_4218 = torch.constant.int 0
%int1_4219 = torch.constant.int 1
%3725 = torch.aten.slice.Tensor %3722, %int0_4217, %int0_4218, %3724, %int1_4219 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3725, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4220 = torch.constant.int 1
%int0_4221 = torch.constant.int 0
%int9223372036854775807_4222 = torch.constant.int 9223372036854775807
%int1_4223 = torch.constant.int 1
%3726 = torch.aten.slice.Tensor %3725, %int1_4220, %int0_4221, %int9223372036854775807_4222, %int1_4223 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3726, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4224 = torch.constant.int 0
%3727 = torch.aten.unsqueeze %3726, %int0_4224 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3727, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4225 = torch.constant.int 2
%3728 = torch.aten.unsqueeze %3727, %int2_4225 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3728, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4226 = torch.constant.int 3
%int0_4227 = torch.constant.int 0
%int9223372036854775807_4228 = torch.constant.int 9223372036854775807
%int1_4229 = torch.constant.int 1
%3729 = torch.aten.slice.Tensor %3728, %int3_4226, %int0_4227, %int9223372036854775807_4228, %int1_4229 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3729, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3730 = torch_c.to_builtin_tensor %3704 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_4230 = arith.constant 1 : index
%dim_4231 = tensor.dim %3730, %c1_4230 : tensor<4x?x32x128xf16>
%3731 = flow.tensor.bitcast %3730 : tensor<4x?x32x128xf16>{%dim_4231} -> tensor<4x?x32x64xcomplex<f16>>{%dim_4231}
%3732 = torch_c.from_builtin_tensor %3731 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %3732, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%3733 = torch.aten.mul.Tensor %3732, %3729 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %3733, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%3734 = torch_c.to_builtin_tensor %3733 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_4232 = arith.constant 1 : index
%dim_4233 = tensor.dim %3734, %c1_4232 : tensor<4x?x32x64xcomplex<f32>>
%3735 = flow.tensor.bitcast %3734 : tensor<4x?x32x64xcomplex<f32>>{%dim_4233} -> tensor<4x?x32x128xf32>{%dim_4233}
%3736 = torch_c.from_builtin_tensor %3735 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %3736, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_4234 = torch.constant.int 5
%3737 = torch.prims.convert_element_type %3736, %int5_4234 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3737, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_4235 = torch.constant.int 131072
%none_4236 = torch.constant.none
%none_4237 = torch.constant.none
%cpu_4238 = torch.constant.device "cpu"
%false_4239 = torch.constant.bool false
%3738 = torch.aten.arange %int131072_4235, %none_4236, %none_4237, %cpu_4238, %false_4239 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4240 = torch.constant.int 0
%int128_4241 = torch.constant.int 128
%int2_4242 = torch.constant.int 2
%none_4243 = torch.constant.none
%none_4244 = torch.constant.none
%cpu_4245 = torch.constant.device "cpu"
%false_4246 = torch.constant.bool false
%3739 = torch.aten.arange.start_step %int0_4240, %int128_4241, %int2_4242, %none_4243, %none_4244, %cpu_4245, %false_4246 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4247 = torch.constant.int 0
%int0_4248 = torch.constant.int 0
%int64_4249 = torch.constant.int 64
%int1_4250 = torch.constant.int 1
%3740 = torch.aten.slice.Tensor %3739, %int0_4247, %int0_4248, %int64_4249, %int1_4250 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4251 = torch.constant.int 6
%3741 = torch.prims.convert_element_type %3740, %int6_4251 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4252 = torch.constant.int 128
%3742 = torch.aten.div.Scalar %3741, %int128_4252 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4253 = torch.constant.float 5.000000e+05
%3743 = torch.aten.pow.Scalar %float5.000000e05_4253, %3742 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3744 = torch.aten.reciprocal %3743 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4254 = torch.constant.float 1.000000e+00
%3745 = torch.aten.mul.Scalar %3744, %float1.000000e00_4254 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4255 = torch.constant.int 131072
%int1_4256 = torch.constant.int 1
%3746 = torch.prim.ListConstruct %int131072_4255, %int1_4256 : (!torch.int, !torch.int) -> !torch.list<int>
%3747 = torch.aten.view %3738, %3746 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3748 = torch.aten.mul.Tensor %3747, %3745 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3749 = torch.aten.cos %3748 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3750 = torch.aten.sin %3748 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3751 = torch.aten.complex %3749, %3750 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4257 = torch.constant.int 1
%3752 = torch.aten.size.int %3695, %int1_4257 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_4258 = torch.constant.int 0
%3753 = torch.aten.add.int %int0_4258, %3752 : !torch.int, !torch.int -> !torch.int
%int0_4259 = torch.constant.int 0
%int0_4260 = torch.constant.int 0
%int1_4261 = torch.constant.int 1
%3754 = torch.aten.slice.Tensor %3751, %int0_4259, %int0_4260, %3753, %int1_4261 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3754, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4262 = torch.constant.int 1
%int0_4263 = torch.constant.int 0
%int9223372036854775807_4264 = torch.constant.int 9223372036854775807
%int1_4265 = torch.constant.int 1
%3755 = torch.aten.slice.Tensor %3754, %int1_4262, %int0_4263, %int9223372036854775807_4264, %int1_4265 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3755, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4266 = torch.constant.int 0
%3756 = torch.aten.unsqueeze %3755, %int0_4266 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3756, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4267 = torch.constant.int 2
%3757 = torch.aten.unsqueeze %3756, %int2_4267 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3757, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4268 = torch.constant.int 3
%int0_4269 = torch.constant.int 0
%int9223372036854775807_4270 = torch.constant.int 9223372036854775807
%int1_4271 = torch.constant.int 1
%3758 = torch.aten.slice.Tensor %3757, %int3_4268, %int0_4269, %int9223372036854775807_4270, %int1_4271 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3758, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3759 = torch_c.to_builtin_tensor %3706 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_4272 = arith.constant 1 : index
%dim_4273 = tensor.dim %3759, %c1_4272 : tensor<4x?x8x128xf16>
%3760 = flow.tensor.bitcast %3759 : tensor<4x?x8x128xf16>{%dim_4273} -> tensor<4x?x8x64xcomplex<f16>>{%dim_4273}
%3761 = torch_c.from_builtin_tensor %3760 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %3761, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%3762 = torch.aten.mul.Tensor %3761, %3758 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %3762, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%3763 = torch_c.to_builtin_tensor %3762 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_4274 = arith.constant 1 : index
%dim_4275 = tensor.dim %3763, %c1_4274 : tensor<4x?x8x64xcomplex<f32>>
%3764 = flow.tensor.bitcast %3763 : tensor<4x?x8x64xcomplex<f32>>{%dim_4275} -> tensor<4x?x8x128xf32>{%dim_4275}
%3765 = torch_c.from_builtin_tensor %3764 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %3765, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_4276 = torch.constant.int 5
%3766 = torch.prims.convert_element_type %3765, %int5_4276 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3766, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_4277 = torch.constant.int 64
%3767 = torch.aten.mul.Scalar %arg2, %int64_4277 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3767, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int34 = torch.constant.int 34
%int1_4278 = torch.constant.int 1
%3768 = torch.aten.add.Scalar %3767, %int34, %int1_4278 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3768, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4279 = torch.constant.int 4
%int32_4280 = torch.constant.int 32
%int8_4281 = torch.constant.int 8
%int128_4282 = torch.constant.int 128
%3769 = torch.prim.ListConstruct %int4_4279, %425, %int32_4280, %int8_4281, %int128_4282 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3770 = torch.aten.view %3766, %3769 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3770, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4283 = torch.constant.int 4
%3771 = torch.aten.mul.int %int4_4283, %425 : !torch.int, !torch.int -> !torch.int
%int32_4284 = torch.constant.int 32
%int8_4285 = torch.constant.int 8
%int128_4286 = torch.constant.int 128
%3772 = torch.prim.ListConstruct %3771, %int32_4284, %int8_4285, %int128_4286 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3773 = torch.aten.view %3770, %3772 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3773, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4287 = torch.constant.int 4
%3774 = torch.aten.mul.int %int4_4287, %425 : !torch.int, !torch.int -> !torch.int
%3775 = torch.prim.ListConstruct %3774 : (!torch.int) -> !torch.list<int>
%3776 = torch.aten.view %3768, %3775 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3776, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_4288 = torch.constant.int 32
%int2_4289 = torch.constant.int 2
%int32_4290 = torch.constant.int 32
%int8_4291 = torch.constant.int 8
%int128_4292 = torch.constant.int 128
%3777 = torch.prim.ListConstruct %416, %int32_4288, %int2_4289, %int32_4290, %int8_4291, %int128_4292 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3778 = torch.aten.view %3610, %3777 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3778, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4293 = torch.constant.int 32
%3779 = torch.aten.mul.int %416, %int32_4293 : !torch.int, !torch.int -> !torch.int
%int2_4294 = torch.constant.int 2
%3780 = torch.aten.mul.int %3779, %int2_4294 : !torch.int, !torch.int -> !torch.int
%int32_4295 = torch.constant.int 32
%int8_4296 = torch.constant.int 8
%int128_4297 = torch.constant.int 128
%3781 = torch.prim.ListConstruct %3780, %int32_4295, %int8_4296, %int128_4297 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3782 = torch.aten.view %3778, %3781 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3782, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%3783 = torch.prim.ListConstruct %3776 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4298 = torch.constant.bool false
%3784 = torch.aten.index_put %3782, %3783, %3773, %false_4298 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3784, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4299 = torch.constant.int 32
%int2_4300 = torch.constant.int 2
%int32_4301 = torch.constant.int 32
%int8_4302 = torch.constant.int 8
%int128_4303 = torch.constant.int 128
%3785 = torch.prim.ListConstruct %416, %int32_4299, %int2_4300, %int32_4301, %int8_4302, %int128_4303 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3786 = torch.aten.view %3784, %3785 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3786, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4304 = torch.constant.int 2097152
%3787 = torch.prim.ListConstruct %416, %int2097152_4304 : (!torch.int, !torch.int) -> !torch.list<int>
%3788 = torch.aten.view %3786, %3787 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3788, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_4305 = torch.constant.int 32
%int2_4306 = torch.constant.int 2
%int32_4307 = torch.constant.int 32
%int8_4308 = torch.constant.int 8
%int128_4309 = torch.constant.int 128
%3789 = torch.prim.ListConstruct %416, %int32_4305, %int2_4306, %int32_4307, %int8_4308, %int128_4309 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3790 = torch.aten.view %3788, %3789 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3790, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4310 = torch.constant.int 32
%int8_4311 = torch.constant.int 8
%int128_4312 = torch.constant.int 128
%3791 = torch.prim.ListConstruct %3780, %int32_4310, %int8_4311, %int128_4312 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3792 = torch.aten.view %3790, %3791 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3792, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4313 = torch.constant.int 4
%int32_4314 = torch.constant.int 32
%int8_4315 = torch.constant.int 8
%int128_4316 = torch.constant.int 128
%3793 = torch.prim.ListConstruct %int4_4313, %425, %int32_4314, %int8_4315, %int128_4316 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3794 = torch.aten.view %3708, %3793 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3794, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4317 = torch.constant.int 4
%3795 = torch.aten.mul.int %int4_4317, %425 : !torch.int, !torch.int -> !torch.int
%int32_4318 = torch.constant.int 32
%int8_4319 = torch.constant.int 8
%int128_4320 = torch.constant.int 128
%3796 = torch.prim.ListConstruct %3795, %int32_4318, %int8_4319, %int128_4320 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3797 = torch.aten.view %3794, %3796 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3797, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_4321 = torch.constant.int 1
%int1_4322 = torch.constant.int 1
%3798 = torch.aten.add.Scalar %3768, %int1_4321, %int1_4322 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3798, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4323 = torch.constant.int 4
%3799 = torch.aten.mul.int %int4_4323, %425 : !torch.int, !torch.int -> !torch.int
%3800 = torch.prim.ListConstruct %3799 : (!torch.int) -> !torch.list<int>
%3801 = torch.aten.view %3798, %3800 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3801, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3802 = torch.prim.ListConstruct %3801 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4324 = torch.constant.bool false
%3803 = torch.aten.index_put %3792, %3802, %3797, %false_4324 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3803, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4325 = torch.constant.int 32
%int2_4326 = torch.constant.int 2
%int32_4327 = torch.constant.int 32
%int8_4328 = torch.constant.int 8
%int128_4329 = torch.constant.int 128
%3804 = torch.prim.ListConstruct %416, %int32_4325, %int2_4326, %int32_4327, %int8_4328, %int128_4329 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3805 = torch.aten.view %3803, %3804 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3805, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4330 = torch.constant.int 2097152
%3806 = torch.prim.ListConstruct %416, %int2097152_4330 : (!torch.int, !torch.int) -> !torch.list<int>
%3807 = torch.aten.view %3805, %3806 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3807, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_4331 = torch.constant.int -2
%3808 = torch.aten.unsqueeze %3766, %int-2_4331 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3808, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_4332 = torch.constant.int 4
%int8_4333 = torch.constant.int 8
%int4_4334 = torch.constant.int 4
%int128_4335 = torch.constant.int 128
%3809 = torch.prim.ListConstruct %int4_4332, %3752, %int8_4333, %int4_4334, %int128_4335 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4336 = torch.constant.bool false
%3810 = torch.aten.expand %3808, %3809, %false_4336 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3810, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4337 = torch.constant.int 0
%3811 = torch.aten.clone %3810, %int0_4337 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3811, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4338 = torch.constant.int 4
%int32_4339 = torch.constant.int 32
%int128_4340 = torch.constant.int 128
%3812 = torch.prim.ListConstruct %int4_4338, %3752, %int32_4339, %int128_4340 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3813 = torch.aten._unsafe_view %3811, %3812 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3813, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_4341 = torch.constant.int -2
%3814 = torch.aten.unsqueeze %3708, %int-2_4341 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %3814, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_4342 = torch.constant.int 1
%3815 = torch.aten.size.int %3702, %int1_4342 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_4343 = torch.constant.int 4
%int8_4344 = torch.constant.int 8
%int4_4345 = torch.constant.int 4
%int128_4346 = torch.constant.int 128
%3816 = torch.prim.ListConstruct %int4_4343, %3815, %int8_4344, %int4_4345, %int128_4346 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4347 = torch.constant.bool false
%3817 = torch.aten.expand %3814, %3816, %false_4347 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3817, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4348 = torch.constant.int 0
%3818 = torch.aten.clone %3817, %int0_4348 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %3818, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4349 = torch.constant.int 4
%int32_4350 = torch.constant.int 32
%int128_4351 = torch.constant.int 128
%3819 = torch.prim.ListConstruct %int4_4349, %3815, %int32_4350, %int128_4351 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3820 = torch.aten._unsafe_view %3818, %3819 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3820, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_4352 = torch.constant.int 1
%int2_4353 = torch.constant.int 2
%3821 = torch.aten.transpose.int %3737, %int1_4352, %int2_4353 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3821, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4354 = torch.constant.int 1
%int2_4355 = torch.constant.int 2
%3822 = torch.aten.transpose.int %3813, %int1_4354, %int2_4355 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3822, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4356 = torch.constant.int 1
%int2_4357 = torch.constant.int 2
%3823 = torch.aten.transpose.int %3820, %int1_4356, %int2_4357 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %3823, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_4358 = torch.constant.float 0.000000e+00
%false_4359 = torch.constant.bool false
%none_4360 = torch.constant.none
%3824:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%3821, %3822, %3823, %float0.000000e00_4358, %false_4359, %320, %none_4360) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %3824#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4361 = torch.constant.int 1
%int2_4362 = torch.constant.int 2
%3825 = torch.aten.transpose.int %3824#0, %int1_4361, %int2_4362 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3825, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4363 = torch.constant.int 4
%int4096_4364 = torch.constant.int 4096
%3826 = torch.prim.ListConstruct %int4_4363, %3723, %int4096_4364 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3827 = torch.aten.view %3825, %3826 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3827, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4365 = torch.constant.int -2
%int-1_4366 = torch.constant.int -1
%3828 = torch.aten.transpose.int %158, %int-2_4365, %int-1_4366 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4367 = torch.constant.int 4
%3829 = torch.aten.mul.int %int4_4367, %3723 : !torch.int, !torch.int -> !torch.int
%int4096_4368 = torch.constant.int 4096
%3830 = torch.prim.ListConstruct %3829, %int4096_4368 : (!torch.int, !torch.int) -> !torch.list<int>
%3831 = torch.aten.view %3827, %3830 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3831, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3832 = torch.aten.mm %3831, %3828 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3832, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4369 = torch.constant.int 4
%int4096_4370 = torch.constant.int 4096
%3833 = torch.prim.ListConstruct %int4_4369, %3723, %int4096_4370 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3834 = torch.aten.view %3832, %3833 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3834, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4371 = torch.constant.int 1
%3835 = torch.aten.add.Tensor %3672, %3834, %int1_4371 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3835, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4372 = torch.constant.int 6
%3836 = torch.prims.convert_element_type %3835, %int6_4372 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3836, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4373 = torch.constant.int 2
%3837 = torch.aten.pow.Tensor_Scalar %3836, %int2_4373 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3837, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4374 = torch.constant.int -1
%3838 = torch.prim.ListConstruct %int-1_4374 : (!torch.int) -> !torch.list<int>
%true_4375 = torch.constant.bool true
%none_4376 = torch.constant.none
%3839 = torch.aten.mean.dim %3837, %3838, %true_4375, %none_4376 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3839, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4377 = torch.constant.float 9.9999997473787516E-6
%int1_4378 = torch.constant.int 1
%3840 = torch.aten.add.Scalar %3839, %float9.999990e-06_4377, %int1_4378 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3840, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3841 = torch.aten.rsqrt %3840 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3841, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3842 = torch.aten.mul.Tensor %3836, %3841 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3842, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3843 = torch.aten.mul.Tensor %159, %3842 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3843, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4379 = torch.constant.int 5
%3844 = torch.prims.convert_element_type %3843, %int5_4379 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3844, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4380 = torch.constant.int -2
%int-1_4381 = torch.constant.int -1
%3845 = torch.aten.transpose.int %160, %int-2_4380, %int-1_4381 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4382 = torch.constant.int 4
%3846 = torch.aten.mul.int %int4_4382, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4383 = torch.constant.int 4096
%3847 = torch.prim.ListConstruct %3846, %int4096_4383 : (!torch.int, !torch.int) -> !torch.list<int>
%3848 = torch.aten.view %3844, %3847 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3848, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3849 = torch.aten.mm %3848, %3845 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3849, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4384 = torch.constant.int 4
%int14336_4385 = torch.constant.int 14336
%3850 = torch.prim.ListConstruct %int4_4384, %294, %int14336_4385 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3851 = torch.aten.view %3849, %3850 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3851, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3852 = torch.aten.silu %3851 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3852, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4386 = torch.constant.int -2
%int-1_4387 = torch.constant.int -1
%3853 = torch.aten.transpose.int %161, %int-2_4386, %int-1_4387 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4388 = torch.constant.int 4
%3854 = torch.aten.mul.int %int4_4388, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4389 = torch.constant.int 4096
%3855 = torch.prim.ListConstruct %3854, %int4096_4389 : (!torch.int, !torch.int) -> !torch.list<int>
%3856 = torch.aten.view %3844, %3855 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3856, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3857 = torch.aten.mm %3856, %3853 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3857, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4390 = torch.constant.int 4
%int14336_4391 = torch.constant.int 14336
%3858 = torch.prim.ListConstruct %int4_4390, %294, %int14336_4391 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3859 = torch.aten.view %3857, %3858 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3859, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%3860 = torch.aten.mul.Tensor %3852, %3859 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %3860, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4392 = torch.constant.int -2
%int-1_4393 = torch.constant.int -1
%3861 = torch.aten.transpose.int %162, %int-2_4392, %int-1_4393 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_4394 = torch.constant.int 1
%3862 = torch.aten.size.int %3851, %int1_4394 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_4395 = torch.constant.int 4
%3863 = torch.aten.mul.int %int4_4395, %3862 : !torch.int, !torch.int -> !torch.int
%int14336_4396 = torch.constant.int 14336
%3864 = torch.prim.ListConstruct %3863, %int14336_4396 : (!torch.int, !torch.int) -> !torch.list<int>
%3865 = torch.aten.view %3860, %3864 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %3865, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%3866 = torch.aten.mm %3865, %3861 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3866, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4397 = torch.constant.int 4
%int4096_4398 = torch.constant.int 4096
%3867 = torch.prim.ListConstruct %int4_4397, %3862, %int4096_4398 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3868 = torch.aten.view %3866, %3867 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3868, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4399 = torch.constant.int 1
%3869 = torch.aten.add.Tensor %3835, %3868, %int1_4399 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3869, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4400 = torch.constant.int 6
%3870 = torch.prims.convert_element_type %3869, %int6_4400 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3870, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4401 = torch.constant.int 2
%3871 = torch.aten.pow.Tensor_Scalar %3870, %int2_4401 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3871, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4402 = torch.constant.int -1
%3872 = torch.prim.ListConstruct %int-1_4402 : (!torch.int) -> !torch.list<int>
%true_4403 = torch.constant.bool true
%none_4404 = torch.constant.none
%3873 = torch.aten.mean.dim %3871, %3872, %true_4403, %none_4404 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3873, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4405 = torch.constant.float 9.9999997473787516E-6
%int1_4406 = torch.constant.int 1
%3874 = torch.aten.add.Scalar %3873, %float9.999990e-06_4405, %int1_4406 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3874, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3875 = torch.aten.rsqrt %3874 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %3875, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%3876 = torch.aten.mul.Tensor %3870, %3875 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3876, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%3877 = torch.aten.mul.Tensor %163, %3876 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %3877, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4407 = torch.constant.int 5
%3878 = torch.prims.convert_element_type %3877, %int5_4407 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3878, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4408 = torch.constant.int -2
%int-1_4409 = torch.constant.int -1
%3879 = torch.aten.transpose.int %164, %int-2_4408, %int-1_4409 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4410 = torch.constant.int 4
%3880 = torch.aten.mul.int %int4_4410, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4411 = torch.constant.int 4096
%3881 = torch.prim.ListConstruct %3880, %int4096_4411 : (!torch.int, !torch.int) -> !torch.list<int>
%3882 = torch.aten.view %3878, %3881 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3882, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3883 = torch.aten.mm %3882, %3879 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3883, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4412 = torch.constant.int 4
%int4096_4413 = torch.constant.int 4096
%3884 = torch.prim.ListConstruct %int4_4412, %294, %int4096_4413 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3885 = torch.aten.view %3883, %3884 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %3885, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4414 = torch.constant.int -2
%int-1_4415 = torch.constant.int -1
%3886 = torch.aten.transpose.int %165, %int-2_4414, %int-1_4415 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4416 = torch.constant.int 4
%3887 = torch.aten.mul.int %int4_4416, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4417 = torch.constant.int 4096
%3888 = torch.prim.ListConstruct %3887, %int4096_4417 : (!torch.int, !torch.int) -> !torch.list<int>
%3889 = torch.aten.view %3878, %3888 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3889, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3890 = torch.aten.mm %3889, %3886 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3890, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4418 = torch.constant.int 4
%int1024_4419 = torch.constant.int 1024
%3891 = torch.prim.ListConstruct %int4_4418, %294, %int1024_4419 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3892 = torch.aten.view %3890, %3891 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3892, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_4420 = torch.constant.int -2
%int-1_4421 = torch.constant.int -1
%3893 = torch.aten.transpose.int %166, %int-2_4420, %int-1_4421 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4422 = torch.constant.int 4
%3894 = torch.aten.mul.int %int4_4422, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4423 = torch.constant.int 4096
%3895 = torch.prim.ListConstruct %3894, %int4096_4423 : (!torch.int, !torch.int) -> !torch.list<int>
%3896 = torch.aten.view %3878, %3895 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %3896, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%3897 = torch.aten.mm %3896, %3893 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %3897, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4424 = torch.constant.int 4
%int1024_4425 = torch.constant.int 1024
%3898 = torch.prim.ListConstruct %int4_4424, %294, %int1024_4425 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3899 = torch.aten.view %3897, %3898 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %3899, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_4426 = torch.constant.int 4
%int32_4427 = torch.constant.int 32
%int128_4428 = torch.constant.int 128
%3900 = torch.prim.ListConstruct %int4_4426, %294, %int32_4427, %int128_4428 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3901 = torch.aten.view %3885, %3900 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3901, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4429 = torch.constant.int 4
%int8_4430 = torch.constant.int 8
%int128_4431 = torch.constant.int 128
%3902 = torch.prim.ListConstruct %int4_4429, %294, %int8_4430, %int128_4431 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3903 = torch.aten.view %3892, %3902 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3903, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_4432 = torch.constant.int 4
%int8_4433 = torch.constant.int 8
%int128_4434 = torch.constant.int 128
%3904 = torch.prim.ListConstruct %int4_4432, %294, %int8_4433, %int128_4434 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3905 = torch.aten.view %3899, %3904 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3905, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_4435 = torch.constant.int 131072
%none_4436 = torch.constant.none
%none_4437 = torch.constant.none
%cpu_4438 = torch.constant.device "cpu"
%false_4439 = torch.constant.bool false
%3906 = torch.aten.arange %int131072_4435, %none_4436, %none_4437, %cpu_4438, %false_4439 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4440 = torch.constant.int 0
%int128_4441 = torch.constant.int 128
%int2_4442 = torch.constant.int 2
%none_4443 = torch.constant.none
%none_4444 = torch.constant.none
%cpu_4445 = torch.constant.device "cpu"
%false_4446 = torch.constant.bool false
%3907 = torch.aten.arange.start_step %int0_4440, %int128_4441, %int2_4442, %none_4443, %none_4444, %cpu_4445, %false_4446 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4447 = torch.constant.int 0
%int0_4448 = torch.constant.int 0
%int64_4449 = torch.constant.int 64
%int1_4450 = torch.constant.int 1
%3908 = torch.aten.slice.Tensor %3907, %int0_4447, %int0_4448, %int64_4449, %int1_4450 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4451 = torch.constant.int 6
%3909 = torch.prims.convert_element_type %3908, %int6_4451 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4452 = torch.constant.int 128
%3910 = torch.aten.div.Scalar %3909, %int128_4452 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4453 = torch.constant.float 5.000000e+05
%3911 = torch.aten.pow.Scalar %float5.000000e05_4453, %3910 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3912 = torch.aten.reciprocal %3911 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4454 = torch.constant.float 1.000000e+00
%3913 = torch.aten.mul.Scalar %3912, %float1.000000e00_4454 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4455 = torch.constant.int 131072
%int1_4456 = torch.constant.int 1
%3914 = torch.prim.ListConstruct %int131072_4455, %int1_4456 : (!torch.int, !torch.int) -> !torch.list<int>
%3915 = torch.aten.view %3906, %3914 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3916 = torch.aten.mul.Tensor %3915, %3913 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3917 = torch.aten.cos %3916 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3918 = torch.aten.sin %3916 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3919 = torch.aten.complex %3917, %3918 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4457 = torch.constant.int 1
%3920 = torch.aten.size.int %3885, %int1_4457 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_4458 = torch.constant.int 0
%3921 = torch.aten.add.int %int0_4458, %3920 : !torch.int, !torch.int -> !torch.int
%int0_4459 = torch.constant.int 0
%int0_4460 = torch.constant.int 0
%int1_4461 = torch.constant.int 1
%3922 = torch.aten.slice.Tensor %3919, %int0_4459, %int0_4460, %3921, %int1_4461 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3922, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4462 = torch.constant.int 1
%int0_4463 = torch.constant.int 0
%int9223372036854775807_4464 = torch.constant.int 9223372036854775807
%int1_4465 = torch.constant.int 1
%3923 = torch.aten.slice.Tensor %3922, %int1_4462, %int0_4463, %int9223372036854775807_4464, %int1_4465 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3923, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4466 = torch.constant.int 0
%3924 = torch.aten.unsqueeze %3923, %int0_4466 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3924, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4467 = torch.constant.int 2
%3925 = torch.aten.unsqueeze %3924, %int2_4467 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3925, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4468 = torch.constant.int 3
%int0_4469 = torch.constant.int 0
%int9223372036854775807_4470 = torch.constant.int 9223372036854775807
%int1_4471 = torch.constant.int 1
%3926 = torch.aten.slice.Tensor %3925, %int3_4468, %int0_4469, %int9223372036854775807_4470, %int1_4471 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3926, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3927 = torch_c.to_builtin_tensor %3901 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_4472 = arith.constant 1 : index
%dim_4473 = tensor.dim %3927, %c1_4472 : tensor<4x?x32x128xf16>
%3928 = flow.tensor.bitcast %3927 : tensor<4x?x32x128xf16>{%dim_4473} -> tensor<4x?x32x64xcomplex<f16>>{%dim_4473}
%3929 = torch_c.from_builtin_tensor %3928 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %3929, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%3930 = torch.aten.mul.Tensor %3929, %3926 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %3930, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%3931 = torch_c.to_builtin_tensor %3930 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_4474 = arith.constant 1 : index
%dim_4475 = tensor.dim %3931, %c1_4474 : tensor<4x?x32x64xcomplex<f32>>
%3932 = flow.tensor.bitcast %3931 : tensor<4x?x32x64xcomplex<f32>>{%dim_4475} -> tensor<4x?x32x128xf32>{%dim_4475}
%3933 = torch_c.from_builtin_tensor %3932 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %3933, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_4476 = torch.constant.int 5
%3934 = torch.prims.convert_element_type %3933, %int5_4476 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %3934, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_4477 = torch.constant.int 131072
%none_4478 = torch.constant.none
%none_4479 = torch.constant.none
%cpu_4480 = torch.constant.device "cpu"
%false_4481 = torch.constant.bool false
%3935 = torch.aten.arange %int131072_4477, %none_4478, %none_4479, %cpu_4480, %false_4481 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4482 = torch.constant.int 0
%int128_4483 = torch.constant.int 128
%int2_4484 = torch.constant.int 2
%none_4485 = torch.constant.none
%none_4486 = torch.constant.none
%cpu_4487 = torch.constant.device "cpu"
%false_4488 = torch.constant.bool false
%3936 = torch.aten.arange.start_step %int0_4482, %int128_4483, %int2_4484, %none_4485, %none_4486, %cpu_4487, %false_4488 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4489 = torch.constant.int 0
%int0_4490 = torch.constant.int 0
%int64_4491 = torch.constant.int 64
%int1_4492 = torch.constant.int 1
%3937 = torch.aten.slice.Tensor %3936, %int0_4489, %int0_4490, %int64_4491, %int1_4492 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4493 = torch.constant.int 6
%3938 = torch.prims.convert_element_type %3937, %int6_4493 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4494 = torch.constant.int 128
%3939 = torch.aten.div.Scalar %3938, %int128_4494 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4495 = torch.constant.float 5.000000e+05
%3940 = torch.aten.pow.Scalar %float5.000000e05_4495, %3939 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%3941 = torch.aten.reciprocal %3940 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4496 = torch.constant.float 1.000000e+00
%3942 = torch.aten.mul.Scalar %3941, %float1.000000e00_4496 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4497 = torch.constant.int 131072
%int1_4498 = torch.constant.int 1
%3943 = torch.prim.ListConstruct %int131072_4497, %int1_4498 : (!torch.int, !torch.int) -> !torch.list<int>
%3944 = torch.aten.view %3935, %3943 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%3945 = torch.aten.mul.Tensor %3944, %3942 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%3946 = torch.aten.cos %3945 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3947 = torch.aten.sin %3945 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%3948 = torch.aten.complex %3946, %3947 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4499 = torch.constant.int 1
%3949 = torch.aten.size.int %3892, %int1_4499 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_4500 = torch.constant.int 0
%3950 = torch.aten.add.int %int0_4500, %3949 : !torch.int, !torch.int -> !torch.int
%int0_4501 = torch.constant.int 0
%int0_4502 = torch.constant.int 0
%int1_4503 = torch.constant.int 1
%3951 = torch.aten.slice.Tensor %3948, %int0_4501, %int0_4502, %3950, %int1_4503 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3951, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4504 = torch.constant.int 1
%int0_4505 = torch.constant.int 0
%int9223372036854775807_4506 = torch.constant.int 9223372036854775807
%int1_4507 = torch.constant.int 1
%3952 = torch.aten.slice.Tensor %3951, %int1_4504, %int0_4505, %int9223372036854775807_4506, %int1_4507 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %3952, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4508 = torch.constant.int 0
%3953 = torch.aten.unsqueeze %3952, %int0_4508 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %3953, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4509 = torch.constant.int 2
%3954 = torch.aten.unsqueeze %3953, %int2_4509 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3954, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4510 = torch.constant.int 3
%int0_4511 = torch.constant.int 0
%int9223372036854775807_4512 = torch.constant.int 9223372036854775807
%int1_4513 = torch.constant.int 1
%3955 = torch.aten.slice.Tensor %3954, %int3_4510, %int0_4511, %int9223372036854775807_4512, %int1_4513 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %3955, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%3956 = torch_c.to_builtin_tensor %3903 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_4514 = arith.constant 1 : index
%dim_4515 = tensor.dim %3956, %c1_4514 : tensor<4x?x8x128xf16>
%3957 = flow.tensor.bitcast %3956 : tensor<4x?x8x128xf16>{%dim_4515} -> tensor<4x?x8x64xcomplex<f16>>{%dim_4515}
%3958 = torch_c.from_builtin_tensor %3957 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %3958, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%3959 = torch.aten.mul.Tensor %3958, %3955 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %3959, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%3960 = torch_c.to_builtin_tensor %3959 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_4516 = arith.constant 1 : index
%dim_4517 = tensor.dim %3960, %c1_4516 : tensor<4x?x8x64xcomplex<f32>>
%3961 = flow.tensor.bitcast %3960 : tensor<4x?x8x64xcomplex<f32>>{%dim_4517} -> tensor<4x?x8x128xf32>{%dim_4517}
%3962 = torch_c.from_builtin_tensor %3961 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %3962, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_4518 = torch.constant.int 5
%3963 = torch.prims.convert_element_type %3962, %int5_4518 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %3963, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_4519 = torch.constant.int 64
%3964 = torch.aten.mul.Scalar %arg2, %int64_4519 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3964, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int36 = torch.constant.int 36
%int1_4520 = torch.constant.int 1
%3965 = torch.aten.add.Scalar %3964, %int36, %int1_4520 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3965, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4521 = torch.constant.int 4
%int32_4522 = torch.constant.int 32
%int8_4523 = torch.constant.int 8
%int128_4524 = torch.constant.int 128
%3966 = torch.prim.ListConstruct %int4_4521, %425, %int32_4522, %int8_4523, %int128_4524 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3967 = torch.aten.view %3963, %3966 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3967, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4525 = torch.constant.int 4
%3968 = torch.aten.mul.int %int4_4525, %425 : !torch.int, !torch.int -> !torch.int
%int32_4526 = torch.constant.int 32
%int8_4527 = torch.constant.int 8
%int128_4528 = torch.constant.int 128
%3969 = torch.prim.ListConstruct %3968, %int32_4526, %int8_4527, %int128_4528 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3970 = torch.aten.view %3967, %3969 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3970, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4529 = torch.constant.int 4
%3971 = torch.aten.mul.int %int4_4529, %425 : !torch.int, !torch.int -> !torch.int
%3972 = torch.prim.ListConstruct %3971 : (!torch.int) -> !torch.list<int>
%3973 = torch.aten.view %3965, %3972 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3973, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_4530 = torch.constant.int 32
%int2_4531 = torch.constant.int 2
%int32_4532 = torch.constant.int 32
%int8_4533 = torch.constant.int 8
%int128_4534 = torch.constant.int 128
%3974 = torch.prim.ListConstruct %416, %int32_4530, %int2_4531, %int32_4532, %int8_4533, %int128_4534 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3975 = torch.aten.view %3807, %3974 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3975, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4535 = torch.constant.int 32
%3976 = torch.aten.mul.int %416, %int32_4535 : !torch.int, !torch.int -> !torch.int
%int2_4536 = torch.constant.int 2
%3977 = torch.aten.mul.int %3976, %int2_4536 : !torch.int, !torch.int -> !torch.int
%int32_4537 = torch.constant.int 32
%int8_4538 = torch.constant.int 8
%int128_4539 = torch.constant.int 128
%3978 = torch.prim.ListConstruct %3977, %int32_4537, %int8_4538, %int128_4539 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3979 = torch.aten.view %3975, %3978 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3979, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%3980 = torch.prim.ListConstruct %3973 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4540 = torch.constant.bool false
%3981 = torch.aten.index_put %3979, %3980, %3970, %false_4540 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3981, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4541 = torch.constant.int 32
%int2_4542 = torch.constant.int 2
%int32_4543 = torch.constant.int 32
%int8_4544 = torch.constant.int 8
%int128_4545 = torch.constant.int 128
%3982 = torch.prim.ListConstruct %416, %int32_4541, %int2_4542, %int32_4543, %int8_4544, %int128_4545 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3983 = torch.aten.view %3981, %3982 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3983, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4546 = torch.constant.int 2097152
%3984 = torch.prim.ListConstruct %416, %int2097152_4546 : (!torch.int, !torch.int) -> !torch.list<int>
%3985 = torch.aten.view %3983, %3984 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %3985, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_4547 = torch.constant.int 32
%int2_4548 = torch.constant.int 2
%int32_4549 = torch.constant.int 32
%int8_4550 = torch.constant.int 8
%int128_4551 = torch.constant.int 128
%3986 = torch.prim.ListConstruct %416, %int32_4547, %int2_4548, %int32_4549, %int8_4550, %int128_4551 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3987 = torch.aten.view %3985, %3986 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %3987, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4552 = torch.constant.int 32
%int8_4553 = torch.constant.int 8
%int128_4554 = torch.constant.int 128
%3988 = torch.prim.ListConstruct %3977, %int32_4552, %int8_4553, %int128_4554 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3989 = torch.aten.view %3987, %3988 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3989, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4555 = torch.constant.int 4
%int32_4556 = torch.constant.int 32
%int8_4557 = torch.constant.int 8
%int128_4558 = torch.constant.int 128
%3990 = torch.prim.ListConstruct %int4_4555, %425, %int32_4556, %int8_4557, %int128_4558 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3991 = torch.aten.view %3905, %3990 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %3991, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4559 = torch.constant.int 4
%3992 = torch.aten.mul.int %int4_4559, %425 : !torch.int, !torch.int -> !torch.int
%int32_4560 = torch.constant.int 32
%int8_4561 = torch.constant.int 8
%int128_4562 = torch.constant.int 128
%3993 = torch.prim.ListConstruct %3992, %int32_4560, %int8_4561, %int128_4562 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%3994 = torch.aten.view %3991, %3993 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %3994, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_4563 = torch.constant.int 1
%int1_4564 = torch.constant.int 1
%3995 = torch.aten.add.Scalar %3965, %int1_4563, %int1_4564 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %3995, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4565 = torch.constant.int 4
%3996 = torch.aten.mul.int %int4_4565, %425 : !torch.int, !torch.int -> !torch.int
%3997 = torch.prim.ListConstruct %3996 : (!torch.int) -> !torch.list<int>
%3998 = torch.aten.view %3995, %3997 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %3998, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%3999 = torch.prim.ListConstruct %3998 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4566 = torch.constant.bool false
%4000 = torch.aten.index_put %3989, %3999, %3994, %false_4566 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4000, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4567 = torch.constant.int 32
%int2_4568 = torch.constant.int 2
%int32_4569 = torch.constant.int 32
%int8_4570 = torch.constant.int 8
%int128_4571 = torch.constant.int 128
%4001 = torch.prim.ListConstruct %416, %int32_4567, %int2_4568, %int32_4569, %int8_4570, %int128_4571 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4002 = torch.aten.view %4000, %4001 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4002, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4572 = torch.constant.int 2097152
%4003 = torch.prim.ListConstruct %416, %int2097152_4572 : (!torch.int, !torch.int) -> !torch.list<int>
%4004 = torch.aten.view %4002, %4003 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4004, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_4573 = torch.constant.int -2
%4005 = torch.aten.unsqueeze %3963, %int-2_4573 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4005, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_4574 = torch.constant.int 4
%int8_4575 = torch.constant.int 8
%int4_4576 = torch.constant.int 4
%int128_4577 = torch.constant.int 128
%4006 = torch.prim.ListConstruct %int4_4574, %3949, %int8_4575, %int4_4576, %int128_4577 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4578 = torch.constant.bool false
%4007 = torch.aten.expand %4005, %4006, %false_4578 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4007, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4579 = torch.constant.int 0
%4008 = torch.aten.clone %4007, %int0_4579 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4008, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4580 = torch.constant.int 4
%int32_4581 = torch.constant.int 32
%int128_4582 = torch.constant.int 128
%4009 = torch.prim.ListConstruct %int4_4580, %3949, %int32_4581, %int128_4582 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4010 = torch.aten._unsafe_view %4008, %4009 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4010, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_4583 = torch.constant.int -2
%4011 = torch.aten.unsqueeze %3905, %int-2_4583 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4011, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_4584 = torch.constant.int 1
%4012 = torch.aten.size.int %3899, %int1_4584 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_4585 = torch.constant.int 4
%int8_4586 = torch.constant.int 8
%int4_4587 = torch.constant.int 4
%int128_4588 = torch.constant.int 128
%4013 = torch.prim.ListConstruct %int4_4585, %4012, %int8_4586, %int4_4587, %int128_4588 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4589 = torch.constant.bool false
%4014 = torch.aten.expand %4011, %4013, %false_4589 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4014, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4590 = torch.constant.int 0
%4015 = torch.aten.clone %4014, %int0_4590 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4015, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4591 = torch.constant.int 4
%int32_4592 = torch.constant.int 32
%int128_4593 = torch.constant.int 128
%4016 = torch.prim.ListConstruct %int4_4591, %4012, %int32_4592, %int128_4593 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4017 = torch.aten._unsafe_view %4015, %4016 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4017, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_4594 = torch.constant.int 1
%int2_4595 = torch.constant.int 2
%4018 = torch.aten.transpose.int %3934, %int1_4594, %int2_4595 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4018, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4596 = torch.constant.int 1
%int2_4597 = torch.constant.int 2
%4019 = torch.aten.transpose.int %4010, %int1_4596, %int2_4597 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4019, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4598 = torch.constant.int 1
%int2_4599 = torch.constant.int 2
%4020 = torch.aten.transpose.int %4017, %int1_4598, %int2_4599 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4020, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_4600 = torch.constant.float 0.000000e+00
%false_4601 = torch.constant.bool false
%none_4602 = torch.constant.none
%4021:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%4018, %4019, %4020, %float0.000000e00_4600, %false_4601, %320, %none_4602) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %4021#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4603 = torch.constant.int 1
%int2_4604 = torch.constant.int 2
%4022 = torch.aten.transpose.int %4021#0, %int1_4603, %int2_4604 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4022, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4605 = torch.constant.int 4
%int4096_4606 = torch.constant.int 4096
%4023 = torch.prim.ListConstruct %int4_4605, %3920, %int4096_4606 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4024 = torch.aten.view %4022, %4023 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4024, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4607 = torch.constant.int -2
%int-1_4608 = torch.constant.int -1
%4025 = torch.aten.transpose.int %167, %int-2_4607, %int-1_4608 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4609 = torch.constant.int 4
%4026 = torch.aten.mul.int %int4_4609, %3920 : !torch.int, !torch.int -> !torch.int
%int4096_4610 = torch.constant.int 4096
%4027 = torch.prim.ListConstruct %4026, %int4096_4610 : (!torch.int, !torch.int) -> !torch.list<int>
%4028 = torch.aten.view %4024, %4027 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4028, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4029 = torch.aten.mm %4028, %4025 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4029, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4611 = torch.constant.int 4
%int4096_4612 = torch.constant.int 4096
%4030 = torch.prim.ListConstruct %int4_4611, %3920, %int4096_4612 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4031 = torch.aten.view %4029, %4030 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4031, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4613 = torch.constant.int 1
%4032 = torch.aten.add.Tensor %3869, %4031, %int1_4613 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4032, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4614 = torch.constant.int 6
%4033 = torch.prims.convert_element_type %4032, %int6_4614 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4033, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4615 = torch.constant.int 2
%4034 = torch.aten.pow.Tensor_Scalar %4033, %int2_4615 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4034, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4616 = torch.constant.int -1
%4035 = torch.prim.ListConstruct %int-1_4616 : (!torch.int) -> !torch.list<int>
%true_4617 = torch.constant.bool true
%none_4618 = torch.constant.none
%4036 = torch.aten.mean.dim %4034, %4035, %true_4617, %none_4618 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4036, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4619 = torch.constant.float 9.9999997473787516E-6
%int1_4620 = torch.constant.int 1
%4037 = torch.aten.add.Scalar %4036, %float9.999990e-06_4619, %int1_4620 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4037, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4038 = torch.aten.rsqrt %4037 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4038, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4039 = torch.aten.mul.Tensor %4033, %4038 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4039, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4040 = torch.aten.mul.Tensor %168, %4039 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4040, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4621 = torch.constant.int 5
%4041 = torch.prims.convert_element_type %4040, %int5_4621 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4041, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4622 = torch.constant.int -2
%int-1_4623 = torch.constant.int -1
%4042 = torch.aten.transpose.int %169, %int-2_4622, %int-1_4623 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4624 = torch.constant.int 4
%4043 = torch.aten.mul.int %int4_4624, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4625 = torch.constant.int 4096
%4044 = torch.prim.ListConstruct %4043, %int4096_4625 : (!torch.int, !torch.int) -> !torch.list<int>
%4045 = torch.aten.view %4041, %4044 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4045, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4046 = torch.aten.mm %4045, %4042 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4046, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4626 = torch.constant.int 4
%int14336_4627 = torch.constant.int 14336
%4047 = torch.prim.ListConstruct %int4_4626, %294, %int14336_4627 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4048 = torch.aten.view %4046, %4047 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4048, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4049 = torch.aten.silu %4048 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4049, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4628 = torch.constant.int -2
%int-1_4629 = torch.constant.int -1
%4050 = torch.aten.transpose.int %170, %int-2_4628, %int-1_4629 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4630 = torch.constant.int 4
%4051 = torch.aten.mul.int %int4_4630, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4631 = torch.constant.int 4096
%4052 = torch.prim.ListConstruct %4051, %int4096_4631 : (!torch.int, !torch.int) -> !torch.list<int>
%4053 = torch.aten.view %4041, %4052 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4053, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4054 = torch.aten.mm %4053, %4050 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4054, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4632 = torch.constant.int 4
%int14336_4633 = torch.constant.int 14336
%4055 = torch.prim.ListConstruct %int4_4632, %294, %int14336_4633 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4056 = torch.aten.view %4054, %4055 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4056, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4057 = torch.aten.mul.Tensor %4049, %4056 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4057, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4634 = torch.constant.int -2
%int-1_4635 = torch.constant.int -1
%4058 = torch.aten.transpose.int %171, %int-2_4634, %int-1_4635 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_4636 = torch.constant.int 1
%4059 = torch.aten.size.int %4048, %int1_4636 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_4637 = torch.constant.int 4
%4060 = torch.aten.mul.int %int4_4637, %4059 : !torch.int, !torch.int -> !torch.int
%int14336_4638 = torch.constant.int 14336
%4061 = torch.prim.ListConstruct %4060, %int14336_4638 : (!torch.int, !torch.int) -> !torch.list<int>
%4062 = torch.aten.view %4057, %4061 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4062, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%4063 = torch.aten.mm %4062, %4058 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4063, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4639 = torch.constant.int 4
%int4096_4640 = torch.constant.int 4096
%4064 = torch.prim.ListConstruct %int4_4639, %4059, %int4096_4640 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4065 = torch.aten.view %4063, %4064 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4065, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4641 = torch.constant.int 1
%4066 = torch.aten.add.Tensor %4032, %4065, %int1_4641 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4066, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4642 = torch.constant.int 6
%4067 = torch.prims.convert_element_type %4066, %int6_4642 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4067, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4643 = torch.constant.int 2
%4068 = torch.aten.pow.Tensor_Scalar %4067, %int2_4643 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4068, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4644 = torch.constant.int -1
%4069 = torch.prim.ListConstruct %int-1_4644 : (!torch.int) -> !torch.list<int>
%true_4645 = torch.constant.bool true
%none_4646 = torch.constant.none
%4070 = torch.aten.mean.dim %4068, %4069, %true_4645, %none_4646 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4070, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4647 = torch.constant.float 9.9999997473787516E-6
%int1_4648 = torch.constant.int 1
%4071 = torch.aten.add.Scalar %4070, %float9.999990e-06_4647, %int1_4648 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4071, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4072 = torch.aten.rsqrt %4071 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4072, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4073 = torch.aten.mul.Tensor %4067, %4072 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4073, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4074 = torch.aten.mul.Tensor %172, %4073 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4074, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4649 = torch.constant.int 5
%4075 = torch.prims.convert_element_type %4074, %int5_4649 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4075, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4650 = torch.constant.int -2
%int-1_4651 = torch.constant.int -1
%4076 = torch.aten.transpose.int %173, %int-2_4650, %int-1_4651 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4652 = torch.constant.int 4
%4077 = torch.aten.mul.int %int4_4652, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4653 = torch.constant.int 4096
%4078 = torch.prim.ListConstruct %4077, %int4096_4653 : (!torch.int, !torch.int) -> !torch.list<int>
%4079 = torch.aten.view %4075, %4078 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4079, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4080 = torch.aten.mm %4079, %4076 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4080, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4654 = torch.constant.int 4
%int4096_4655 = torch.constant.int 4096
%4081 = torch.prim.ListConstruct %int4_4654, %294, %int4096_4655 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4082 = torch.aten.view %4080, %4081 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4082, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4656 = torch.constant.int -2
%int-1_4657 = torch.constant.int -1
%4083 = torch.aten.transpose.int %174, %int-2_4656, %int-1_4657 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4658 = torch.constant.int 4
%4084 = torch.aten.mul.int %int4_4658, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4659 = torch.constant.int 4096
%4085 = torch.prim.ListConstruct %4084, %int4096_4659 : (!torch.int, !torch.int) -> !torch.list<int>
%4086 = torch.aten.view %4075, %4085 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4086, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4087 = torch.aten.mm %4086, %4083 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4087, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4660 = torch.constant.int 4
%int1024_4661 = torch.constant.int 1024
%4088 = torch.prim.ListConstruct %int4_4660, %294, %int1024_4661 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4089 = torch.aten.view %4087, %4088 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4089, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_4662 = torch.constant.int -2
%int-1_4663 = torch.constant.int -1
%4090 = torch.aten.transpose.int %175, %int-2_4662, %int-1_4663 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4664 = torch.constant.int 4
%4091 = torch.aten.mul.int %int4_4664, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4665 = torch.constant.int 4096
%4092 = torch.prim.ListConstruct %4091, %int4096_4665 : (!torch.int, !torch.int) -> !torch.list<int>
%4093 = torch.aten.view %4075, %4092 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4093, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4094 = torch.aten.mm %4093, %4090 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4094, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4666 = torch.constant.int 4
%int1024_4667 = torch.constant.int 1024
%4095 = torch.prim.ListConstruct %int4_4666, %294, %int1024_4667 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4096 = torch.aten.view %4094, %4095 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4096, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_4668 = torch.constant.int 4
%int32_4669 = torch.constant.int 32
%int128_4670 = torch.constant.int 128
%4097 = torch.prim.ListConstruct %int4_4668, %294, %int32_4669, %int128_4670 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4098 = torch.aten.view %4082, %4097 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4098, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4671 = torch.constant.int 4
%int8_4672 = torch.constant.int 8
%int128_4673 = torch.constant.int 128
%4099 = torch.prim.ListConstruct %int4_4671, %294, %int8_4672, %int128_4673 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4100 = torch.aten.view %4089, %4099 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4100, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_4674 = torch.constant.int 4
%int8_4675 = torch.constant.int 8
%int128_4676 = torch.constant.int 128
%4101 = torch.prim.ListConstruct %int4_4674, %294, %int8_4675, %int128_4676 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4102 = torch.aten.view %4096, %4101 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4102, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_4677 = torch.constant.int 131072
%none_4678 = torch.constant.none
%none_4679 = torch.constant.none
%cpu_4680 = torch.constant.device "cpu"
%false_4681 = torch.constant.bool false
%4103 = torch.aten.arange %int131072_4677, %none_4678, %none_4679, %cpu_4680, %false_4681 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4682 = torch.constant.int 0
%int128_4683 = torch.constant.int 128
%int2_4684 = torch.constant.int 2
%none_4685 = torch.constant.none
%none_4686 = torch.constant.none
%cpu_4687 = torch.constant.device "cpu"
%false_4688 = torch.constant.bool false
%4104 = torch.aten.arange.start_step %int0_4682, %int128_4683, %int2_4684, %none_4685, %none_4686, %cpu_4687, %false_4688 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4689 = torch.constant.int 0
%int0_4690 = torch.constant.int 0
%int64_4691 = torch.constant.int 64
%int1_4692 = torch.constant.int 1
%4105 = torch.aten.slice.Tensor %4104, %int0_4689, %int0_4690, %int64_4691, %int1_4692 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4693 = torch.constant.int 6
%4106 = torch.prims.convert_element_type %4105, %int6_4693 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4694 = torch.constant.int 128
%4107 = torch.aten.div.Scalar %4106, %int128_4694 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4695 = torch.constant.float 5.000000e+05
%4108 = torch.aten.pow.Scalar %float5.000000e05_4695, %4107 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4109 = torch.aten.reciprocal %4108 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4696 = torch.constant.float 1.000000e+00
%4110 = torch.aten.mul.Scalar %4109, %float1.000000e00_4696 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4697 = torch.constant.int 131072
%int1_4698 = torch.constant.int 1
%4111 = torch.prim.ListConstruct %int131072_4697, %int1_4698 : (!torch.int, !torch.int) -> !torch.list<int>
%4112 = torch.aten.view %4103, %4111 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4113 = torch.aten.mul.Tensor %4112, %4110 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4114 = torch.aten.cos %4113 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4115 = torch.aten.sin %4113 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4116 = torch.aten.complex %4114, %4115 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4699 = torch.constant.int 1
%4117 = torch.aten.size.int %4082, %int1_4699 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_4700 = torch.constant.int 0
%4118 = torch.aten.add.int %int0_4700, %4117 : !torch.int, !torch.int -> !torch.int
%int0_4701 = torch.constant.int 0
%int0_4702 = torch.constant.int 0
%int1_4703 = torch.constant.int 1
%4119 = torch.aten.slice.Tensor %4116, %int0_4701, %int0_4702, %4118, %int1_4703 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4119, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4704 = torch.constant.int 1
%int0_4705 = torch.constant.int 0
%int9223372036854775807_4706 = torch.constant.int 9223372036854775807
%int1_4707 = torch.constant.int 1
%4120 = torch.aten.slice.Tensor %4119, %int1_4704, %int0_4705, %int9223372036854775807_4706, %int1_4707 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4120, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4708 = torch.constant.int 0
%4121 = torch.aten.unsqueeze %4120, %int0_4708 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4121, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4709 = torch.constant.int 2
%4122 = torch.aten.unsqueeze %4121, %int2_4709 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4122, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4710 = torch.constant.int 3
%int0_4711 = torch.constant.int 0
%int9223372036854775807_4712 = torch.constant.int 9223372036854775807
%int1_4713 = torch.constant.int 1
%4123 = torch.aten.slice.Tensor %4122, %int3_4710, %int0_4711, %int9223372036854775807_4712, %int1_4713 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4123, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4124 = torch_c.to_builtin_tensor %4098 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_4714 = arith.constant 1 : index
%dim_4715 = tensor.dim %4124, %c1_4714 : tensor<4x?x32x128xf16>
%4125 = flow.tensor.bitcast %4124 : tensor<4x?x32x128xf16>{%dim_4715} -> tensor<4x?x32x64xcomplex<f16>>{%dim_4715}
%4126 = torch_c.from_builtin_tensor %4125 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %4126, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%4127 = torch.aten.mul.Tensor %4126, %4123 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %4127, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%4128 = torch_c.to_builtin_tensor %4127 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_4716 = arith.constant 1 : index
%dim_4717 = tensor.dim %4128, %c1_4716 : tensor<4x?x32x64xcomplex<f32>>
%4129 = flow.tensor.bitcast %4128 : tensor<4x?x32x64xcomplex<f32>>{%dim_4717} -> tensor<4x?x32x128xf32>{%dim_4717}
%4130 = torch_c.from_builtin_tensor %4129 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %4130, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_4718 = torch.constant.int 5
%4131 = torch.prims.convert_element_type %4130, %int5_4718 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4131, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_4719 = torch.constant.int 131072
%none_4720 = torch.constant.none
%none_4721 = torch.constant.none
%cpu_4722 = torch.constant.device "cpu"
%false_4723 = torch.constant.bool false
%4132 = torch.aten.arange %int131072_4719, %none_4720, %none_4721, %cpu_4722, %false_4723 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4724 = torch.constant.int 0
%int128_4725 = torch.constant.int 128
%int2_4726 = torch.constant.int 2
%none_4727 = torch.constant.none
%none_4728 = torch.constant.none
%cpu_4729 = torch.constant.device "cpu"
%false_4730 = torch.constant.bool false
%4133 = torch.aten.arange.start_step %int0_4724, %int128_4725, %int2_4726, %none_4727, %none_4728, %cpu_4729, %false_4730 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4731 = torch.constant.int 0
%int0_4732 = torch.constant.int 0
%int64_4733 = torch.constant.int 64
%int1_4734 = torch.constant.int 1
%4134 = torch.aten.slice.Tensor %4133, %int0_4731, %int0_4732, %int64_4733, %int1_4734 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4735 = torch.constant.int 6
%4135 = torch.prims.convert_element_type %4134, %int6_4735 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4736 = torch.constant.int 128
%4136 = torch.aten.div.Scalar %4135, %int128_4736 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4737 = torch.constant.float 5.000000e+05
%4137 = torch.aten.pow.Scalar %float5.000000e05_4737, %4136 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4138 = torch.aten.reciprocal %4137 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4738 = torch.constant.float 1.000000e+00
%4139 = torch.aten.mul.Scalar %4138, %float1.000000e00_4738 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4739 = torch.constant.int 131072
%int1_4740 = torch.constant.int 1
%4140 = torch.prim.ListConstruct %int131072_4739, %int1_4740 : (!torch.int, !torch.int) -> !torch.list<int>
%4141 = torch.aten.view %4132, %4140 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4142 = torch.aten.mul.Tensor %4141, %4139 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4143 = torch.aten.cos %4142 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4144 = torch.aten.sin %4142 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4145 = torch.aten.complex %4143, %4144 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4741 = torch.constant.int 1
%4146 = torch.aten.size.int %4089, %int1_4741 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_4742 = torch.constant.int 0
%4147 = torch.aten.add.int %int0_4742, %4146 : !torch.int, !torch.int -> !torch.int
%int0_4743 = torch.constant.int 0
%int0_4744 = torch.constant.int 0
%int1_4745 = torch.constant.int 1
%4148 = torch.aten.slice.Tensor %4145, %int0_4743, %int0_4744, %4147, %int1_4745 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4148, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4746 = torch.constant.int 1
%int0_4747 = torch.constant.int 0
%int9223372036854775807_4748 = torch.constant.int 9223372036854775807
%int1_4749 = torch.constant.int 1
%4149 = torch.aten.slice.Tensor %4148, %int1_4746, %int0_4747, %int9223372036854775807_4748, %int1_4749 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4149, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4750 = torch.constant.int 0
%4150 = torch.aten.unsqueeze %4149, %int0_4750 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4150, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4751 = torch.constant.int 2
%4151 = torch.aten.unsqueeze %4150, %int2_4751 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4151, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4752 = torch.constant.int 3
%int0_4753 = torch.constant.int 0
%int9223372036854775807_4754 = torch.constant.int 9223372036854775807
%int1_4755 = torch.constant.int 1
%4152 = torch.aten.slice.Tensor %4151, %int3_4752, %int0_4753, %int9223372036854775807_4754, %int1_4755 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4152, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4153 = torch_c.to_builtin_tensor %4100 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_4756 = arith.constant 1 : index
%dim_4757 = tensor.dim %4153, %c1_4756 : tensor<4x?x8x128xf16>
%4154 = flow.tensor.bitcast %4153 : tensor<4x?x8x128xf16>{%dim_4757} -> tensor<4x?x8x64xcomplex<f16>>{%dim_4757}
%4155 = torch_c.from_builtin_tensor %4154 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %4155, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%4156 = torch.aten.mul.Tensor %4155, %4152 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %4156, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%4157 = torch_c.to_builtin_tensor %4156 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_4758 = arith.constant 1 : index
%dim_4759 = tensor.dim %4157, %c1_4758 : tensor<4x?x8x64xcomplex<f32>>
%4158 = flow.tensor.bitcast %4157 : tensor<4x?x8x64xcomplex<f32>>{%dim_4759} -> tensor<4x?x8x128xf32>{%dim_4759}
%4159 = torch_c.from_builtin_tensor %4158 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %4159, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_4760 = torch.constant.int 5
%4160 = torch.prims.convert_element_type %4159, %int5_4760 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4160, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_4761 = torch.constant.int 64
%4161 = torch.aten.mul.Scalar %arg2, %int64_4761 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4161, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int38 = torch.constant.int 38
%int1_4762 = torch.constant.int 1
%4162 = torch.aten.add.Scalar %4161, %int38, %int1_4762 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4162, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4763 = torch.constant.int 4
%int32_4764 = torch.constant.int 32
%int8_4765 = torch.constant.int 8
%int128_4766 = torch.constant.int 128
%4163 = torch.prim.ListConstruct %int4_4763, %425, %int32_4764, %int8_4765, %int128_4766 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4164 = torch.aten.view %4160, %4163 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4164, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4767 = torch.constant.int 4
%4165 = torch.aten.mul.int %int4_4767, %425 : !torch.int, !torch.int -> !torch.int
%int32_4768 = torch.constant.int 32
%int8_4769 = torch.constant.int 8
%int128_4770 = torch.constant.int 128
%4166 = torch.prim.ListConstruct %4165, %int32_4768, %int8_4769, %int128_4770 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4167 = torch.aten.view %4164, %4166 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4167, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4771 = torch.constant.int 4
%4168 = torch.aten.mul.int %int4_4771, %425 : !torch.int, !torch.int -> !torch.int
%4169 = torch.prim.ListConstruct %4168 : (!torch.int) -> !torch.list<int>
%4170 = torch.aten.view %4162, %4169 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4170, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_4772 = torch.constant.int 32
%int2_4773 = torch.constant.int 2
%int32_4774 = torch.constant.int 32
%int8_4775 = torch.constant.int 8
%int128_4776 = torch.constant.int 128
%4171 = torch.prim.ListConstruct %416, %int32_4772, %int2_4773, %int32_4774, %int8_4775, %int128_4776 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4172 = torch.aten.view %4004, %4171 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4172, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4777 = torch.constant.int 32
%4173 = torch.aten.mul.int %416, %int32_4777 : !torch.int, !torch.int -> !torch.int
%int2_4778 = torch.constant.int 2
%4174 = torch.aten.mul.int %4173, %int2_4778 : !torch.int, !torch.int -> !torch.int
%int32_4779 = torch.constant.int 32
%int8_4780 = torch.constant.int 8
%int128_4781 = torch.constant.int 128
%4175 = torch.prim.ListConstruct %4174, %int32_4779, %int8_4780, %int128_4781 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4176 = torch.aten.view %4172, %4175 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4176, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%4177 = torch.prim.ListConstruct %4170 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4782 = torch.constant.bool false
%4178 = torch.aten.index_put %4176, %4177, %4167, %false_4782 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4178, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4783 = torch.constant.int 32
%int2_4784 = torch.constant.int 2
%int32_4785 = torch.constant.int 32
%int8_4786 = torch.constant.int 8
%int128_4787 = torch.constant.int 128
%4179 = torch.prim.ListConstruct %416, %int32_4783, %int2_4784, %int32_4785, %int8_4786, %int128_4787 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4180 = torch.aten.view %4178, %4179 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4180, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4788 = torch.constant.int 2097152
%4181 = torch.prim.ListConstruct %416, %int2097152_4788 : (!torch.int, !torch.int) -> !torch.list<int>
%4182 = torch.aten.view %4180, %4181 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4182, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_4789 = torch.constant.int 32
%int2_4790 = torch.constant.int 2
%int32_4791 = torch.constant.int 32
%int8_4792 = torch.constant.int 8
%int128_4793 = torch.constant.int 128
%4183 = torch.prim.ListConstruct %416, %int32_4789, %int2_4790, %int32_4791, %int8_4792, %int128_4793 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4184 = torch.aten.view %4182, %4183 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4184, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_4794 = torch.constant.int 32
%int8_4795 = torch.constant.int 8
%int128_4796 = torch.constant.int 128
%4185 = torch.prim.ListConstruct %4174, %int32_4794, %int8_4795, %int128_4796 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4186 = torch.aten.view %4184, %4185 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4186, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_4797 = torch.constant.int 4
%int32_4798 = torch.constant.int 32
%int8_4799 = torch.constant.int 8
%int128_4800 = torch.constant.int 128
%4187 = torch.prim.ListConstruct %int4_4797, %425, %int32_4798, %int8_4799, %int128_4800 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4188 = torch.aten.view %4102, %4187 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4188, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_4801 = torch.constant.int 4
%4189 = torch.aten.mul.int %int4_4801, %425 : !torch.int, !torch.int -> !torch.int
%int32_4802 = torch.constant.int 32
%int8_4803 = torch.constant.int 8
%int128_4804 = torch.constant.int 128
%4190 = torch.prim.ListConstruct %4189, %int32_4802, %int8_4803, %int128_4804 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4191 = torch.aten.view %4188, %4190 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4191, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_4805 = torch.constant.int 1
%int1_4806 = torch.constant.int 1
%4192 = torch.aten.add.Scalar %4162, %int1_4805, %int1_4806 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4192, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_4807 = torch.constant.int 4
%4193 = torch.aten.mul.int %int4_4807, %425 : !torch.int, !torch.int -> !torch.int
%4194 = torch.prim.ListConstruct %4193 : (!torch.int) -> !torch.list<int>
%4195 = torch.aten.view %4192, %4194 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4195, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%4196 = torch.prim.ListConstruct %4195 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_4808 = torch.constant.bool false
%4197 = torch.aten.index_put %4186, %4196, %4191, %false_4808 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4197, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_4809 = torch.constant.int 32
%int2_4810 = torch.constant.int 2
%int32_4811 = torch.constant.int 32
%int8_4812 = torch.constant.int 8
%int128_4813 = torch.constant.int 128
%4198 = torch.prim.ListConstruct %416, %int32_4809, %int2_4810, %int32_4811, %int8_4812, %int128_4813 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4199 = torch.aten.view %4197, %4198 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4199, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_4814 = torch.constant.int 2097152
%4200 = torch.prim.ListConstruct %416, %int2097152_4814 : (!torch.int, !torch.int) -> !torch.list<int>
%4201 = torch.aten.view %4199, %4200 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4201, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_4815 = torch.constant.int -2
%4202 = torch.aten.unsqueeze %4160, %int-2_4815 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4202, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_4816 = torch.constant.int 4
%int8_4817 = torch.constant.int 8
%int4_4818 = torch.constant.int 4
%int128_4819 = torch.constant.int 128
%4203 = torch.prim.ListConstruct %int4_4816, %4146, %int8_4817, %int4_4818, %int128_4819 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4820 = torch.constant.bool false
%4204 = torch.aten.expand %4202, %4203, %false_4820 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4204, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4821 = torch.constant.int 0
%4205 = torch.aten.clone %4204, %int0_4821 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4205, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4822 = torch.constant.int 4
%int32_4823 = torch.constant.int 32
%int128_4824 = torch.constant.int 128
%4206 = torch.prim.ListConstruct %int4_4822, %4146, %int32_4823, %int128_4824 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4207 = torch.aten._unsafe_view %4205, %4206 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4207, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_4825 = torch.constant.int -2
%4208 = torch.aten.unsqueeze %4102, %int-2_4825 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4208, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_4826 = torch.constant.int 1
%4209 = torch.aten.size.int %4096, %int1_4826 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_4827 = torch.constant.int 4
%int8_4828 = torch.constant.int 8
%int4_4829 = torch.constant.int 4
%int128_4830 = torch.constant.int 128
%4210 = torch.prim.ListConstruct %int4_4827, %4209, %int8_4828, %int4_4829, %int128_4830 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_4831 = torch.constant.bool false
%4211 = torch.aten.expand %4208, %4210, %false_4831 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4211, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_4832 = torch.constant.int 0
%4212 = torch.aten.clone %4211, %int0_4832 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4212, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_4833 = torch.constant.int 4
%int32_4834 = torch.constant.int 32
%int128_4835 = torch.constant.int 128
%4213 = torch.prim.ListConstruct %int4_4833, %4209, %int32_4834, %int128_4835 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4214 = torch.aten._unsafe_view %4212, %4213 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4214, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_4836 = torch.constant.int 1
%int2_4837 = torch.constant.int 2
%4215 = torch.aten.transpose.int %4131, %int1_4836, %int2_4837 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4215, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4838 = torch.constant.int 1
%int2_4839 = torch.constant.int 2
%4216 = torch.aten.transpose.int %4207, %int1_4838, %int2_4839 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4216, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4840 = torch.constant.int 1
%int2_4841 = torch.constant.int 2
%4217 = torch.aten.transpose.int %4214, %int1_4840, %int2_4841 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4217, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_4842 = torch.constant.float 0.000000e+00
%false_4843 = torch.constant.bool false
%none_4844 = torch.constant.none
%4218:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%4215, %4216, %4217, %float0.000000e00_4842, %false_4843, %320, %none_4844) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %4218#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_4845 = torch.constant.int 1
%int2_4846 = torch.constant.int 2
%4219 = torch.aten.transpose.int %4218#0, %int1_4845, %int2_4846 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4219, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4847 = torch.constant.int 4
%int4096_4848 = torch.constant.int 4096
%4220 = torch.prim.ListConstruct %int4_4847, %4117, %int4096_4848 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4221 = torch.aten.view %4219, %4220 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4221, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4849 = torch.constant.int -2
%int-1_4850 = torch.constant.int -1
%4222 = torch.aten.transpose.int %176, %int-2_4849, %int-1_4850 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4851 = torch.constant.int 4
%4223 = torch.aten.mul.int %int4_4851, %4117 : !torch.int, !torch.int -> !torch.int
%int4096_4852 = torch.constant.int 4096
%4224 = torch.prim.ListConstruct %4223, %int4096_4852 : (!torch.int, !torch.int) -> !torch.list<int>
%4225 = torch.aten.view %4221, %4224 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4225, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4226 = torch.aten.mm %4225, %4222 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4226, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4853 = torch.constant.int 4
%int4096_4854 = torch.constant.int 4096
%4227 = torch.prim.ListConstruct %int4_4853, %4117, %int4096_4854 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4228 = torch.aten.view %4226, %4227 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4228, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4855 = torch.constant.int 1
%4229 = torch.aten.add.Tensor %4066, %4228, %int1_4855 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4229, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4856 = torch.constant.int 6
%4230 = torch.prims.convert_element_type %4229, %int6_4856 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4230, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4857 = torch.constant.int 2
%4231 = torch.aten.pow.Tensor_Scalar %4230, %int2_4857 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4231, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4858 = torch.constant.int -1
%4232 = torch.prim.ListConstruct %int-1_4858 : (!torch.int) -> !torch.list<int>
%true_4859 = torch.constant.bool true
%none_4860 = torch.constant.none
%4233 = torch.aten.mean.dim %4231, %4232, %true_4859, %none_4860 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4233, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4861 = torch.constant.float 9.9999997473787516E-6
%int1_4862 = torch.constant.int 1
%4234 = torch.aten.add.Scalar %4233, %float9.999990e-06_4861, %int1_4862 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4234, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4235 = torch.aten.rsqrt %4234 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4235, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4236 = torch.aten.mul.Tensor %4230, %4235 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4236, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4237 = torch.aten.mul.Tensor %177, %4236 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4237, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4863 = torch.constant.int 5
%4238 = torch.prims.convert_element_type %4237, %int5_4863 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4238, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4864 = torch.constant.int -2
%int-1_4865 = torch.constant.int -1
%4239 = torch.aten.transpose.int %178, %int-2_4864, %int-1_4865 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4866 = torch.constant.int 4
%4240 = torch.aten.mul.int %int4_4866, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4867 = torch.constant.int 4096
%4241 = torch.prim.ListConstruct %4240, %int4096_4867 : (!torch.int, !torch.int) -> !torch.list<int>
%4242 = torch.aten.view %4238, %4241 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4242, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4243 = torch.aten.mm %4242, %4239 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4243, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4868 = torch.constant.int 4
%int14336_4869 = torch.constant.int 14336
%4244 = torch.prim.ListConstruct %int4_4868, %294, %int14336_4869 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4245 = torch.aten.view %4243, %4244 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4245, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4246 = torch.aten.silu %4245 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4246, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4870 = torch.constant.int -2
%int-1_4871 = torch.constant.int -1
%4247 = torch.aten.transpose.int %179, %int-2_4870, %int-1_4871 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_4872 = torch.constant.int 4
%4248 = torch.aten.mul.int %int4_4872, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4873 = torch.constant.int 4096
%4249 = torch.prim.ListConstruct %4248, %int4096_4873 : (!torch.int, !torch.int) -> !torch.list<int>
%4250 = torch.aten.view %4238, %4249 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4250, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4251 = torch.aten.mm %4250, %4247 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4251, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_4874 = torch.constant.int 4
%int14336_4875 = torch.constant.int 14336
%4252 = torch.prim.ListConstruct %int4_4874, %294, %int14336_4875 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4253 = torch.aten.view %4251, %4252 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4253, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4254 = torch.aten.mul.Tensor %4246, %4253 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4254, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_4876 = torch.constant.int -2
%int-1_4877 = torch.constant.int -1
%4255 = torch.aten.transpose.int %180, %int-2_4876, %int-1_4877 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_4878 = torch.constant.int 1
%4256 = torch.aten.size.int %4245, %int1_4878 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_4879 = torch.constant.int 4
%4257 = torch.aten.mul.int %int4_4879, %4256 : !torch.int, !torch.int -> !torch.int
%int14336_4880 = torch.constant.int 14336
%4258 = torch.prim.ListConstruct %4257, %int14336_4880 : (!torch.int, !torch.int) -> !torch.list<int>
%4259 = torch.aten.view %4254, %4258 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4259, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%4260 = torch.aten.mm %4259, %4255 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4260, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4881 = torch.constant.int 4
%int4096_4882 = torch.constant.int 4096
%4261 = torch.prim.ListConstruct %int4_4881, %4256, %int4096_4882 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4262 = torch.aten.view %4260, %4261 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4262, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_4883 = torch.constant.int 1
%4263 = torch.aten.add.Tensor %4229, %4262, %int1_4883 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4263, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_4884 = torch.constant.int 6
%4264 = torch.prims.convert_element_type %4263, %int6_4884 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4264, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_4885 = torch.constant.int 2
%4265 = torch.aten.pow.Tensor_Scalar %4264, %int2_4885 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4265, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_4886 = torch.constant.int -1
%4266 = torch.prim.ListConstruct %int-1_4886 : (!torch.int) -> !torch.list<int>
%true_4887 = torch.constant.bool true
%none_4888 = torch.constant.none
%4267 = torch.aten.mean.dim %4265, %4266, %true_4887, %none_4888 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4267, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_4889 = torch.constant.float 9.9999997473787516E-6
%int1_4890 = torch.constant.int 1
%4268 = torch.aten.add.Scalar %4267, %float9.999990e-06_4889, %int1_4890 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4268, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4269 = torch.aten.rsqrt %4268 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4269, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4270 = torch.aten.mul.Tensor %4264, %4269 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4270, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4271 = torch.aten.mul.Tensor %181, %4270 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4271, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_4891 = torch.constant.int 5
%4272 = torch.prims.convert_element_type %4271, %int5_4891 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4272, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4892 = torch.constant.int -2
%int-1_4893 = torch.constant.int -1
%4273 = torch.aten.transpose.int %182, %int-2_4892, %int-1_4893 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_4894 = torch.constant.int 4
%4274 = torch.aten.mul.int %int4_4894, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4895 = torch.constant.int 4096
%4275 = torch.prim.ListConstruct %4274, %int4096_4895 : (!torch.int, !torch.int) -> !torch.list<int>
%4276 = torch.aten.view %4272, %4275 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4276, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4277 = torch.aten.mm %4276, %4273 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4277, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_4896 = torch.constant.int 4
%int4096_4897 = torch.constant.int 4096
%4278 = torch.prim.ListConstruct %int4_4896, %294, %int4096_4897 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4279 = torch.aten.view %4277, %4278 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4279, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_4898 = torch.constant.int -2
%int-1_4899 = torch.constant.int -1
%4280 = torch.aten.transpose.int %183, %int-2_4898, %int-1_4899 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4900 = torch.constant.int 4
%4281 = torch.aten.mul.int %int4_4900, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4901 = torch.constant.int 4096
%4282 = torch.prim.ListConstruct %4281, %int4096_4901 : (!torch.int, !torch.int) -> !torch.list<int>
%4283 = torch.aten.view %4272, %4282 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4283, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4284 = torch.aten.mm %4283, %4280 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4284, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4902 = torch.constant.int 4
%int1024_4903 = torch.constant.int 1024
%4285 = torch.prim.ListConstruct %int4_4902, %294, %int1024_4903 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4286 = torch.aten.view %4284, %4285 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4286, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_4904 = torch.constant.int -2
%int-1_4905 = torch.constant.int -1
%4287 = torch.aten.transpose.int %184, %int-2_4904, %int-1_4905 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_4906 = torch.constant.int 4
%4288 = torch.aten.mul.int %int4_4906, %294 : !torch.int, !torch.int -> !torch.int
%int4096_4907 = torch.constant.int 4096
%4289 = torch.prim.ListConstruct %4288, %int4096_4907 : (!torch.int, !torch.int) -> !torch.list<int>
%4290 = torch.aten.view %4272, %4289 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4290, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4291 = torch.aten.mm %4290, %4287 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4291, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_4908 = torch.constant.int 4
%int1024_4909 = torch.constant.int 1024
%4292 = torch.prim.ListConstruct %int4_4908, %294, %int1024_4909 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4293 = torch.aten.view %4291, %4292 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4293, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_4910 = torch.constant.int 4
%int32_4911 = torch.constant.int 32
%int128_4912 = torch.constant.int 128
%4294 = torch.prim.ListConstruct %int4_4910, %294, %int32_4911, %int128_4912 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4295 = torch.aten.view %4279, %4294 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4295, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_4913 = torch.constant.int 4
%int8_4914 = torch.constant.int 8
%int128_4915 = torch.constant.int 128
%4296 = torch.prim.ListConstruct %int4_4913, %294, %int8_4914, %int128_4915 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4297 = torch.aten.view %4286, %4296 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4297, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_4916 = torch.constant.int 4
%int8_4917 = torch.constant.int 8
%int128_4918 = torch.constant.int 128
%4298 = torch.prim.ListConstruct %int4_4916, %294, %int8_4917, %int128_4918 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4299 = torch.aten.view %4293, %4298 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4299, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_4919 = torch.constant.int 131072
%none_4920 = torch.constant.none
%none_4921 = torch.constant.none
%cpu_4922 = torch.constant.device "cpu"
%false_4923 = torch.constant.bool false
%4300 = torch.aten.arange %int131072_4919, %none_4920, %none_4921, %cpu_4922, %false_4923 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4924 = torch.constant.int 0
%int128_4925 = torch.constant.int 128
%int2_4926 = torch.constant.int 2
%none_4927 = torch.constant.none
%none_4928 = torch.constant.none
%cpu_4929 = torch.constant.device "cpu"
%false_4930 = torch.constant.bool false
%4301 = torch.aten.arange.start_step %int0_4924, %int128_4925, %int2_4926, %none_4927, %none_4928, %cpu_4929, %false_4930 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4931 = torch.constant.int 0
%int0_4932 = torch.constant.int 0
%int64_4933 = torch.constant.int 64
%int1_4934 = torch.constant.int 1
%4302 = torch.aten.slice.Tensor %4301, %int0_4931, %int0_4932, %int64_4933, %int1_4934 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4935 = torch.constant.int 6
%4303 = torch.prims.convert_element_type %4302, %int6_4935 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4936 = torch.constant.int 128
%4304 = torch.aten.div.Scalar %4303, %int128_4936 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4937 = torch.constant.float 5.000000e+05
%4305 = torch.aten.pow.Scalar %float5.000000e05_4937, %4304 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4306 = torch.aten.reciprocal %4305 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4938 = torch.constant.float 1.000000e+00
%4307 = torch.aten.mul.Scalar %4306, %float1.000000e00_4938 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4939 = torch.constant.int 131072
%int1_4940 = torch.constant.int 1
%4308 = torch.prim.ListConstruct %int131072_4939, %int1_4940 : (!torch.int, !torch.int) -> !torch.list<int>
%4309 = torch.aten.view %4300, %4308 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4310 = torch.aten.mul.Tensor %4309, %4307 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4311 = torch.aten.cos %4310 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4312 = torch.aten.sin %4310 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4313 = torch.aten.complex %4311, %4312 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4941 = torch.constant.int 1
%4314 = torch.aten.size.int %4279, %int1_4941 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_4942 = torch.constant.int 0
%4315 = torch.aten.add.int %int0_4942, %4314 : !torch.int, !torch.int -> !torch.int
%int0_4943 = torch.constant.int 0
%int0_4944 = torch.constant.int 0
%int1_4945 = torch.constant.int 1
%4316 = torch.aten.slice.Tensor %4313, %int0_4943, %int0_4944, %4315, %int1_4945 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4316, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4946 = torch.constant.int 1
%int0_4947 = torch.constant.int 0
%int9223372036854775807_4948 = torch.constant.int 9223372036854775807
%int1_4949 = torch.constant.int 1
%4317 = torch.aten.slice.Tensor %4316, %int1_4946, %int0_4947, %int9223372036854775807_4948, %int1_4949 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4317, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4950 = torch.constant.int 0
%4318 = torch.aten.unsqueeze %4317, %int0_4950 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4318, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4951 = torch.constant.int 2
%4319 = torch.aten.unsqueeze %4318, %int2_4951 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4319, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4952 = torch.constant.int 3
%int0_4953 = torch.constant.int 0
%int9223372036854775807_4954 = torch.constant.int 9223372036854775807
%int1_4955 = torch.constant.int 1
%4320 = torch.aten.slice.Tensor %4319, %int3_4952, %int0_4953, %int9223372036854775807_4954, %int1_4955 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4320, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4321 = torch_c.to_builtin_tensor %4295 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_4956 = arith.constant 1 : index
%dim_4957 = tensor.dim %4321, %c1_4956 : tensor<4x?x32x128xf16>
%4322 = flow.tensor.bitcast %4321 : tensor<4x?x32x128xf16>{%dim_4957} -> tensor<4x?x32x64xcomplex<f16>>{%dim_4957}
%4323 = torch_c.from_builtin_tensor %4322 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %4323, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%4324 = torch.aten.mul.Tensor %4323, %4320 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %4324, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%4325 = torch_c.to_builtin_tensor %4324 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_4958 = arith.constant 1 : index
%dim_4959 = tensor.dim %4325, %c1_4958 : tensor<4x?x32x64xcomplex<f32>>
%4326 = flow.tensor.bitcast %4325 : tensor<4x?x32x64xcomplex<f32>>{%dim_4959} -> tensor<4x?x32x128xf32>{%dim_4959}
%4327 = torch_c.from_builtin_tensor %4326 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %4327, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_4960 = torch.constant.int 5
%4328 = torch.prims.convert_element_type %4327, %int5_4960 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4328, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_4961 = torch.constant.int 131072
%none_4962 = torch.constant.none
%none_4963 = torch.constant.none
%cpu_4964 = torch.constant.device "cpu"
%false_4965 = torch.constant.bool false
%4329 = torch.aten.arange %int131072_4961, %none_4962, %none_4963, %cpu_4964, %false_4965 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_4966 = torch.constant.int 0
%int128_4967 = torch.constant.int 128
%int2_4968 = torch.constant.int 2
%none_4969 = torch.constant.none
%none_4970 = torch.constant.none
%cpu_4971 = torch.constant.device "cpu"
%false_4972 = torch.constant.bool false
%4330 = torch.aten.arange.start_step %int0_4966, %int128_4967, %int2_4968, %none_4969, %none_4970, %cpu_4971, %false_4972 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_4973 = torch.constant.int 0
%int0_4974 = torch.constant.int 0
%int64_4975 = torch.constant.int 64
%int1_4976 = torch.constant.int 1
%4331 = torch.aten.slice.Tensor %4330, %int0_4973, %int0_4974, %int64_4975, %int1_4976 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_4977 = torch.constant.int 6
%4332 = torch.prims.convert_element_type %4331, %int6_4977 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_4978 = torch.constant.int 128
%4333 = torch.aten.div.Scalar %4332, %int128_4978 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_4979 = torch.constant.float 5.000000e+05
%4334 = torch.aten.pow.Scalar %float5.000000e05_4979, %4333 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4335 = torch.aten.reciprocal %4334 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_4980 = torch.constant.float 1.000000e+00
%4336 = torch.aten.mul.Scalar %4335, %float1.000000e00_4980 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_4981 = torch.constant.int 131072
%int1_4982 = torch.constant.int 1
%4337 = torch.prim.ListConstruct %int131072_4981, %int1_4982 : (!torch.int, !torch.int) -> !torch.list<int>
%4338 = torch.aten.view %4329, %4337 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4339 = torch.aten.mul.Tensor %4338, %4336 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4340 = torch.aten.cos %4339 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4341 = torch.aten.sin %4339 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4342 = torch.aten.complex %4340, %4341 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_4983 = torch.constant.int 1
%4343 = torch.aten.size.int %4286, %int1_4983 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_4984 = torch.constant.int 0
%4344 = torch.aten.add.int %int0_4984, %4343 : !torch.int, !torch.int -> !torch.int
%int0_4985 = torch.constant.int 0
%int0_4986 = torch.constant.int 0
%int1_4987 = torch.constant.int 1
%4345 = torch.aten.slice.Tensor %4342, %int0_4985, %int0_4986, %4344, %int1_4987 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4345, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_4988 = torch.constant.int 1
%int0_4989 = torch.constant.int 0
%int9223372036854775807_4990 = torch.constant.int 9223372036854775807
%int1_4991 = torch.constant.int 1
%4346 = torch.aten.slice.Tensor %4345, %int1_4988, %int0_4989, %int9223372036854775807_4990, %int1_4991 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4346, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_4992 = torch.constant.int 0
%4347 = torch.aten.unsqueeze %4346, %int0_4992 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4347, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_4993 = torch.constant.int 2
%4348 = torch.aten.unsqueeze %4347, %int2_4993 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4348, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_4994 = torch.constant.int 3
%int0_4995 = torch.constant.int 0
%int9223372036854775807_4996 = torch.constant.int 9223372036854775807
%int1_4997 = torch.constant.int 1
%4349 = torch.aten.slice.Tensor %4348, %int3_4994, %int0_4995, %int9223372036854775807_4996, %int1_4997 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4349, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4350 = torch_c.to_builtin_tensor %4297 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_4998 = arith.constant 1 : index
%dim_4999 = tensor.dim %4350, %c1_4998 : tensor<4x?x8x128xf16>
%4351 = flow.tensor.bitcast %4350 : tensor<4x?x8x128xf16>{%dim_4999} -> tensor<4x?x8x64xcomplex<f16>>{%dim_4999}
%4352 = torch_c.from_builtin_tensor %4351 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %4352, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%4353 = torch.aten.mul.Tensor %4352, %4349 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %4353, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%4354 = torch_c.to_builtin_tensor %4353 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_5000 = arith.constant 1 : index
%dim_5001 = tensor.dim %4354, %c1_5000 : tensor<4x?x8x64xcomplex<f32>>
%4355 = flow.tensor.bitcast %4354 : tensor<4x?x8x64xcomplex<f32>>{%dim_5001} -> tensor<4x?x8x128xf32>{%dim_5001}
%4356 = torch_c.from_builtin_tensor %4355 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %4356, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_5002 = torch.constant.int 5
%4357 = torch.prims.convert_element_type %4356, %int5_5002 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4357, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_5003 = torch.constant.int 64
%4358 = torch.aten.mul.Scalar %arg2, %int64_5003 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4358, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int40 = torch.constant.int 40
%int1_5004 = torch.constant.int 1
%4359 = torch.aten.add.Scalar %4358, %int40, %int1_5004 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4359, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5005 = torch.constant.int 4
%int32_5006 = torch.constant.int 32
%int8_5007 = torch.constant.int 8
%int128_5008 = torch.constant.int 128
%4360 = torch.prim.ListConstruct %int4_5005, %425, %int32_5006, %int8_5007, %int128_5008 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4361 = torch.aten.view %4357, %4360 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4361, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5009 = torch.constant.int 4
%4362 = torch.aten.mul.int %int4_5009, %425 : !torch.int, !torch.int -> !torch.int
%int32_5010 = torch.constant.int 32
%int8_5011 = torch.constant.int 8
%int128_5012 = torch.constant.int 128
%4363 = torch.prim.ListConstruct %4362, %int32_5010, %int8_5011, %int128_5012 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4364 = torch.aten.view %4361, %4363 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4364, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5013 = torch.constant.int 4
%4365 = torch.aten.mul.int %int4_5013, %425 : !torch.int, !torch.int -> !torch.int
%4366 = torch.prim.ListConstruct %4365 : (!torch.int) -> !torch.list<int>
%4367 = torch.aten.view %4359, %4366 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4367, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_5014 = torch.constant.int 32
%int2_5015 = torch.constant.int 2
%int32_5016 = torch.constant.int 32
%int8_5017 = torch.constant.int 8
%int128_5018 = torch.constant.int 128
%4368 = torch.prim.ListConstruct %416, %int32_5014, %int2_5015, %int32_5016, %int8_5017, %int128_5018 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4369 = torch.aten.view %4201, %4368 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4369, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5019 = torch.constant.int 32
%4370 = torch.aten.mul.int %416, %int32_5019 : !torch.int, !torch.int -> !torch.int
%int2_5020 = torch.constant.int 2
%4371 = torch.aten.mul.int %4370, %int2_5020 : !torch.int, !torch.int -> !torch.int
%int32_5021 = torch.constant.int 32
%int8_5022 = torch.constant.int 8
%int128_5023 = torch.constant.int 128
%4372 = torch.prim.ListConstruct %4371, %int32_5021, %int8_5022, %int128_5023 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4373 = torch.aten.view %4369, %4372 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4373, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%4374 = torch.prim.ListConstruct %4367 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5024 = torch.constant.bool false
%4375 = torch.aten.index_put %4373, %4374, %4364, %false_5024 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4375, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5025 = torch.constant.int 32
%int2_5026 = torch.constant.int 2
%int32_5027 = torch.constant.int 32
%int8_5028 = torch.constant.int 8
%int128_5029 = torch.constant.int 128
%4376 = torch.prim.ListConstruct %416, %int32_5025, %int2_5026, %int32_5027, %int8_5028, %int128_5029 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4377 = torch.aten.view %4375, %4376 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4377, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5030 = torch.constant.int 2097152
%4378 = torch.prim.ListConstruct %416, %int2097152_5030 : (!torch.int, !torch.int) -> !torch.list<int>
%4379 = torch.aten.view %4377, %4378 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4379, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_5031 = torch.constant.int 32
%int2_5032 = torch.constant.int 2
%int32_5033 = torch.constant.int 32
%int8_5034 = torch.constant.int 8
%int128_5035 = torch.constant.int 128
%4380 = torch.prim.ListConstruct %416, %int32_5031, %int2_5032, %int32_5033, %int8_5034, %int128_5035 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4381 = torch.aten.view %4379, %4380 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4381, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5036 = torch.constant.int 32
%int8_5037 = torch.constant.int 8
%int128_5038 = torch.constant.int 128
%4382 = torch.prim.ListConstruct %4371, %int32_5036, %int8_5037, %int128_5038 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4383 = torch.aten.view %4381, %4382 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4383, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5039 = torch.constant.int 4
%int32_5040 = torch.constant.int 32
%int8_5041 = torch.constant.int 8
%int128_5042 = torch.constant.int 128
%4384 = torch.prim.ListConstruct %int4_5039, %425, %int32_5040, %int8_5041, %int128_5042 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4385 = torch.aten.view %4299, %4384 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4385, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5043 = torch.constant.int 4
%4386 = torch.aten.mul.int %int4_5043, %425 : !torch.int, !torch.int -> !torch.int
%int32_5044 = torch.constant.int 32
%int8_5045 = torch.constant.int 8
%int128_5046 = torch.constant.int 128
%4387 = torch.prim.ListConstruct %4386, %int32_5044, %int8_5045, %int128_5046 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4388 = torch.aten.view %4385, %4387 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4388, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_5047 = torch.constant.int 1
%int1_5048 = torch.constant.int 1
%4389 = torch.aten.add.Scalar %4359, %int1_5047, %int1_5048 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4389, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5049 = torch.constant.int 4
%4390 = torch.aten.mul.int %int4_5049, %425 : !torch.int, !torch.int -> !torch.int
%4391 = torch.prim.ListConstruct %4390 : (!torch.int) -> !torch.list<int>
%4392 = torch.aten.view %4389, %4391 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4392, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%4393 = torch.prim.ListConstruct %4392 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5050 = torch.constant.bool false
%4394 = torch.aten.index_put %4383, %4393, %4388, %false_5050 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4394, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5051 = torch.constant.int 32
%int2_5052 = torch.constant.int 2
%int32_5053 = torch.constant.int 32
%int8_5054 = torch.constant.int 8
%int128_5055 = torch.constant.int 128
%4395 = torch.prim.ListConstruct %416, %int32_5051, %int2_5052, %int32_5053, %int8_5054, %int128_5055 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4396 = torch.aten.view %4394, %4395 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4396, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5056 = torch.constant.int 2097152
%4397 = torch.prim.ListConstruct %416, %int2097152_5056 : (!torch.int, !torch.int) -> !torch.list<int>
%4398 = torch.aten.view %4396, %4397 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4398, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_5057 = torch.constant.int -2
%4399 = torch.aten.unsqueeze %4357, %int-2_5057 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4399, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_5058 = torch.constant.int 4
%int8_5059 = torch.constant.int 8
%int4_5060 = torch.constant.int 4
%int128_5061 = torch.constant.int 128
%4400 = torch.prim.ListConstruct %int4_5058, %4343, %int8_5059, %int4_5060, %int128_5061 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5062 = torch.constant.bool false
%4401 = torch.aten.expand %4399, %4400, %false_5062 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4401, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5063 = torch.constant.int 0
%4402 = torch.aten.clone %4401, %int0_5063 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4402, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5064 = torch.constant.int 4
%int32_5065 = torch.constant.int 32
%int128_5066 = torch.constant.int 128
%4403 = torch.prim.ListConstruct %int4_5064, %4343, %int32_5065, %int128_5066 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4404 = torch.aten._unsafe_view %4402, %4403 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4404, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_5067 = torch.constant.int -2
%4405 = torch.aten.unsqueeze %4299, %int-2_5067 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4405, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_5068 = torch.constant.int 1
%4406 = torch.aten.size.int %4293, %int1_5068 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_5069 = torch.constant.int 4
%int8_5070 = torch.constant.int 8
%int4_5071 = torch.constant.int 4
%int128_5072 = torch.constant.int 128
%4407 = torch.prim.ListConstruct %int4_5069, %4406, %int8_5070, %int4_5071, %int128_5072 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5073 = torch.constant.bool false
%4408 = torch.aten.expand %4405, %4407, %false_5073 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4408, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5074 = torch.constant.int 0
%4409 = torch.aten.clone %4408, %int0_5074 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4409, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5075 = torch.constant.int 4
%int32_5076 = torch.constant.int 32
%int128_5077 = torch.constant.int 128
%4410 = torch.prim.ListConstruct %int4_5075, %4406, %int32_5076, %int128_5077 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4411 = torch.aten._unsafe_view %4409, %4410 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4411, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_5078 = torch.constant.int 1
%int2_5079 = torch.constant.int 2
%4412 = torch.aten.transpose.int %4328, %int1_5078, %int2_5079 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4412, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5080 = torch.constant.int 1
%int2_5081 = torch.constant.int 2
%4413 = torch.aten.transpose.int %4404, %int1_5080, %int2_5081 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4413, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5082 = torch.constant.int 1
%int2_5083 = torch.constant.int 2
%4414 = torch.aten.transpose.int %4411, %int1_5082, %int2_5083 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4414, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_5084 = torch.constant.float 0.000000e+00
%false_5085 = torch.constant.bool false
%none_5086 = torch.constant.none
%4415:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%4412, %4413, %4414, %float0.000000e00_5084, %false_5085, %320, %none_5086) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %4415#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5087 = torch.constant.int 1
%int2_5088 = torch.constant.int 2
%4416 = torch.aten.transpose.int %4415#0, %int1_5087, %int2_5088 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4416, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5089 = torch.constant.int 4
%int4096_5090 = torch.constant.int 4096
%4417 = torch.prim.ListConstruct %int4_5089, %4314, %int4096_5090 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4418 = torch.aten.view %4416, %4417 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4418, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5091 = torch.constant.int -2
%int-1_5092 = torch.constant.int -1
%4419 = torch.aten.transpose.int %185, %int-2_5091, %int-1_5092 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5093 = torch.constant.int 4
%4420 = torch.aten.mul.int %int4_5093, %4314 : !torch.int, !torch.int -> !torch.int
%int4096_5094 = torch.constant.int 4096
%4421 = torch.prim.ListConstruct %4420, %int4096_5094 : (!torch.int, !torch.int) -> !torch.list<int>
%4422 = torch.aten.view %4418, %4421 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4422, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4423 = torch.aten.mm %4422, %4419 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4423, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5095 = torch.constant.int 4
%int4096_5096 = torch.constant.int 4096
%4424 = torch.prim.ListConstruct %int4_5095, %4314, %int4096_5096 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4425 = torch.aten.view %4423, %4424 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4425, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5097 = torch.constant.int 1
%4426 = torch.aten.add.Tensor %4263, %4425, %int1_5097 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4426, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5098 = torch.constant.int 6
%4427 = torch.prims.convert_element_type %4426, %int6_5098 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4427, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5099 = torch.constant.int 2
%4428 = torch.aten.pow.Tensor_Scalar %4427, %int2_5099 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4428, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5100 = torch.constant.int -1
%4429 = torch.prim.ListConstruct %int-1_5100 : (!torch.int) -> !torch.list<int>
%true_5101 = torch.constant.bool true
%none_5102 = torch.constant.none
%4430 = torch.aten.mean.dim %4428, %4429, %true_5101, %none_5102 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4430, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5103 = torch.constant.float 9.9999997473787516E-6
%int1_5104 = torch.constant.int 1
%4431 = torch.aten.add.Scalar %4430, %float9.999990e-06_5103, %int1_5104 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4431, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4432 = torch.aten.rsqrt %4431 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4432, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4433 = torch.aten.mul.Tensor %4427, %4432 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4433, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4434 = torch.aten.mul.Tensor %186, %4433 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4434, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5105 = torch.constant.int 5
%4435 = torch.prims.convert_element_type %4434, %int5_5105 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4435, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5106 = torch.constant.int -2
%int-1_5107 = torch.constant.int -1
%4436 = torch.aten.transpose.int %187, %int-2_5106, %int-1_5107 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5108 = torch.constant.int 4
%4437 = torch.aten.mul.int %int4_5108, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5109 = torch.constant.int 4096
%4438 = torch.prim.ListConstruct %4437, %int4096_5109 : (!torch.int, !torch.int) -> !torch.list<int>
%4439 = torch.aten.view %4435, %4438 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4439, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4440 = torch.aten.mm %4439, %4436 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4440, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5110 = torch.constant.int 4
%int14336_5111 = torch.constant.int 14336
%4441 = torch.prim.ListConstruct %int4_5110, %294, %int14336_5111 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4442 = torch.aten.view %4440, %4441 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4442, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4443 = torch.aten.silu %4442 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4443, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5112 = torch.constant.int -2
%int-1_5113 = torch.constant.int -1
%4444 = torch.aten.transpose.int %188, %int-2_5112, %int-1_5113 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5114 = torch.constant.int 4
%4445 = torch.aten.mul.int %int4_5114, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5115 = torch.constant.int 4096
%4446 = torch.prim.ListConstruct %4445, %int4096_5115 : (!torch.int, !torch.int) -> !torch.list<int>
%4447 = torch.aten.view %4435, %4446 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4447, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4448 = torch.aten.mm %4447, %4444 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4448, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5116 = torch.constant.int 4
%int14336_5117 = torch.constant.int 14336
%4449 = torch.prim.ListConstruct %int4_5116, %294, %int14336_5117 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4450 = torch.aten.view %4448, %4449 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4450, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4451 = torch.aten.mul.Tensor %4443, %4450 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4451, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5118 = torch.constant.int -2
%int-1_5119 = torch.constant.int -1
%4452 = torch.aten.transpose.int %189, %int-2_5118, %int-1_5119 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_5120 = torch.constant.int 1
%4453 = torch.aten.size.int %4442, %int1_5120 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_5121 = torch.constant.int 4
%4454 = torch.aten.mul.int %int4_5121, %4453 : !torch.int, !torch.int -> !torch.int
%int14336_5122 = torch.constant.int 14336
%4455 = torch.prim.ListConstruct %4454, %int14336_5122 : (!torch.int, !torch.int) -> !torch.list<int>
%4456 = torch.aten.view %4451, %4455 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4456, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%4457 = torch.aten.mm %4456, %4452 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4457, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5123 = torch.constant.int 4
%int4096_5124 = torch.constant.int 4096
%4458 = torch.prim.ListConstruct %int4_5123, %4453, %int4096_5124 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4459 = torch.aten.view %4457, %4458 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4459, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5125 = torch.constant.int 1
%4460 = torch.aten.add.Tensor %4426, %4459, %int1_5125 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4460, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5126 = torch.constant.int 6
%4461 = torch.prims.convert_element_type %4460, %int6_5126 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4461, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5127 = torch.constant.int 2
%4462 = torch.aten.pow.Tensor_Scalar %4461, %int2_5127 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4462, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5128 = torch.constant.int -1
%4463 = torch.prim.ListConstruct %int-1_5128 : (!torch.int) -> !torch.list<int>
%true_5129 = torch.constant.bool true
%none_5130 = torch.constant.none
%4464 = torch.aten.mean.dim %4462, %4463, %true_5129, %none_5130 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4464, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5131 = torch.constant.float 9.9999997473787516E-6
%int1_5132 = torch.constant.int 1
%4465 = torch.aten.add.Scalar %4464, %float9.999990e-06_5131, %int1_5132 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4465, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4466 = torch.aten.rsqrt %4465 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4466, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4467 = torch.aten.mul.Tensor %4461, %4466 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4467, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4468 = torch.aten.mul.Tensor %190, %4467 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4468, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5133 = torch.constant.int 5
%4469 = torch.prims.convert_element_type %4468, %int5_5133 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4469, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5134 = torch.constant.int -2
%int-1_5135 = torch.constant.int -1
%4470 = torch.aten.transpose.int %191, %int-2_5134, %int-1_5135 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5136 = torch.constant.int 4
%4471 = torch.aten.mul.int %int4_5136, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5137 = torch.constant.int 4096
%4472 = torch.prim.ListConstruct %4471, %int4096_5137 : (!torch.int, !torch.int) -> !torch.list<int>
%4473 = torch.aten.view %4469, %4472 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4473, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4474 = torch.aten.mm %4473, %4470 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4474, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5138 = torch.constant.int 4
%int4096_5139 = torch.constant.int 4096
%4475 = torch.prim.ListConstruct %int4_5138, %294, %int4096_5139 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4476 = torch.aten.view %4474, %4475 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4476, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5140 = torch.constant.int -2
%int-1_5141 = torch.constant.int -1
%4477 = torch.aten.transpose.int %192, %int-2_5140, %int-1_5141 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5142 = torch.constant.int 4
%4478 = torch.aten.mul.int %int4_5142, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5143 = torch.constant.int 4096
%4479 = torch.prim.ListConstruct %4478, %int4096_5143 : (!torch.int, !torch.int) -> !torch.list<int>
%4480 = torch.aten.view %4469, %4479 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4480, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4481 = torch.aten.mm %4480, %4477 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4481, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5144 = torch.constant.int 4
%int1024_5145 = torch.constant.int 1024
%4482 = torch.prim.ListConstruct %int4_5144, %294, %int1024_5145 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4483 = torch.aten.view %4481, %4482 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4483, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_5146 = torch.constant.int -2
%int-1_5147 = torch.constant.int -1
%4484 = torch.aten.transpose.int %193, %int-2_5146, %int-1_5147 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5148 = torch.constant.int 4
%4485 = torch.aten.mul.int %int4_5148, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5149 = torch.constant.int 4096
%4486 = torch.prim.ListConstruct %4485, %int4096_5149 : (!torch.int, !torch.int) -> !torch.list<int>
%4487 = torch.aten.view %4469, %4486 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4487, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4488 = torch.aten.mm %4487, %4484 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4488, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5150 = torch.constant.int 4
%int1024_5151 = torch.constant.int 1024
%4489 = torch.prim.ListConstruct %int4_5150, %294, %int1024_5151 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4490 = torch.aten.view %4488, %4489 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4490, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_5152 = torch.constant.int 4
%int32_5153 = torch.constant.int 32
%int128_5154 = torch.constant.int 128
%4491 = torch.prim.ListConstruct %int4_5152, %294, %int32_5153, %int128_5154 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4492 = torch.aten.view %4476, %4491 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4492, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5155 = torch.constant.int 4
%int8_5156 = torch.constant.int 8
%int128_5157 = torch.constant.int 128
%4493 = torch.prim.ListConstruct %int4_5155, %294, %int8_5156, %int128_5157 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4494 = torch.aten.view %4483, %4493 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4494, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_5158 = torch.constant.int 4
%int8_5159 = torch.constant.int 8
%int128_5160 = torch.constant.int 128
%4495 = torch.prim.ListConstruct %int4_5158, %294, %int8_5159, %int128_5160 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4496 = torch.aten.view %4490, %4495 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4496, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_5161 = torch.constant.int 131072
%none_5162 = torch.constant.none
%none_5163 = torch.constant.none
%cpu_5164 = torch.constant.device "cpu"
%false_5165 = torch.constant.bool false
%4497 = torch.aten.arange %int131072_5161, %none_5162, %none_5163, %cpu_5164, %false_5165 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5166 = torch.constant.int 0
%int128_5167 = torch.constant.int 128
%int2_5168 = torch.constant.int 2
%none_5169 = torch.constant.none
%none_5170 = torch.constant.none
%cpu_5171 = torch.constant.device "cpu"
%false_5172 = torch.constant.bool false
%4498 = torch.aten.arange.start_step %int0_5166, %int128_5167, %int2_5168, %none_5169, %none_5170, %cpu_5171, %false_5172 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5173 = torch.constant.int 0
%int0_5174 = torch.constant.int 0
%int64_5175 = torch.constant.int 64
%int1_5176 = torch.constant.int 1
%4499 = torch.aten.slice.Tensor %4498, %int0_5173, %int0_5174, %int64_5175, %int1_5176 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5177 = torch.constant.int 6
%4500 = torch.prims.convert_element_type %4499, %int6_5177 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5178 = torch.constant.int 128
%4501 = torch.aten.div.Scalar %4500, %int128_5178 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5179 = torch.constant.float 5.000000e+05
%4502 = torch.aten.pow.Scalar %float5.000000e05_5179, %4501 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4503 = torch.aten.reciprocal %4502 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5180 = torch.constant.float 1.000000e+00
%4504 = torch.aten.mul.Scalar %4503, %float1.000000e00_5180 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5181 = torch.constant.int 131072
%int1_5182 = torch.constant.int 1
%4505 = torch.prim.ListConstruct %int131072_5181, %int1_5182 : (!torch.int, !torch.int) -> !torch.list<int>
%4506 = torch.aten.view %4497, %4505 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4507 = torch.aten.mul.Tensor %4506, %4504 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4508 = torch.aten.cos %4507 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4509 = torch.aten.sin %4507 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4510 = torch.aten.complex %4508, %4509 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5183 = torch.constant.int 1
%4511 = torch.aten.size.int %4476, %int1_5183 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_5184 = torch.constant.int 0
%4512 = torch.aten.add.int %int0_5184, %4511 : !torch.int, !torch.int -> !torch.int
%int0_5185 = torch.constant.int 0
%int0_5186 = torch.constant.int 0
%int1_5187 = torch.constant.int 1
%4513 = torch.aten.slice.Tensor %4510, %int0_5185, %int0_5186, %4512, %int1_5187 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4513, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5188 = torch.constant.int 1
%int0_5189 = torch.constant.int 0
%int9223372036854775807_5190 = torch.constant.int 9223372036854775807
%int1_5191 = torch.constant.int 1
%4514 = torch.aten.slice.Tensor %4513, %int1_5188, %int0_5189, %int9223372036854775807_5190, %int1_5191 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4514, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5192 = torch.constant.int 0
%4515 = torch.aten.unsqueeze %4514, %int0_5192 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4515, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5193 = torch.constant.int 2
%4516 = torch.aten.unsqueeze %4515, %int2_5193 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4516, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5194 = torch.constant.int 3
%int0_5195 = torch.constant.int 0
%int9223372036854775807_5196 = torch.constant.int 9223372036854775807
%int1_5197 = torch.constant.int 1
%4517 = torch.aten.slice.Tensor %4516, %int3_5194, %int0_5195, %int9223372036854775807_5196, %int1_5197 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4517, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4518 = torch_c.to_builtin_tensor %4492 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_5198 = arith.constant 1 : index
%dim_5199 = tensor.dim %4518, %c1_5198 : tensor<4x?x32x128xf16>
%4519 = flow.tensor.bitcast %4518 : tensor<4x?x32x128xf16>{%dim_5199} -> tensor<4x?x32x64xcomplex<f16>>{%dim_5199}
%4520 = torch_c.from_builtin_tensor %4519 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %4520, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%4521 = torch.aten.mul.Tensor %4520, %4517 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %4521, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%4522 = torch_c.to_builtin_tensor %4521 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_5200 = arith.constant 1 : index
%dim_5201 = tensor.dim %4522, %c1_5200 : tensor<4x?x32x64xcomplex<f32>>
%4523 = flow.tensor.bitcast %4522 : tensor<4x?x32x64xcomplex<f32>>{%dim_5201} -> tensor<4x?x32x128xf32>{%dim_5201}
%4524 = torch_c.from_builtin_tensor %4523 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %4524, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_5202 = torch.constant.int 5
%4525 = torch.prims.convert_element_type %4524, %int5_5202 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4525, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_5203 = torch.constant.int 131072
%none_5204 = torch.constant.none
%none_5205 = torch.constant.none
%cpu_5206 = torch.constant.device "cpu"
%false_5207 = torch.constant.bool false
%4526 = torch.aten.arange %int131072_5203, %none_5204, %none_5205, %cpu_5206, %false_5207 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5208 = torch.constant.int 0
%int128_5209 = torch.constant.int 128
%int2_5210 = torch.constant.int 2
%none_5211 = torch.constant.none
%none_5212 = torch.constant.none
%cpu_5213 = torch.constant.device "cpu"
%false_5214 = torch.constant.bool false
%4527 = torch.aten.arange.start_step %int0_5208, %int128_5209, %int2_5210, %none_5211, %none_5212, %cpu_5213, %false_5214 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5215 = torch.constant.int 0
%int0_5216 = torch.constant.int 0
%int64_5217 = torch.constant.int 64
%int1_5218 = torch.constant.int 1
%4528 = torch.aten.slice.Tensor %4527, %int0_5215, %int0_5216, %int64_5217, %int1_5218 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5219 = torch.constant.int 6
%4529 = torch.prims.convert_element_type %4528, %int6_5219 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5220 = torch.constant.int 128
%4530 = torch.aten.div.Scalar %4529, %int128_5220 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5221 = torch.constant.float 5.000000e+05
%4531 = torch.aten.pow.Scalar %float5.000000e05_5221, %4530 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4532 = torch.aten.reciprocal %4531 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5222 = torch.constant.float 1.000000e+00
%4533 = torch.aten.mul.Scalar %4532, %float1.000000e00_5222 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5223 = torch.constant.int 131072
%int1_5224 = torch.constant.int 1
%4534 = torch.prim.ListConstruct %int131072_5223, %int1_5224 : (!torch.int, !torch.int) -> !torch.list<int>
%4535 = torch.aten.view %4526, %4534 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4536 = torch.aten.mul.Tensor %4535, %4533 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4537 = torch.aten.cos %4536 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4538 = torch.aten.sin %4536 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4539 = torch.aten.complex %4537, %4538 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5225 = torch.constant.int 1
%4540 = torch.aten.size.int %4483, %int1_5225 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_5226 = torch.constant.int 0
%4541 = torch.aten.add.int %int0_5226, %4540 : !torch.int, !torch.int -> !torch.int
%int0_5227 = torch.constant.int 0
%int0_5228 = torch.constant.int 0
%int1_5229 = torch.constant.int 1
%4542 = torch.aten.slice.Tensor %4539, %int0_5227, %int0_5228, %4541, %int1_5229 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4542, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5230 = torch.constant.int 1
%int0_5231 = torch.constant.int 0
%int9223372036854775807_5232 = torch.constant.int 9223372036854775807
%int1_5233 = torch.constant.int 1
%4543 = torch.aten.slice.Tensor %4542, %int1_5230, %int0_5231, %int9223372036854775807_5232, %int1_5233 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4543, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5234 = torch.constant.int 0
%4544 = torch.aten.unsqueeze %4543, %int0_5234 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4544, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5235 = torch.constant.int 2
%4545 = torch.aten.unsqueeze %4544, %int2_5235 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4545, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5236 = torch.constant.int 3
%int0_5237 = torch.constant.int 0
%int9223372036854775807_5238 = torch.constant.int 9223372036854775807
%int1_5239 = torch.constant.int 1
%4546 = torch.aten.slice.Tensor %4545, %int3_5236, %int0_5237, %int9223372036854775807_5238, %int1_5239 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4546, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4547 = torch_c.to_builtin_tensor %4494 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_5240 = arith.constant 1 : index
%dim_5241 = tensor.dim %4547, %c1_5240 : tensor<4x?x8x128xf16>
%4548 = flow.tensor.bitcast %4547 : tensor<4x?x8x128xf16>{%dim_5241} -> tensor<4x?x8x64xcomplex<f16>>{%dim_5241}
%4549 = torch_c.from_builtin_tensor %4548 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %4549, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%4550 = torch.aten.mul.Tensor %4549, %4546 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %4550, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%4551 = torch_c.to_builtin_tensor %4550 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_5242 = arith.constant 1 : index
%dim_5243 = tensor.dim %4551, %c1_5242 : tensor<4x?x8x64xcomplex<f32>>
%4552 = flow.tensor.bitcast %4551 : tensor<4x?x8x64xcomplex<f32>>{%dim_5243} -> tensor<4x?x8x128xf32>{%dim_5243}
%4553 = torch_c.from_builtin_tensor %4552 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %4553, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_5244 = torch.constant.int 5
%4554 = torch.prims.convert_element_type %4553, %int5_5244 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4554, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_5245 = torch.constant.int 64
%4555 = torch.aten.mul.Scalar %arg2, %int64_5245 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4555, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int42 = torch.constant.int 42
%int1_5246 = torch.constant.int 1
%4556 = torch.aten.add.Scalar %4555, %int42, %int1_5246 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4556, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5247 = torch.constant.int 4
%int32_5248 = torch.constant.int 32
%int8_5249 = torch.constant.int 8
%int128_5250 = torch.constant.int 128
%4557 = torch.prim.ListConstruct %int4_5247, %425, %int32_5248, %int8_5249, %int128_5250 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4558 = torch.aten.view %4554, %4557 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4558, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5251 = torch.constant.int 4
%4559 = torch.aten.mul.int %int4_5251, %425 : !torch.int, !torch.int -> !torch.int
%int32_5252 = torch.constant.int 32
%int8_5253 = torch.constant.int 8
%int128_5254 = torch.constant.int 128
%4560 = torch.prim.ListConstruct %4559, %int32_5252, %int8_5253, %int128_5254 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4561 = torch.aten.view %4558, %4560 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4561, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5255 = torch.constant.int 4
%4562 = torch.aten.mul.int %int4_5255, %425 : !torch.int, !torch.int -> !torch.int
%4563 = torch.prim.ListConstruct %4562 : (!torch.int) -> !torch.list<int>
%4564 = torch.aten.view %4556, %4563 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4564, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_5256 = torch.constant.int 32
%int2_5257 = torch.constant.int 2
%int32_5258 = torch.constant.int 32
%int8_5259 = torch.constant.int 8
%int128_5260 = torch.constant.int 128
%4565 = torch.prim.ListConstruct %416, %int32_5256, %int2_5257, %int32_5258, %int8_5259, %int128_5260 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4566 = torch.aten.view %4398, %4565 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4566, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5261 = torch.constant.int 32
%4567 = torch.aten.mul.int %416, %int32_5261 : !torch.int, !torch.int -> !torch.int
%int2_5262 = torch.constant.int 2
%4568 = torch.aten.mul.int %4567, %int2_5262 : !torch.int, !torch.int -> !torch.int
%int32_5263 = torch.constant.int 32
%int8_5264 = torch.constant.int 8
%int128_5265 = torch.constant.int 128
%4569 = torch.prim.ListConstruct %4568, %int32_5263, %int8_5264, %int128_5265 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4570 = torch.aten.view %4566, %4569 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4570, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%4571 = torch.prim.ListConstruct %4564 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5266 = torch.constant.bool false
%4572 = torch.aten.index_put %4570, %4571, %4561, %false_5266 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4572, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5267 = torch.constant.int 32
%int2_5268 = torch.constant.int 2
%int32_5269 = torch.constant.int 32
%int8_5270 = torch.constant.int 8
%int128_5271 = torch.constant.int 128
%4573 = torch.prim.ListConstruct %416, %int32_5267, %int2_5268, %int32_5269, %int8_5270, %int128_5271 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4574 = torch.aten.view %4572, %4573 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4574, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5272 = torch.constant.int 2097152
%4575 = torch.prim.ListConstruct %416, %int2097152_5272 : (!torch.int, !torch.int) -> !torch.list<int>
%4576 = torch.aten.view %4574, %4575 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4576, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_5273 = torch.constant.int 32
%int2_5274 = torch.constant.int 2
%int32_5275 = torch.constant.int 32
%int8_5276 = torch.constant.int 8
%int128_5277 = torch.constant.int 128
%4577 = torch.prim.ListConstruct %416, %int32_5273, %int2_5274, %int32_5275, %int8_5276, %int128_5277 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4578 = torch.aten.view %4576, %4577 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4578, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5278 = torch.constant.int 32
%int8_5279 = torch.constant.int 8
%int128_5280 = torch.constant.int 128
%4579 = torch.prim.ListConstruct %4568, %int32_5278, %int8_5279, %int128_5280 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4580 = torch.aten.view %4578, %4579 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4580, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5281 = torch.constant.int 4
%int32_5282 = torch.constant.int 32
%int8_5283 = torch.constant.int 8
%int128_5284 = torch.constant.int 128
%4581 = torch.prim.ListConstruct %int4_5281, %425, %int32_5282, %int8_5283, %int128_5284 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4582 = torch.aten.view %4496, %4581 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4582, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5285 = torch.constant.int 4
%4583 = torch.aten.mul.int %int4_5285, %425 : !torch.int, !torch.int -> !torch.int
%int32_5286 = torch.constant.int 32
%int8_5287 = torch.constant.int 8
%int128_5288 = torch.constant.int 128
%4584 = torch.prim.ListConstruct %4583, %int32_5286, %int8_5287, %int128_5288 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4585 = torch.aten.view %4582, %4584 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4585, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_5289 = torch.constant.int 1
%int1_5290 = torch.constant.int 1
%4586 = torch.aten.add.Scalar %4556, %int1_5289, %int1_5290 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4586, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5291 = torch.constant.int 4
%4587 = torch.aten.mul.int %int4_5291, %425 : !torch.int, !torch.int -> !torch.int
%4588 = torch.prim.ListConstruct %4587 : (!torch.int) -> !torch.list<int>
%4589 = torch.aten.view %4586, %4588 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4589, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%4590 = torch.prim.ListConstruct %4589 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5292 = torch.constant.bool false
%4591 = torch.aten.index_put %4580, %4590, %4585, %false_5292 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4591, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5293 = torch.constant.int 32
%int2_5294 = torch.constant.int 2
%int32_5295 = torch.constant.int 32
%int8_5296 = torch.constant.int 8
%int128_5297 = torch.constant.int 128
%4592 = torch.prim.ListConstruct %416, %int32_5293, %int2_5294, %int32_5295, %int8_5296, %int128_5297 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4593 = torch.aten.view %4591, %4592 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4593, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5298 = torch.constant.int 2097152
%4594 = torch.prim.ListConstruct %416, %int2097152_5298 : (!torch.int, !torch.int) -> !torch.list<int>
%4595 = torch.aten.view %4593, %4594 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4595, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_5299 = torch.constant.int -2
%4596 = torch.aten.unsqueeze %4554, %int-2_5299 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4596, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_5300 = torch.constant.int 4
%int8_5301 = torch.constant.int 8
%int4_5302 = torch.constant.int 4
%int128_5303 = torch.constant.int 128
%4597 = torch.prim.ListConstruct %int4_5300, %4540, %int8_5301, %int4_5302, %int128_5303 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5304 = torch.constant.bool false
%4598 = torch.aten.expand %4596, %4597, %false_5304 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4598, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5305 = torch.constant.int 0
%4599 = torch.aten.clone %4598, %int0_5305 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4599, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5306 = torch.constant.int 4
%int32_5307 = torch.constant.int 32
%int128_5308 = torch.constant.int 128
%4600 = torch.prim.ListConstruct %int4_5306, %4540, %int32_5307, %int128_5308 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4601 = torch.aten._unsafe_view %4599, %4600 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4601, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_5309 = torch.constant.int -2
%4602 = torch.aten.unsqueeze %4496, %int-2_5309 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4602, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_5310 = torch.constant.int 1
%4603 = torch.aten.size.int %4490, %int1_5310 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_5311 = torch.constant.int 4
%int8_5312 = torch.constant.int 8
%int4_5313 = torch.constant.int 4
%int128_5314 = torch.constant.int 128
%4604 = torch.prim.ListConstruct %int4_5311, %4603, %int8_5312, %int4_5313, %int128_5314 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5315 = torch.constant.bool false
%4605 = torch.aten.expand %4602, %4604, %false_5315 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4605, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5316 = torch.constant.int 0
%4606 = torch.aten.clone %4605, %int0_5316 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4606, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5317 = torch.constant.int 4
%int32_5318 = torch.constant.int 32
%int128_5319 = torch.constant.int 128
%4607 = torch.prim.ListConstruct %int4_5317, %4603, %int32_5318, %int128_5319 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4608 = torch.aten._unsafe_view %4606, %4607 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4608, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_5320 = torch.constant.int 1
%int2_5321 = torch.constant.int 2
%4609 = torch.aten.transpose.int %4525, %int1_5320, %int2_5321 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4609, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5322 = torch.constant.int 1
%int2_5323 = torch.constant.int 2
%4610 = torch.aten.transpose.int %4601, %int1_5322, %int2_5323 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4610, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5324 = torch.constant.int 1
%int2_5325 = torch.constant.int 2
%4611 = torch.aten.transpose.int %4608, %int1_5324, %int2_5325 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4611, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_5326 = torch.constant.float 0.000000e+00
%false_5327 = torch.constant.bool false
%none_5328 = torch.constant.none
%4612:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%4609, %4610, %4611, %float0.000000e00_5326, %false_5327, %320, %none_5328) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %4612#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5329 = torch.constant.int 1
%int2_5330 = torch.constant.int 2
%4613 = torch.aten.transpose.int %4612#0, %int1_5329, %int2_5330 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4613, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5331 = torch.constant.int 4
%int4096_5332 = torch.constant.int 4096
%4614 = torch.prim.ListConstruct %int4_5331, %4511, %int4096_5332 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4615 = torch.aten.view %4613, %4614 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4615, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5333 = torch.constant.int -2
%int-1_5334 = torch.constant.int -1
%4616 = torch.aten.transpose.int %194, %int-2_5333, %int-1_5334 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5335 = torch.constant.int 4
%4617 = torch.aten.mul.int %int4_5335, %4511 : !torch.int, !torch.int -> !torch.int
%int4096_5336 = torch.constant.int 4096
%4618 = torch.prim.ListConstruct %4617, %int4096_5336 : (!torch.int, !torch.int) -> !torch.list<int>
%4619 = torch.aten.view %4615, %4618 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4619, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4620 = torch.aten.mm %4619, %4616 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4620, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5337 = torch.constant.int 4
%int4096_5338 = torch.constant.int 4096
%4621 = torch.prim.ListConstruct %int4_5337, %4511, %int4096_5338 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4622 = torch.aten.view %4620, %4621 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4622, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5339 = torch.constant.int 1
%4623 = torch.aten.add.Tensor %4460, %4622, %int1_5339 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4623, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5340 = torch.constant.int 6
%4624 = torch.prims.convert_element_type %4623, %int6_5340 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4624, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5341 = torch.constant.int 2
%4625 = torch.aten.pow.Tensor_Scalar %4624, %int2_5341 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4625, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5342 = torch.constant.int -1
%4626 = torch.prim.ListConstruct %int-1_5342 : (!torch.int) -> !torch.list<int>
%true_5343 = torch.constant.bool true
%none_5344 = torch.constant.none
%4627 = torch.aten.mean.dim %4625, %4626, %true_5343, %none_5344 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4627, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5345 = torch.constant.float 9.9999997473787516E-6
%int1_5346 = torch.constant.int 1
%4628 = torch.aten.add.Scalar %4627, %float9.999990e-06_5345, %int1_5346 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4628, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4629 = torch.aten.rsqrt %4628 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4629, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4630 = torch.aten.mul.Tensor %4624, %4629 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4630, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4631 = torch.aten.mul.Tensor %195, %4630 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4631, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5347 = torch.constant.int 5
%4632 = torch.prims.convert_element_type %4631, %int5_5347 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4632, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5348 = torch.constant.int -2
%int-1_5349 = torch.constant.int -1
%4633 = torch.aten.transpose.int %196, %int-2_5348, %int-1_5349 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5350 = torch.constant.int 4
%4634 = torch.aten.mul.int %int4_5350, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5351 = torch.constant.int 4096
%4635 = torch.prim.ListConstruct %4634, %int4096_5351 : (!torch.int, !torch.int) -> !torch.list<int>
%4636 = torch.aten.view %4632, %4635 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4636, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4637 = torch.aten.mm %4636, %4633 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4637, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5352 = torch.constant.int 4
%int14336_5353 = torch.constant.int 14336
%4638 = torch.prim.ListConstruct %int4_5352, %294, %int14336_5353 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4639 = torch.aten.view %4637, %4638 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4639, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4640 = torch.aten.silu %4639 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4640, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5354 = torch.constant.int -2
%int-1_5355 = torch.constant.int -1
%4641 = torch.aten.transpose.int %197, %int-2_5354, %int-1_5355 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5356 = torch.constant.int 4
%4642 = torch.aten.mul.int %int4_5356, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5357 = torch.constant.int 4096
%4643 = torch.prim.ListConstruct %4642, %int4096_5357 : (!torch.int, !torch.int) -> !torch.list<int>
%4644 = torch.aten.view %4632, %4643 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4644, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4645 = torch.aten.mm %4644, %4641 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4645, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5358 = torch.constant.int 4
%int14336_5359 = torch.constant.int 14336
%4646 = torch.prim.ListConstruct %int4_5358, %294, %int14336_5359 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4647 = torch.aten.view %4645, %4646 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4647, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4648 = torch.aten.mul.Tensor %4640, %4647 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4648, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5360 = torch.constant.int -2
%int-1_5361 = torch.constant.int -1
%4649 = torch.aten.transpose.int %198, %int-2_5360, %int-1_5361 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_5362 = torch.constant.int 1
%4650 = torch.aten.size.int %4639, %int1_5362 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_5363 = torch.constant.int 4
%4651 = torch.aten.mul.int %int4_5363, %4650 : !torch.int, !torch.int -> !torch.int
%int14336_5364 = torch.constant.int 14336
%4652 = torch.prim.ListConstruct %4651, %int14336_5364 : (!torch.int, !torch.int) -> !torch.list<int>
%4653 = torch.aten.view %4648, %4652 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4653, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%4654 = torch.aten.mm %4653, %4649 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4654, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5365 = torch.constant.int 4
%int4096_5366 = torch.constant.int 4096
%4655 = torch.prim.ListConstruct %int4_5365, %4650, %int4096_5366 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4656 = torch.aten.view %4654, %4655 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4656, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5367 = torch.constant.int 1
%4657 = torch.aten.add.Tensor %4623, %4656, %int1_5367 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4657, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5368 = torch.constant.int 6
%4658 = torch.prims.convert_element_type %4657, %int6_5368 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4658, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5369 = torch.constant.int 2
%4659 = torch.aten.pow.Tensor_Scalar %4658, %int2_5369 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4659, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5370 = torch.constant.int -1
%4660 = torch.prim.ListConstruct %int-1_5370 : (!torch.int) -> !torch.list<int>
%true_5371 = torch.constant.bool true
%none_5372 = torch.constant.none
%4661 = torch.aten.mean.dim %4659, %4660, %true_5371, %none_5372 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4661, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5373 = torch.constant.float 9.9999997473787516E-6
%int1_5374 = torch.constant.int 1
%4662 = torch.aten.add.Scalar %4661, %float9.999990e-06_5373, %int1_5374 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4662, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4663 = torch.aten.rsqrt %4662 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4663, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4664 = torch.aten.mul.Tensor %4658, %4663 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4664, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4665 = torch.aten.mul.Tensor %199, %4664 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4665, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5375 = torch.constant.int 5
%4666 = torch.prims.convert_element_type %4665, %int5_5375 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4666, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5376 = torch.constant.int -2
%int-1_5377 = torch.constant.int -1
%4667 = torch.aten.transpose.int %200, %int-2_5376, %int-1_5377 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5378 = torch.constant.int 4
%4668 = torch.aten.mul.int %int4_5378, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5379 = torch.constant.int 4096
%4669 = torch.prim.ListConstruct %4668, %int4096_5379 : (!torch.int, !torch.int) -> !torch.list<int>
%4670 = torch.aten.view %4666, %4669 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4670, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4671 = torch.aten.mm %4670, %4667 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4671, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5380 = torch.constant.int 4
%int4096_5381 = torch.constant.int 4096
%4672 = torch.prim.ListConstruct %int4_5380, %294, %int4096_5381 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4673 = torch.aten.view %4671, %4672 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4673, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5382 = torch.constant.int -2
%int-1_5383 = torch.constant.int -1
%4674 = torch.aten.transpose.int %201, %int-2_5382, %int-1_5383 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5384 = torch.constant.int 4
%4675 = torch.aten.mul.int %int4_5384, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5385 = torch.constant.int 4096
%4676 = torch.prim.ListConstruct %4675, %int4096_5385 : (!torch.int, !torch.int) -> !torch.list<int>
%4677 = torch.aten.view %4666, %4676 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4677, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4678 = torch.aten.mm %4677, %4674 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4678, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5386 = torch.constant.int 4
%int1024_5387 = torch.constant.int 1024
%4679 = torch.prim.ListConstruct %int4_5386, %294, %int1024_5387 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4680 = torch.aten.view %4678, %4679 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4680, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_5388 = torch.constant.int -2
%int-1_5389 = torch.constant.int -1
%4681 = torch.aten.transpose.int %202, %int-2_5388, %int-1_5389 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5390 = torch.constant.int 4
%4682 = torch.aten.mul.int %int4_5390, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5391 = torch.constant.int 4096
%4683 = torch.prim.ListConstruct %4682, %int4096_5391 : (!torch.int, !torch.int) -> !torch.list<int>
%4684 = torch.aten.view %4666, %4683 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4684, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4685 = torch.aten.mm %4684, %4681 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4685, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5392 = torch.constant.int 4
%int1024_5393 = torch.constant.int 1024
%4686 = torch.prim.ListConstruct %int4_5392, %294, %int1024_5393 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4687 = torch.aten.view %4685, %4686 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4687, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_5394 = torch.constant.int 4
%int32_5395 = torch.constant.int 32
%int128_5396 = torch.constant.int 128
%4688 = torch.prim.ListConstruct %int4_5394, %294, %int32_5395, %int128_5396 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4689 = torch.aten.view %4673, %4688 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4689, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5397 = torch.constant.int 4
%int8_5398 = torch.constant.int 8
%int128_5399 = torch.constant.int 128
%4690 = torch.prim.ListConstruct %int4_5397, %294, %int8_5398, %int128_5399 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4691 = torch.aten.view %4680, %4690 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4691, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_5400 = torch.constant.int 4
%int8_5401 = torch.constant.int 8
%int128_5402 = torch.constant.int 128
%4692 = torch.prim.ListConstruct %int4_5400, %294, %int8_5401, %int128_5402 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4693 = torch.aten.view %4687, %4692 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4693, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_5403 = torch.constant.int 131072
%none_5404 = torch.constant.none
%none_5405 = torch.constant.none
%cpu_5406 = torch.constant.device "cpu"
%false_5407 = torch.constant.bool false
%4694 = torch.aten.arange %int131072_5403, %none_5404, %none_5405, %cpu_5406, %false_5407 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5408 = torch.constant.int 0
%int128_5409 = torch.constant.int 128
%int2_5410 = torch.constant.int 2
%none_5411 = torch.constant.none
%none_5412 = torch.constant.none
%cpu_5413 = torch.constant.device "cpu"
%false_5414 = torch.constant.bool false
%4695 = torch.aten.arange.start_step %int0_5408, %int128_5409, %int2_5410, %none_5411, %none_5412, %cpu_5413, %false_5414 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5415 = torch.constant.int 0
%int0_5416 = torch.constant.int 0
%int64_5417 = torch.constant.int 64
%int1_5418 = torch.constant.int 1
%4696 = torch.aten.slice.Tensor %4695, %int0_5415, %int0_5416, %int64_5417, %int1_5418 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5419 = torch.constant.int 6
%4697 = torch.prims.convert_element_type %4696, %int6_5419 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5420 = torch.constant.int 128
%4698 = torch.aten.div.Scalar %4697, %int128_5420 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5421 = torch.constant.float 5.000000e+05
%4699 = torch.aten.pow.Scalar %float5.000000e05_5421, %4698 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4700 = torch.aten.reciprocal %4699 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5422 = torch.constant.float 1.000000e+00
%4701 = torch.aten.mul.Scalar %4700, %float1.000000e00_5422 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5423 = torch.constant.int 131072
%int1_5424 = torch.constant.int 1
%4702 = torch.prim.ListConstruct %int131072_5423, %int1_5424 : (!torch.int, !torch.int) -> !torch.list<int>
%4703 = torch.aten.view %4694, %4702 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4704 = torch.aten.mul.Tensor %4703, %4701 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4705 = torch.aten.cos %4704 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4706 = torch.aten.sin %4704 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4707 = torch.aten.complex %4705, %4706 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5425 = torch.constant.int 1
%4708 = torch.aten.size.int %4673, %int1_5425 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_5426 = torch.constant.int 0
%4709 = torch.aten.add.int %int0_5426, %4708 : !torch.int, !torch.int -> !torch.int
%int0_5427 = torch.constant.int 0
%int0_5428 = torch.constant.int 0
%int1_5429 = torch.constant.int 1
%4710 = torch.aten.slice.Tensor %4707, %int0_5427, %int0_5428, %4709, %int1_5429 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4710, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5430 = torch.constant.int 1
%int0_5431 = torch.constant.int 0
%int9223372036854775807_5432 = torch.constant.int 9223372036854775807
%int1_5433 = torch.constant.int 1
%4711 = torch.aten.slice.Tensor %4710, %int1_5430, %int0_5431, %int9223372036854775807_5432, %int1_5433 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4711, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5434 = torch.constant.int 0
%4712 = torch.aten.unsqueeze %4711, %int0_5434 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4712, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5435 = torch.constant.int 2
%4713 = torch.aten.unsqueeze %4712, %int2_5435 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4713, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5436 = torch.constant.int 3
%int0_5437 = torch.constant.int 0
%int9223372036854775807_5438 = torch.constant.int 9223372036854775807
%int1_5439 = torch.constant.int 1
%4714 = torch.aten.slice.Tensor %4713, %int3_5436, %int0_5437, %int9223372036854775807_5438, %int1_5439 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4714, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4715 = torch_c.to_builtin_tensor %4689 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_5440 = arith.constant 1 : index
%dim_5441 = tensor.dim %4715, %c1_5440 : tensor<4x?x32x128xf16>
%4716 = flow.tensor.bitcast %4715 : tensor<4x?x32x128xf16>{%dim_5441} -> tensor<4x?x32x64xcomplex<f16>>{%dim_5441}
%4717 = torch_c.from_builtin_tensor %4716 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %4717, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%4718 = torch.aten.mul.Tensor %4717, %4714 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %4718, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%4719 = torch_c.to_builtin_tensor %4718 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_5442 = arith.constant 1 : index
%dim_5443 = tensor.dim %4719, %c1_5442 : tensor<4x?x32x64xcomplex<f32>>
%4720 = flow.tensor.bitcast %4719 : tensor<4x?x32x64xcomplex<f32>>{%dim_5443} -> tensor<4x?x32x128xf32>{%dim_5443}
%4721 = torch_c.from_builtin_tensor %4720 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %4721, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_5444 = torch.constant.int 5
%4722 = torch.prims.convert_element_type %4721, %int5_5444 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4722, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_5445 = torch.constant.int 131072
%none_5446 = torch.constant.none
%none_5447 = torch.constant.none
%cpu_5448 = torch.constant.device "cpu"
%false_5449 = torch.constant.bool false
%4723 = torch.aten.arange %int131072_5445, %none_5446, %none_5447, %cpu_5448, %false_5449 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5450 = torch.constant.int 0
%int128_5451 = torch.constant.int 128
%int2_5452 = torch.constant.int 2
%none_5453 = torch.constant.none
%none_5454 = torch.constant.none
%cpu_5455 = torch.constant.device "cpu"
%false_5456 = torch.constant.bool false
%4724 = torch.aten.arange.start_step %int0_5450, %int128_5451, %int2_5452, %none_5453, %none_5454, %cpu_5455, %false_5456 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5457 = torch.constant.int 0
%int0_5458 = torch.constant.int 0
%int64_5459 = torch.constant.int 64
%int1_5460 = torch.constant.int 1
%4725 = torch.aten.slice.Tensor %4724, %int0_5457, %int0_5458, %int64_5459, %int1_5460 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5461 = torch.constant.int 6
%4726 = torch.prims.convert_element_type %4725, %int6_5461 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5462 = torch.constant.int 128
%4727 = torch.aten.div.Scalar %4726, %int128_5462 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5463 = torch.constant.float 5.000000e+05
%4728 = torch.aten.pow.Scalar %float5.000000e05_5463, %4727 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4729 = torch.aten.reciprocal %4728 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5464 = torch.constant.float 1.000000e+00
%4730 = torch.aten.mul.Scalar %4729, %float1.000000e00_5464 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5465 = torch.constant.int 131072
%int1_5466 = torch.constant.int 1
%4731 = torch.prim.ListConstruct %int131072_5465, %int1_5466 : (!torch.int, !torch.int) -> !torch.list<int>
%4732 = torch.aten.view %4723, %4731 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4733 = torch.aten.mul.Tensor %4732, %4730 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4734 = torch.aten.cos %4733 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4735 = torch.aten.sin %4733 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4736 = torch.aten.complex %4734, %4735 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5467 = torch.constant.int 1
%4737 = torch.aten.size.int %4680, %int1_5467 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_5468 = torch.constant.int 0
%4738 = torch.aten.add.int %int0_5468, %4737 : !torch.int, !torch.int -> !torch.int
%int0_5469 = torch.constant.int 0
%int0_5470 = torch.constant.int 0
%int1_5471 = torch.constant.int 1
%4739 = torch.aten.slice.Tensor %4736, %int0_5469, %int0_5470, %4738, %int1_5471 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4739, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5472 = torch.constant.int 1
%int0_5473 = torch.constant.int 0
%int9223372036854775807_5474 = torch.constant.int 9223372036854775807
%int1_5475 = torch.constant.int 1
%4740 = torch.aten.slice.Tensor %4739, %int1_5472, %int0_5473, %int9223372036854775807_5474, %int1_5475 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4740, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5476 = torch.constant.int 0
%4741 = torch.aten.unsqueeze %4740, %int0_5476 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4741, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5477 = torch.constant.int 2
%4742 = torch.aten.unsqueeze %4741, %int2_5477 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4742, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5478 = torch.constant.int 3
%int0_5479 = torch.constant.int 0
%int9223372036854775807_5480 = torch.constant.int 9223372036854775807
%int1_5481 = torch.constant.int 1
%4743 = torch.aten.slice.Tensor %4742, %int3_5478, %int0_5479, %int9223372036854775807_5480, %int1_5481 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4743, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4744 = torch_c.to_builtin_tensor %4691 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_5482 = arith.constant 1 : index
%dim_5483 = tensor.dim %4744, %c1_5482 : tensor<4x?x8x128xf16>
%4745 = flow.tensor.bitcast %4744 : tensor<4x?x8x128xf16>{%dim_5483} -> tensor<4x?x8x64xcomplex<f16>>{%dim_5483}
%4746 = torch_c.from_builtin_tensor %4745 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %4746, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%4747 = torch.aten.mul.Tensor %4746, %4743 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %4747, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%4748 = torch_c.to_builtin_tensor %4747 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_5484 = arith.constant 1 : index
%dim_5485 = tensor.dim %4748, %c1_5484 : tensor<4x?x8x64xcomplex<f32>>
%4749 = flow.tensor.bitcast %4748 : tensor<4x?x8x64xcomplex<f32>>{%dim_5485} -> tensor<4x?x8x128xf32>{%dim_5485}
%4750 = torch_c.from_builtin_tensor %4749 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %4750, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_5486 = torch.constant.int 5
%4751 = torch.prims.convert_element_type %4750, %int5_5486 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4751, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_5487 = torch.constant.int 64
%4752 = torch.aten.mul.Scalar %arg2, %int64_5487 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4752, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int44 = torch.constant.int 44
%int1_5488 = torch.constant.int 1
%4753 = torch.aten.add.Scalar %4752, %int44, %int1_5488 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4753, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5489 = torch.constant.int 4
%int32_5490 = torch.constant.int 32
%int8_5491 = torch.constant.int 8
%int128_5492 = torch.constant.int 128
%4754 = torch.prim.ListConstruct %int4_5489, %425, %int32_5490, %int8_5491, %int128_5492 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4755 = torch.aten.view %4751, %4754 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4755, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5493 = torch.constant.int 4
%4756 = torch.aten.mul.int %int4_5493, %425 : !torch.int, !torch.int -> !torch.int
%int32_5494 = torch.constant.int 32
%int8_5495 = torch.constant.int 8
%int128_5496 = torch.constant.int 128
%4757 = torch.prim.ListConstruct %4756, %int32_5494, %int8_5495, %int128_5496 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4758 = torch.aten.view %4755, %4757 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4758, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5497 = torch.constant.int 4
%4759 = torch.aten.mul.int %int4_5497, %425 : !torch.int, !torch.int -> !torch.int
%4760 = torch.prim.ListConstruct %4759 : (!torch.int) -> !torch.list<int>
%4761 = torch.aten.view %4753, %4760 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4761, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_5498 = torch.constant.int 32
%int2_5499 = torch.constant.int 2
%int32_5500 = torch.constant.int 32
%int8_5501 = torch.constant.int 8
%int128_5502 = torch.constant.int 128
%4762 = torch.prim.ListConstruct %416, %int32_5498, %int2_5499, %int32_5500, %int8_5501, %int128_5502 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4763 = torch.aten.view %4595, %4762 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4763, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5503 = torch.constant.int 32
%4764 = torch.aten.mul.int %416, %int32_5503 : !torch.int, !torch.int -> !torch.int
%int2_5504 = torch.constant.int 2
%4765 = torch.aten.mul.int %4764, %int2_5504 : !torch.int, !torch.int -> !torch.int
%int32_5505 = torch.constant.int 32
%int8_5506 = torch.constant.int 8
%int128_5507 = torch.constant.int 128
%4766 = torch.prim.ListConstruct %4765, %int32_5505, %int8_5506, %int128_5507 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4767 = torch.aten.view %4763, %4766 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4767, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%4768 = torch.prim.ListConstruct %4761 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5508 = torch.constant.bool false
%4769 = torch.aten.index_put %4767, %4768, %4758, %false_5508 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4769, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5509 = torch.constant.int 32
%int2_5510 = torch.constant.int 2
%int32_5511 = torch.constant.int 32
%int8_5512 = torch.constant.int 8
%int128_5513 = torch.constant.int 128
%4770 = torch.prim.ListConstruct %416, %int32_5509, %int2_5510, %int32_5511, %int8_5512, %int128_5513 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4771 = torch.aten.view %4769, %4770 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4771, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5514 = torch.constant.int 2097152
%4772 = torch.prim.ListConstruct %416, %int2097152_5514 : (!torch.int, !torch.int) -> !torch.list<int>
%4773 = torch.aten.view %4771, %4772 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4773, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_5515 = torch.constant.int 32
%int2_5516 = torch.constant.int 2
%int32_5517 = torch.constant.int 32
%int8_5518 = torch.constant.int 8
%int128_5519 = torch.constant.int 128
%4774 = torch.prim.ListConstruct %416, %int32_5515, %int2_5516, %int32_5517, %int8_5518, %int128_5519 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4775 = torch.aten.view %4773, %4774 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4775, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5520 = torch.constant.int 32
%int8_5521 = torch.constant.int 8
%int128_5522 = torch.constant.int 128
%4776 = torch.prim.ListConstruct %4765, %int32_5520, %int8_5521, %int128_5522 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4777 = torch.aten.view %4775, %4776 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4777, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5523 = torch.constant.int 4
%int32_5524 = torch.constant.int 32
%int8_5525 = torch.constant.int 8
%int128_5526 = torch.constant.int 128
%4778 = torch.prim.ListConstruct %int4_5523, %425, %int32_5524, %int8_5525, %int128_5526 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4779 = torch.aten.view %4693, %4778 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4779, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5527 = torch.constant.int 4
%4780 = torch.aten.mul.int %int4_5527, %425 : !torch.int, !torch.int -> !torch.int
%int32_5528 = torch.constant.int 32
%int8_5529 = torch.constant.int 8
%int128_5530 = torch.constant.int 128
%4781 = torch.prim.ListConstruct %4780, %int32_5528, %int8_5529, %int128_5530 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4782 = torch.aten.view %4779, %4781 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4782, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_5531 = torch.constant.int 1
%int1_5532 = torch.constant.int 1
%4783 = torch.aten.add.Scalar %4753, %int1_5531, %int1_5532 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4783, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5533 = torch.constant.int 4
%4784 = torch.aten.mul.int %int4_5533, %425 : !torch.int, !torch.int -> !torch.int
%4785 = torch.prim.ListConstruct %4784 : (!torch.int) -> !torch.list<int>
%4786 = torch.aten.view %4783, %4785 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4786, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%4787 = torch.prim.ListConstruct %4786 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5534 = torch.constant.bool false
%4788 = torch.aten.index_put %4777, %4787, %4782, %false_5534 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4788, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5535 = torch.constant.int 32
%int2_5536 = torch.constant.int 2
%int32_5537 = torch.constant.int 32
%int8_5538 = torch.constant.int 8
%int128_5539 = torch.constant.int 128
%4789 = torch.prim.ListConstruct %416, %int32_5535, %int2_5536, %int32_5537, %int8_5538, %int128_5539 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4790 = torch.aten.view %4788, %4789 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4790, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5540 = torch.constant.int 2097152
%4791 = torch.prim.ListConstruct %416, %int2097152_5540 : (!torch.int, !torch.int) -> !torch.list<int>
%4792 = torch.aten.view %4790, %4791 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4792, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_5541 = torch.constant.int -2
%4793 = torch.aten.unsqueeze %4751, %int-2_5541 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4793, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_5542 = torch.constant.int 4
%int8_5543 = torch.constant.int 8
%int4_5544 = torch.constant.int 4
%int128_5545 = torch.constant.int 128
%4794 = torch.prim.ListConstruct %int4_5542, %4737, %int8_5543, %int4_5544, %int128_5545 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5546 = torch.constant.bool false
%4795 = torch.aten.expand %4793, %4794, %false_5546 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4795, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5547 = torch.constant.int 0
%4796 = torch.aten.clone %4795, %int0_5547 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4796, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5548 = torch.constant.int 4
%int32_5549 = torch.constant.int 32
%int128_5550 = torch.constant.int 128
%4797 = torch.prim.ListConstruct %int4_5548, %4737, %int32_5549, %int128_5550 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4798 = torch.aten._unsafe_view %4796, %4797 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4798, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_5551 = torch.constant.int -2
%4799 = torch.aten.unsqueeze %4693, %int-2_5551 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4799, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_5552 = torch.constant.int 1
%4800 = torch.aten.size.int %4687, %int1_5552 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_5553 = torch.constant.int 4
%int8_5554 = torch.constant.int 8
%int4_5555 = torch.constant.int 4
%int128_5556 = torch.constant.int 128
%4801 = torch.prim.ListConstruct %int4_5553, %4800, %int8_5554, %int4_5555, %int128_5556 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5557 = torch.constant.bool false
%4802 = torch.aten.expand %4799, %4801, %false_5557 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4802, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5558 = torch.constant.int 0
%4803 = torch.aten.clone %4802, %int0_5558 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4803, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5559 = torch.constant.int 4
%int32_5560 = torch.constant.int 32
%int128_5561 = torch.constant.int 128
%4804 = torch.prim.ListConstruct %int4_5559, %4800, %int32_5560, %int128_5561 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4805 = torch.aten._unsafe_view %4803, %4804 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4805, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_5562 = torch.constant.int 1
%int2_5563 = torch.constant.int 2
%4806 = torch.aten.transpose.int %4722, %int1_5562, %int2_5563 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4806, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5564 = torch.constant.int 1
%int2_5565 = torch.constant.int 2
%4807 = torch.aten.transpose.int %4798, %int1_5564, %int2_5565 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4807, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5566 = torch.constant.int 1
%int2_5567 = torch.constant.int 2
%4808 = torch.aten.transpose.int %4805, %int1_5566, %int2_5567 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %4808, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_5568 = torch.constant.float 0.000000e+00
%false_5569 = torch.constant.bool false
%none_5570 = torch.constant.none
%4809:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%4806, %4807, %4808, %float0.000000e00_5568, %false_5569, %320, %none_5570) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %4809#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5571 = torch.constant.int 1
%int2_5572 = torch.constant.int 2
%4810 = torch.aten.transpose.int %4809#0, %int1_5571, %int2_5572 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4810, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5573 = torch.constant.int 4
%int4096_5574 = torch.constant.int 4096
%4811 = torch.prim.ListConstruct %int4_5573, %4708, %int4096_5574 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4812 = torch.aten.view %4810, %4811 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4812, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5575 = torch.constant.int -2
%int-1_5576 = torch.constant.int -1
%4813 = torch.aten.transpose.int %203, %int-2_5575, %int-1_5576 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5577 = torch.constant.int 4
%4814 = torch.aten.mul.int %int4_5577, %4708 : !torch.int, !torch.int -> !torch.int
%int4096_5578 = torch.constant.int 4096
%4815 = torch.prim.ListConstruct %4814, %int4096_5578 : (!torch.int, !torch.int) -> !torch.list<int>
%4816 = torch.aten.view %4812, %4815 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4816, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4817 = torch.aten.mm %4816, %4813 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4817, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5579 = torch.constant.int 4
%int4096_5580 = torch.constant.int 4096
%4818 = torch.prim.ListConstruct %int4_5579, %4708, %int4096_5580 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4819 = torch.aten.view %4817, %4818 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4819, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5581 = torch.constant.int 1
%4820 = torch.aten.add.Tensor %4657, %4819, %int1_5581 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4820, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5582 = torch.constant.int 6
%4821 = torch.prims.convert_element_type %4820, %int6_5582 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4821, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5583 = torch.constant.int 2
%4822 = torch.aten.pow.Tensor_Scalar %4821, %int2_5583 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4822, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5584 = torch.constant.int -1
%4823 = torch.prim.ListConstruct %int-1_5584 : (!torch.int) -> !torch.list<int>
%true_5585 = torch.constant.bool true
%none_5586 = torch.constant.none
%4824 = torch.aten.mean.dim %4822, %4823, %true_5585, %none_5586 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4824, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5587 = torch.constant.float 9.9999997473787516E-6
%int1_5588 = torch.constant.int 1
%4825 = torch.aten.add.Scalar %4824, %float9.999990e-06_5587, %int1_5588 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4825, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4826 = torch.aten.rsqrt %4825 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4826, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4827 = torch.aten.mul.Tensor %4821, %4826 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4827, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4828 = torch.aten.mul.Tensor %204, %4827 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4828, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5589 = torch.constant.int 5
%4829 = torch.prims.convert_element_type %4828, %int5_5589 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4829, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5590 = torch.constant.int -2
%int-1_5591 = torch.constant.int -1
%4830 = torch.aten.transpose.int %205, %int-2_5590, %int-1_5591 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5592 = torch.constant.int 4
%4831 = torch.aten.mul.int %int4_5592, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5593 = torch.constant.int 4096
%4832 = torch.prim.ListConstruct %4831, %int4096_5593 : (!torch.int, !torch.int) -> !torch.list<int>
%4833 = torch.aten.view %4829, %4832 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4833, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4834 = torch.aten.mm %4833, %4830 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4834, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5594 = torch.constant.int 4
%int14336_5595 = torch.constant.int 14336
%4835 = torch.prim.ListConstruct %int4_5594, %294, %int14336_5595 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4836 = torch.aten.view %4834, %4835 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4836, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4837 = torch.aten.silu %4836 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4837, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5596 = torch.constant.int -2
%int-1_5597 = torch.constant.int -1
%4838 = torch.aten.transpose.int %206, %int-2_5596, %int-1_5597 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5598 = torch.constant.int 4
%4839 = torch.aten.mul.int %int4_5598, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5599 = torch.constant.int 4096
%4840 = torch.prim.ListConstruct %4839, %int4096_5599 : (!torch.int, !torch.int) -> !torch.list<int>
%4841 = torch.aten.view %4829, %4840 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4841, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4842 = torch.aten.mm %4841, %4838 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4842, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5600 = torch.constant.int 4
%int14336_5601 = torch.constant.int 14336
%4843 = torch.prim.ListConstruct %int4_5600, %294, %int14336_5601 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4844 = torch.aten.view %4842, %4843 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4844, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%4845 = torch.aten.mul.Tensor %4837, %4844 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %4845, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5602 = torch.constant.int -2
%int-1_5603 = torch.constant.int -1
%4846 = torch.aten.transpose.int %207, %int-2_5602, %int-1_5603 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_5604 = torch.constant.int 1
%4847 = torch.aten.size.int %4836, %int1_5604 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_5605 = torch.constant.int 4
%4848 = torch.aten.mul.int %int4_5605, %4847 : !torch.int, !torch.int -> !torch.int
%int14336_5606 = torch.constant.int 14336
%4849 = torch.prim.ListConstruct %4848, %int14336_5606 : (!torch.int, !torch.int) -> !torch.list<int>
%4850 = torch.aten.view %4845, %4849 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %4850, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%4851 = torch.aten.mm %4850, %4846 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4851, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5607 = torch.constant.int 4
%int4096_5608 = torch.constant.int 4096
%4852 = torch.prim.ListConstruct %int4_5607, %4847, %int4096_5608 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4853 = torch.aten.view %4851, %4852 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4853, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5609 = torch.constant.int 1
%4854 = torch.aten.add.Tensor %4820, %4853, %int1_5609 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4854, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5610 = torch.constant.int 6
%4855 = torch.prims.convert_element_type %4854, %int6_5610 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4855, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5611 = torch.constant.int 2
%4856 = torch.aten.pow.Tensor_Scalar %4855, %int2_5611 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4856, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5612 = torch.constant.int -1
%4857 = torch.prim.ListConstruct %int-1_5612 : (!torch.int) -> !torch.list<int>
%true_5613 = torch.constant.bool true
%none_5614 = torch.constant.none
%4858 = torch.aten.mean.dim %4856, %4857, %true_5613, %none_5614 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4858, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5615 = torch.constant.float 9.9999997473787516E-6
%int1_5616 = torch.constant.int 1
%4859 = torch.aten.add.Scalar %4858, %float9.999990e-06_5615, %int1_5616 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4859, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4860 = torch.aten.rsqrt %4859 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %4860, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%4861 = torch.aten.mul.Tensor %4855, %4860 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4861, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%4862 = torch.aten.mul.Tensor %208, %4861 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %4862, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5617 = torch.constant.int 5
%4863 = torch.prims.convert_element_type %4862, %int5_5617 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4863, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5618 = torch.constant.int -2
%int-1_5619 = torch.constant.int -1
%4864 = torch.aten.transpose.int %209, %int-2_5618, %int-1_5619 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5620 = torch.constant.int 4
%4865 = torch.aten.mul.int %int4_5620, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5621 = torch.constant.int 4096
%4866 = torch.prim.ListConstruct %4865, %int4096_5621 : (!torch.int, !torch.int) -> !torch.list<int>
%4867 = torch.aten.view %4863, %4866 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4867, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4868 = torch.aten.mm %4867, %4864 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4868, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5622 = torch.constant.int 4
%int4096_5623 = torch.constant.int 4096
%4869 = torch.prim.ListConstruct %int4_5622, %294, %int4096_5623 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4870 = torch.aten.view %4868, %4869 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %4870, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5624 = torch.constant.int -2
%int-1_5625 = torch.constant.int -1
%4871 = torch.aten.transpose.int %210, %int-2_5624, %int-1_5625 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5626 = torch.constant.int 4
%4872 = torch.aten.mul.int %int4_5626, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5627 = torch.constant.int 4096
%4873 = torch.prim.ListConstruct %4872, %int4096_5627 : (!torch.int, !torch.int) -> !torch.list<int>
%4874 = torch.aten.view %4863, %4873 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4874, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4875 = torch.aten.mm %4874, %4871 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4875, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5628 = torch.constant.int 4
%int1024_5629 = torch.constant.int 1024
%4876 = torch.prim.ListConstruct %int4_5628, %294, %int1024_5629 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4877 = torch.aten.view %4875, %4876 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4877, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_5630 = torch.constant.int -2
%int-1_5631 = torch.constant.int -1
%4878 = torch.aten.transpose.int %211, %int-2_5630, %int-1_5631 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5632 = torch.constant.int 4
%4879 = torch.aten.mul.int %int4_5632, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5633 = torch.constant.int 4096
%4880 = torch.prim.ListConstruct %4879, %int4096_5633 : (!torch.int, !torch.int) -> !torch.list<int>
%4881 = torch.aten.view %4863, %4880 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %4881, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%4882 = torch.aten.mm %4881, %4878 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %4882, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5634 = torch.constant.int 4
%int1024_5635 = torch.constant.int 1024
%4883 = torch.prim.ListConstruct %int4_5634, %294, %int1024_5635 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4884 = torch.aten.view %4882, %4883 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %4884, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_5636 = torch.constant.int 4
%int32_5637 = torch.constant.int 32
%int128_5638 = torch.constant.int 128
%4885 = torch.prim.ListConstruct %int4_5636, %294, %int32_5637, %int128_5638 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4886 = torch.aten.view %4870, %4885 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4886, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5639 = torch.constant.int 4
%int8_5640 = torch.constant.int 8
%int128_5641 = torch.constant.int 128
%4887 = torch.prim.ListConstruct %int4_5639, %294, %int8_5640, %int128_5641 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4888 = torch.aten.view %4877, %4887 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4888, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_5642 = torch.constant.int 4
%int8_5643 = torch.constant.int 8
%int128_5644 = torch.constant.int 128
%4889 = torch.prim.ListConstruct %int4_5642, %294, %int8_5643, %int128_5644 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4890 = torch.aten.view %4884, %4889 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4890, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_5645 = torch.constant.int 131072
%none_5646 = torch.constant.none
%none_5647 = torch.constant.none
%cpu_5648 = torch.constant.device "cpu"
%false_5649 = torch.constant.bool false
%4891 = torch.aten.arange %int131072_5645, %none_5646, %none_5647, %cpu_5648, %false_5649 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5650 = torch.constant.int 0
%int128_5651 = torch.constant.int 128
%int2_5652 = torch.constant.int 2
%none_5653 = torch.constant.none
%none_5654 = torch.constant.none
%cpu_5655 = torch.constant.device "cpu"
%false_5656 = torch.constant.bool false
%4892 = torch.aten.arange.start_step %int0_5650, %int128_5651, %int2_5652, %none_5653, %none_5654, %cpu_5655, %false_5656 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5657 = torch.constant.int 0
%int0_5658 = torch.constant.int 0
%int64_5659 = torch.constant.int 64
%int1_5660 = torch.constant.int 1
%4893 = torch.aten.slice.Tensor %4892, %int0_5657, %int0_5658, %int64_5659, %int1_5660 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5661 = torch.constant.int 6
%4894 = torch.prims.convert_element_type %4893, %int6_5661 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5662 = torch.constant.int 128
%4895 = torch.aten.div.Scalar %4894, %int128_5662 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5663 = torch.constant.float 5.000000e+05
%4896 = torch.aten.pow.Scalar %float5.000000e05_5663, %4895 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4897 = torch.aten.reciprocal %4896 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5664 = torch.constant.float 1.000000e+00
%4898 = torch.aten.mul.Scalar %4897, %float1.000000e00_5664 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5665 = torch.constant.int 131072
%int1_5666 = torch.constant.int 1
%4899 = torch.prim.ListConstruct %int131072_5665, %int1_5666 : (!torch.int, !torch.int) -> !torch.list<int>
%4900 = torch.aten.view %4891, %4899 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4901 = torch.aten.mul.Tensor %4900, %4898 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4902 = torch.aten.cos %4901 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4903 = torch.aten.sin %4901 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4904 = torch.aten.complex %4902, %4903 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5667 = torch.constant.int 1
%4905 = torch.aten.size.int %4870, %int1_5667 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_5668 = torch.constant.int 0
%4906 = torch.aten.add.int %int0_5668, %4905 : !torch.int, !torch.int -> !torch.int
%int0_5669 = torch.constant.int 0
%int0_5670 = torch.constant.int 0
%int1_5671 = torch.constant.int 1
%4907 = torch.aten.slice.Tensor %4904, %int0_5669, %int0_5670, %4906, %int1_5671 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4907, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5672 = torch.constant.int 1
%int0_5673 = torch.constant.int 0
%int9223372036854775807_5674 = torch.constant.int 9223372036854775807
%int1_5675 = torch.constant.int 1
%4908 = torch.aten.slice.Tensor %4907, %int1_5672, %int0_5673, %int9223372036854775807_5674, %int1_5675 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4908, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5676 = torch.constant.int 0
%4909 = torch.aten.unsqueeze %4908, %int0_5676 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4909, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5677 = torch.constant.int 2
%4910 = torch.aten.unsqueeze %4909, %int2_5677 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4910, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5678 = torch.constant.int 3
%int0_5679 = torch.constant.int 0
%int9223372036854775807_5680 = torch.constant.int 9223372036854775807
%int1_5681 = torch.constant.int 1
%4911 = torch.aten.slice.Tensor %4910, %int3_5678, %int0_5679, %int9223372036854775807_5680, %int1_5681 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4911, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4912 = torch_c.to_builtin_tensor %4886 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_5682 = arith.constant 1 : index
%dim_5683 = tensor.dim %4912, %c1_5682 : tensor<4x?x32x128xf16>
%4913 = flow.tensor.bitcast %4912 : tensor<4x?x32x128xf16>{%dim_5683} -> tensor<4x?x32x64xcomplex<f16>>{%dim_5683}
%4914 = torch_c.from_builtin_tensor %4913 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %4914, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%4915 = torch.aten.mul.Tensor %4914, %4911 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %4915, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%4916 = torch_c.to_builtin_tensor %4915 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_5684 = arith.constant 1 : index
%dim_5685 = tensor.dim %4916, %c1_5684 : tensor<4x?x32x64xcomplex<f32>>
%4917 = flow.tensor.bitcast %4916 : tensor<4x?x32x64xcomplex<f32>>{%dim_5685} -> tensor<4x?x32x128xf32>{%dim_5685}
%4918 = torch_c.from_builtin_tensor %4917 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %4918, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_5686 = torch.constant.int 5
%4919 = torch.prims.convert_element_type %4918, %int5_5686 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4919, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_5687 = torch.constant.int 131072
%none_5688 = torch.constant.none
%none_5689 = torch.constant.none
%cpu_5690 = torch.constant.device "cpu"
%false_5691 = torch.constant.bool false
%4920 = torch.aten.arange %int131072_5687, %none_5688, %none_5689, %cpu_5690, %false_5691 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5692 = torch.constant.int 0
%int128_5693 = torch.constant.int 128
%int2_5694 = torch.constant.int 2
%none_5695 = torch.constant.none
%none_5696 = torch.constant.none
%cpu_5697 = torch.constant.device "cpu"
%false_5698 = torch.constant.bool false
%4921 = torch.aten.arange.start_step %int0_5692, %int128_5693, %int2_5694, %none_5695, %none_5696, %cpu_5697, %false_5698 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5699 = torch.constant.int 0
%int0_5700 = torch.constant.int 0
%int64_5701 = torch.constant.int 64
%int1_5702 = torch.constant.int 1
%4922 = torch.aten.slice.Tensor %4921, %int0_5699, %int0_5700, %int64_5701, %int1_5702 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5703 = torch.constant.int 6
%4923 = torch.prims.convert_element_type %4922, %int6_5703 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5704 = torch.constant.int 128
%4924 = torch.aten.div.Scalar %4923, %int128_5704 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5705 = torch.constant.float 5.000000e+05
%4925 = torch.aten.pow.Scalar %float5.000000e05_5705, %4924 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%4926 = torch.aten.reciprocal %4925 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5706 = torch.constant.float 1.000000e+00
%4927 = torch.aten.mul.Scalar %4926, %float1.000000e00_5706 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5707 = torch.constant.int 131072
%int1_5708 = torch.constant.int 1
%4928 = torch.prim.ListConstruct %int131072_5707, %int1_5708 : (!torch.int, !torch.int) -> !torch.list<int>
%4929 = torch.aten.view %4920, %4928 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%4930 = torch.aten.mul.Tensor %4929, %4927 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%4931 = torch.aten.cos %4930 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4932 = torch.aten.sin %4930 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%4933 = torch.aten.complex %4931, %4932 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5709 = torch.constant.int 1
%4934 = torch.aten.size.int %4877, %int1_5709 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_5710 = torch.constant.int 0
%4935 = torch.aten.add.int %int0_5710, %4934 : !torch.int, !torch.int -> !torch.int
%int0_5711 = torch.constant.int 0
%int0_5712 = torch.constant.int 0
%int1_5713 = torch.constant.int 1
%4936 = torch.aten.slice.Tensor %4933, %int0_5711, %int0_5712, %4935, %int1_5713 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4936, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5714 = torch.constant.int 1
%int0_5715 = torch.constant.int 0
%int9223372036854775807_5716 = torch.constant.int 9223372036854775807
%int1_5717 = torch.constant.int 1
%4937 = torch.aten.slice.Tensor %4936, %int1_5714, %int0_5715, %int9223372036854775807_5716, %int1_5717 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %4937, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5718 = torch.constant.int 0
%4938 = torch.aten.unsqueeze %4937, %int0_5718 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %4938, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5719 = torch.constant.int 2
%4939 = torch.aten.unsqueeze %4938, %int2_5719 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4939, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5720 = torch.constant.int 3
%int0_5721 = torch.constant.int 0
%int9223372036854775807_5722 = torch.constant.int 9223372036854775807
%int1_5723 = torch.constant.int 1
%4940 = torch.aten.slice.Tensor %4939, %int3_5720, %int0_5721, %int9223372036854775807_5722, %int1_5723 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %4940, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%4941 = torch_c.to_builtin_tensor %4888 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_5724 = arith.constant 1 : index
%dim_5725 = tensor.dim %4941, %c1_5724 : tensor<4x?x8x128xf16>
%4942 = flow.tensor.bitcast %4941 : tensor<4x?x8x128xf16>{%dim_5725} -> tensor<4x?x8x64xcomplex<f16>>{%dim_5725}
%4943 = torch_c.from_builtin_tensor %4942 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %4943, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%4944 = torch.aten.mul.Tensor %4943, %4940 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %4944, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%4945 = torch_c.to_builtin_tensor %4944 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_5726 = arith.constant 1 : index
%dim_5727 = tensor.dim %4945, %c1_5726 : tensor<4x?x8x64xcomplex<f32>>
%4946 = flow.tensor.bitcast %4945 : tensor<4x?x8x64xcomplex<f32>>{%dim_5727} -> tensor<4x?x8x128xf32>{%dim_5727}
%4947 = torch_c.from_builtin_tensor %4946 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %4947, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_5728 = torch.constant.int 5
%4948 = torch.prims.convert_element_type %4947, %int5_5728 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %4948, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_5729 = torch.constant.int 64
%4949 = torch.aten.mul.Scalar %arg2, %int64_5729 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4949, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int46 = torch.constant.int 46
%int1_5730 = torch.constant.int 1
%4950 = torch.aten.add.Scalar %4949, %int46, %int1_5730 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4950, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5731 = torch.constant.int 4
%int32_5732 = torch.constant.int 32
%int8_5733 = torch.constant.int 8
%int128_5734 = torch.constant.int 128
%4951 = torch.prim.ListConstruct %int4_5731, %425, %int32_5732, %int8_5733, %int128_5734 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4952 = torch.aten.view %4948, %4951 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4952, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5735 = torch.constant.int 4
%4953 = torch.aten.mul.int %int4_5735, %425 : !torch.int, !torch.int -> !torch.int
%int32_5736 = torch.constant.int 32
%int8_5737 = torch.constant.int 8
%int128_5738 = torch.constant.int 128
%4954 = torch.prim.ListConstruct %4953, %int32_5736, %int8_5737, %int128_5738 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4955 = torch.aten.view %4952, %4954 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4955, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5739 = torch.constant.int 4
%4956 = torch.aten.mul.int %int4_5739, %425 : !torch.int, !torch.int -> !torch.int
%4957 = torch.prim.ListConstruct %4956 : (!torch.int) -> !torch.list<int>
%4958 = torch.aten.view %4950, %4957 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4958, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_5740 = torch.constant.int 32
%int2_5741 = torch.constant.int 2
%int32_5742 = torch.constant.int 32
%int8_5743 = torch.constant.int 8
%int128_5744 = torch.constant.int 128
%4959 = torch.prim.ListConstruct %416, %int32_5740, %int2_5741, %int32_5742, %int8_5743, %int128_5744 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4960 = torch.aten.view %4792, %4959 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4960, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5745 = torch.constant.int 32
%4961 = torch.aten.mul.int %416, %int32_5745 : !torch.int, !torch.int -> !torch.int
%int2_5746 = torch.constant.int 2
%4962 = torch.aten.mul.int %4961, %int2_5746 : !torch.int, !torch.int -> !torch.int
%int32_5747 = torch.constant.int 32
%int8_5748 = torch.constant.int 8
%int128_5749 = torch.constant.int 128
%4963 = torch.prim.ListConstruct %4962, %int32_5747, %int8_5748, %int128_5749 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4964 = torch.aten.view %4960, %4963 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4964, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%4965 = torch.prim.ListConstruct %4958 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5750 = torch.constant.bool false
%4966 = torch.aten.index_put %4964, %4965, %4955, %false_5750 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4966, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5751 = torch.constant.int 32
%int2_5752 = torch.constant.int 2
%int32_5753 = torch.constant.int 32
%int8_5754 = torch.constant.int 8
%int128_5755 = torch.constant.int 128
%4967 = torch.prim.ListConstruct %416, %int32_5751, %int2_5752, %int32_5753, %int8_5754, %int128_5755 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4968 = torch.aten.view %4966, %4967 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4968, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5756 = torch.constant.int 2097152
%4969 = torch.prim.ListConstruct %416, %int2097152_5756 : (!torch.int, !torch.int) -> !torch.list<int>
%4970 = torch.aten.view %4968, %4969 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4970, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_5757 = torch.constant.int 32
%int2_5758 = torch.constant.int 2
%int32_5759 = torch.constant.int 32
%int8_5760 = torch.constant.int 8
%int128_5761 = torch.constant.int 128
%4971 = torch.prim.ListConstruct %416, %int32_5757, %int2_5758, %int32_5759, %int8_5760, %int128_5761 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4972 = torch.aten.view %4970, %4971 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4972, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5762 = torch.constant.int 32
%int8_5763 = torch.constant.int 8
%int128_5764 = torch.constant.int 128
%4973 = torch.prim.ListConstruct %4962, %int32_5762, %int8_5763, %int128_5764 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4974 = torch.aten.view %4972, %4973 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4974, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5765 = torch.constant.int 4
%int32_5766 = torch.constant.int 32
%int8_5767 = torch.constant.int 8
%int128_5768 = torch.constant.int 128
%4975 = torch.prim.ListConstruct %int4_5765, %425, %int32_5766, %int8_5767, %int128_5768 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4976 = torch.aten.view %4890, %4975 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %4976, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5769 = torch.constant.int 4
%4977 = torch.aten.mul.int %int4_5769, %425 : !torch.int, !torch.int -> !torch.int
%int32_5770 = torch.constant.int 32
%int8_5771 = torch.constant.int 8
%int128_5772 = torch.constant.int 128
%4978 = torch.prim.ListConstruct %4977, %int32_5770, %int8_5771, %int128_5772 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4979 = torch.aten.view %4976, %4978 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4979, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_5773 = torch.constant.int 1
%int1_5774 = torch.constant.int 1
%4980 = torch.aten.add.Scalar %4950, %int1_5773, %int1_5774 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %4980, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5775 = torch.constant.int 4
%4981 = torch.aten.mul.int %int4_5775, %425 : !torch.int, !torch.int -> !torch.int
%4982 = torch.prim.ListConstruct %4981 : (!torch.int) -> !torch.list<int>
%4983 = torch.aten.view %4980, %4982 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %4983, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%4984 = torch.prim.ListConstruct %4983 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5776 = torch.constant.bool false
%4985 = torch.aten.index_put %4974, %4984, %4979, %false_5776 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %4985, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5777 = torch.constant.int 32
%int2_5778 = torch.constant.int 2
%int32_5779 = torch.constant.int 32
%int8_5780 = torch.constant.int 8
%int128_5781 = torch.constant.int 128
%4986 = torch.prim.ListConstruct %416, %int32_5777, %int2_5778, %int32_5779, %int8_5780, %int128_5781 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4987 = torch.aten.view %4985, %4986 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %4987, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5782 = torch.constant.int 2097152
%4988 = torch.prim.ListConstruct %416, %int2097152_5782 : (!torch.int, !torch.int) -> !torch.list<int>
%4989 = torch.aten.view %4987, %4988 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %4989, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_5783 = torch.constant.int -2
%4990 = torch.aten.unsqueeze %4948, %int-2_5783 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4990, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_5784 = torch.constant.int 4
%int8_5785 = torch.constant.int 8
%int4_5786 = torch.constant.int 4
%int128_5787 = torch.constant.int 128
%4991 = torch.prim.ListConstruct %int4_5784, %4934, %int8_5785, %int4_5786, %int128_5787 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5788 = torch.constant.bool false
%4992 = torch.aten.expand %4990, %4991, %false_5788 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4992, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5789 = torch.constant.int 0
%4993 = torch.aten.clone %4992, %int0_5789 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4993, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5790 = torch.constant.int 4
%int32_5791 = torch.constant.int 32
%int128_5792 = torch.constant.int 128
%4994 = torch.prim.ListConstruct %int4_5790, %4934, %int32_5791, %int128_5792 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%4995 = torch.aten._unsafe_view %4993, %4994 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %4995, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_5793 = torch.constant.int -2
%4996 = torch.aten.unsqueeze %4890, %int-2_5793 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %4996, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_5794 = torch.constant.int 1
%4997 = torch.aten.size.int %4884, %int1_5794 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_5795 = torch.constant.int 4
%int8_5796 = torch.constant.int 8
%int4_5797 = torch.constant.int 4
%int128_5798 = torch.constant.int 128
%4998 = torch.prim.ListConstruct %int4_5795, %4997, %int8_5796, %int4_5797, %int128_5798 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_5799 = torch.constant.bool false
%4999 = torch.aten.expand %4996, %4998, %false_5799 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %4999, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_5800 = torch.constant.int 0
%5000 = torch.aten.clone %4999, %int0_5800 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5000, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_5801 = torch.constant.int 4
%int32_5802 = torch.constant.int 32
%int128_5803 = torch.constant.int 128
%5001 = torch.prim.ListConstruct %int4_5801, %4997, %int32_5802, %int128_5803 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5002 = torch.aten._unsafe_view %5000, %5001 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5002, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_5804 = torch.constant.int 1
%int2_5805 = torch.constant.int 2
%5003 = torch.aten.transpose.int %4919, %int1_5804, %int2_5805 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5003, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5806 = torch.constant.int 1
%int2_5807 = torch.constant.int 2
%5004 = torch.aten.transpose.int %4995, %int1_5806, %int2_5807 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5004, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5808 = torch.constant.int 1
%int2_5809 = torch.constant.int 2
%5005 = torch.aten.transpose.int %5002, %int1_5808, %int2_5809 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5005, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_5810 = torch.constant.float 0.000000e+00
%false_5811 = torch.constant.bool false
%none_5812 = torch.constant.none
%5006:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5003, %5004, %5005, %float0.000000e00_5810, %false_5811, %320, %none_5812) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5006#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_5813 = torch.constant.int 1
%int2_5814 = torch.constant.int 2
%5007 = torch.aten.transpose.int %5006#0, %int1_5813, %int2_5814 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5007, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5815 = torch.constant.int 4
%int4096_5816 = torch.constant.int 4096
%5008 = torch.prim.ListConstruct %int4_5815, %4905, %int4096_5816 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5009 = torch.aten.view %5007, %5008 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5009, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5817 = torch.constant.int -2
%int-1_5818 = torch.constant.int -1
%5010 = torch.aten.transpose.int %212, %int-2_5817, %int-1_5818 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5819 = torch.constant.int 4
%5011 = torch.aten.mul.int %int4_5819, %4905 : !torch.int, !torch.int -> !torch.int
%int4096_5820 = torch.constant.int 4096
%5012 = torch.prim.ListConstruct %5011, %int4096_5820 : (!torch.int, !torch.int) -> !torch.list<int>
%5013 = torch.aten.view %5009, %5012 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5013, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5014 = torch.aten.mm %5013, %5010 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5014, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5821 = torch.constant.int 4
%int4096_5822 = torch.constant.int 4096
%5015 = torch.prim.ListConstruct %int4_5821, %4905, %int4096_5822 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5016 = torch.aten.view %5014, %5015 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5016, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5823 = torch.constant.int 1
%5017 = torch.aten.add.Tensor %4854, %5016, %int1_5823 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5017, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5824 = torch.constant.int 6
%5018 = torch.prims.convert_element_type %5017, %int6_5824 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5018, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5825 = torch.constant.int 2
%5019 = torch.aten.pow.Tensor_Scalar %5018, %int2_5825 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5019, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5826 = torch.constant.int -1
%5020 = torch.prim.ListConstruct %int-1_5826 : (!torch.int) -> !torch.list<int>
%true_5827 = torch.constant.bool true
%none_5828 = torch.constant.none
%5021 = torch.aten.mean.dim %5019, %5020, %true_5827, %none_5828 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5021, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5829 = torch.constant.float 9.9999997473787516E-6
%int1_5830 = torch.constant.int 1
%5022 = torch.aten.add.Scalar %5021, %float9.999990e-06_5829, %int1_5830 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5022, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5023 = torch.aten.rsqrt %5022 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5023, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5024 = torch.aten.mul.Tensor %5018, %5023 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5024, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5025 = torch.aten.mul.Tensor %213, %5024 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5025, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5831 = torch.constant.int 5
%5026 = torch.prims.convert_element_type %5025, %int5_5831 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5026, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5832 = torch.constant.int -2
%int-1_5833 = torch.constant.int -1
%5027 = torch.aten.transpose.int %214, %int-2_5832, %int-1_5833 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5834 = torch.constant.int 4
%5028 = torch.aten.mul.int %int4_5834, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5835 = torch.constant.int 4096
%5029 = torch.prim.ListConstruct %5028, %int4096_5835 : (!torch.int, !torch.int) -> !torch.list<int>
%5030 = torch.aten.view %5026, %5029 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5030, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5031 = torch.aten.mm %5030, %5027 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5031, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5836 = torch.constant.int 4
%int14336_5837 = torch.constant.int 14336
%5032 = torch.prim.ListConstruct %int4_5836, %294, %int14336_5837 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5033 = torch.aten.view %5031, %5032 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5033, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5034 = torch.aten.silu %5033 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5034, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5838 = torch.constant.int -2
%int-1_5839 = torch.constant.int -1
%5035 = torch.aten.transpose.int %215, %int-2_5838, %int-1_5839 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_5840 = torch.constant.int 4
%5036 = torch.aten.mul.int %int4_5840, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5841 = torch.constant.int 4096
%5037 = torch.prim.ListConstruct %5036, %int4096_5841 : (!torch.int, !torch.int) -> !torch.list<int>
%5038 = torch.aten.view %5026, %5037 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5038, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5039 = torch.aten.mm %5038, %5035 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5039, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_5842 = torch.constant.int 4
%int14336_5843 = torch.constant.int 14336
%5040 = torch.prim.ListConstruct %int4_5842, %294, %int14336_5843 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5041 = torch.aten.view %5039, %5040 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5041, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5042 = torch.aten.mul.Tensor %5034, %5041 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5042, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_5844 = torch.constant.int -2
%int-1_5845 = torch.constant.int -1
%5043 = torch.aten.transpose.int %216, %int-2_5844, %int-1_5845 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_5846 = torch.constant.int 1
%5044 = torch.aten.size.int %5033, %int1_5846 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_5847 = torch.constant.int 4
%5045 = torch.aten.mul.int %int4_5847, %5044 : !torch.int, !torch.int -> !torch.int
%int14336_5848 = torch.constant.int 14336
%5046 = torch.prim.ListConstruct %5045, %int14336_5848 : (!torch.int, !torch.int) -> !torch.list<int>
%5047 = torch.aten.view %5042, %5046 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5047, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%5048 = torch.aten.mm %5047, %5043 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5048, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5849 = torch.constant.int 4
%int4096_5850 = torch.constant.int 4096
%5049 = torch.prim.ListConstruct %int4_5849, %5044, %int4096_5850 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5050 = torch.aten.view %5048, %5049 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5050, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_5851 = torch.constant.int 1
%5051 = torch.aten.add.Tensor %5017, %5050, %int1_5851 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5051, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_5852 = torch.constant.int 6
%5052 = torch.prims.convert_element_type %5051, %int6_5852 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5052, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_5853 = torch.constant.int 2
%5053 = torch.aten.pow.Tensor_Scalar %5052, %int2_5853 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5053, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_5854 = torch.constant.int -1
%5054 = torch.prim.ListConstruct %int-1_5854 : (!torch.int) -> !torch.list<int>
%true_5855 = torch.constant.bool true
%none_5856 = torch.constant.none
%5055 = torch.aten.mean.dim %5053, %5054, %true_5855, %none_5856 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5055, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_5857 = torch.constant.float 9.9999997473787516E-6
%int1_5858 = torch.constant.int 1
%5056 = torch.aten.add.Scalar %5055, %float9.999990e-06_5857, %int1_5858 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5056, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5057 = torch.aten.rsqrt %5056 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5057, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5058 = torch.aten.mul.Tensor %5052, %5057 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5058, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5059 = torch.aten.mul.Tensor %217, %5058 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5059, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_5859 = torch.constant.int 5
%5060 = torch.prims.convert_element_type %5059, %int5_5859 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5060, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5860 = torch.constant.int -2
%int-1_5861 = torch.constant.int -1
%5061 = torch.aten.transpose.int %218, %int-2_5860, %int-1_5861 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_5862 = torch.constant.int 4
%5062 = torch.aten.mul.int %int4_5862, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5863 = torch.constant.int 4096
%5063 = torch.prim.ListConstruct %5062, %int4096_5863 : (!torch.int, !torch.int) -> !torch.list<int>
%5064 = torch.aten.view %5060, %5063 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5064, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5065 = torch.aten.mm %5064, %5061 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5065, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_5864 = torch.constant.int 4
%int4096_5865 = torch.constant.int 4096
%5066 = torch.prim.ListConstruct %int4_5864, %294, %int4096_5865 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5067 = torch.aten.view %5065, %5066 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5067, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_5866 = torch.constant.int -2
%int-1_5867 = torch.constant.int -1
%5068 = torch.aten.transpose.int %219, %int-2_5866, %int-1_5867 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5868 = torch.constant.int 4
%5069 = torch.aten.mul.int %int4_5868, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5869 = torch.constant.int 4096
%5070 = torch.prim.ListConstruct %5069, %int4096_5869 : (!torch.int, !torch.int) -> !torch.list<int>
%5071 = torch.aten.view %5060, %5070 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5071, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5072 = torch.aten.mm %5071, %5068 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5072, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5870 = torch.constant.int 4
%int1024_5871 = torch.constant.int 1024
%5073 = torch.prim.ListConstruct %int4_5870, %294, %int1024_5871 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5074 = torch.aten.view %5072, %5073 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5074, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_5872 = torch.constant.int -2
%int-1_5873 = torch.constant.int -1
%5075 = torch.aten.transpose.int %220, %int-2_5872, %int-1_5873 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_5874 = torch.constant.int 4
%5076 = torch.aten.mul.int %int4_5874, %294 : !torch.int, !torch.int -> !torch.int
%int4096_5875 = torch.constant.int 4096
%5077 = torch.prim.ListConstruct %5076, %int4096_5875 : (!torch.int, !torch.int) -> !torch.list<int>
%5078 = torch.aten.view %5060, %5077 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5078, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5079 = torch.aten.mm %5078, %5075 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5079, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_5876 = torch.constant.int 4
%int1024_5877 = torch.constant.int 1024
%5080 = torch.prim.ListConstruct %int4_5876, %294, %int1024_5877 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5081 = torch.aten.view %5079, %5080 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5081, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_5878 = torch.constant.int 4
%int32_5879 = torch.constant.int 32
%int128_5880 = torch.constant.int 128
%5082 = torch.prim.ListConstruct %int4_5878, %294, %int32_5879, %int128_5880 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5083 = torch.aten.view %5067, %5082 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5083, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_5881 = torch.constant.int 4
%int8_5882 = torch.constant.int 8
%int128_5883 = torch.constant.int 128
%5084 = torch.prim.ListConstruct %int4_5881, %294, %int8_5882, %int128_5883 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5085 = torch.aten.view %5074, %5084 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5085, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_5884 = torch.constant.int 4
%int8_5885 = torch.constant.int 8
%int128_5886 = torch.constant.int 128
%5086 = torch.prim.ListConstruct %int4_5884, %294, %int8_5885, %int128_5886 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5087 = torch.aten.view %5081, %5086 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5087, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_5887 = torch.constant.int 131072
%none_5888 = torch.constant.none
%none_5889 = torch.constant.none
%cpu_5890 = torch.constant.device "cpu"
%false_5891 = torch.constant.bool false
%5088 = torch.aten.arange %int131072_5887, %none_5888, %none_5889, %cpu_5890, %false_5891 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5892 = torch.constant.int 0
%int128_5893 = torch.constant.int 128
%int2_5894 = torch.constant.int 2
%none_5895 = torch.constant.none
%none_5896 = torch.constant.none
%cpu_5897 = torch.constant.device "cpu"
%false_5898 = torch.constant.bool false
%5089 = torch.aten.arange.start_step %int0_5892, %int128_5893, %int2_5894, %none_5895, %none_5896, %cpu_5897, %false_5898 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5899 = torch.constant.int 0
%int0_5900 = torch.constant.int 0
%int64_5901 = torch.constant.int 64
%int1_5902 = torch.constant.int 1
%5090 = torch.aten.slice.Tensor %5089, %int0_5899, %int0_5900, %int64_5901, %int1_5902 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5903 = torch.constant.int 6
%5091 = torch.prims.convert_element_type %5090, %int6_5903 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5904 = torch.constant.int 128
%5092 = torch.aten.div.Scalar %5091, %int128_5904 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5905 = torch.constant.float 5.000000e+05
%5093 = torch.aten.pow.Scalar %float5.000000e05_5905, %5092 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5094 = torch.aten.reciprocal %5093 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5906 = torch.constant.float 1.000000e+00
%5095 = torch.aten.mul.Scalar %5094, %float1.000000e00_5906 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5907 = torch.constant.int 131072
%int1_5908 = torch.constant.int 1
%5096 = torch.prim.ListConstruct %int131072_5907, %int1_5908 : (!torch.int, !torch.int) -> !torch.list<int>
%5097 = torch.aten.view %5088, %5096 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5098 = torch.aten.mul.Tensor %5097, %5095 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5099 = torch.aten.cos %5098 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5100 = torch.aten.sin %5098 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5101 = torch.aten.complex %5099, %5100 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5909 = torch.constant.int 1
%5102 = torch.aten.size.int %5067, %int1_5909 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_5910 = torch.constant.int 0
%5103 = torch.aten.add.int %int0_5910, %5102 : !torch.int, !torch.int -> !torch.int
%int0_5911 = torch.constant.int 0
%int0_5912 = torch.constant.int 0
%int1_5913 = torch.constant.int 1
%5104 = torch.aten.slice.Tensor %5101, %int0_5911, %int0_5912, %5103, %int1_5913 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5104, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5914 = torch.constant.int 1
%int0_5915 = torch.constant.int 0
%int9223372036854775807_5916 = torch.constant.int 9223372036854775807
%int1_5917 = torch.constant.int 1
%5105 = torch.aten.slice.Tensor %5104, %int1_5914, %int0_5915, %int9223372036854775807_5916, %int1_5917 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5105, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5918 = torch.constant.int 0
%5106 = torch.aten.unsqueeze %5105, %int0_5918 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5106, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5919 = torch.constant.int 2
%5107 = torch.aten.unsqueeze %5106, %int2_5919 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5107, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5920 = torch.constant.int 3
%int0_5921 = torch.constant.int 0
%int9223372036854775807_5922 = torch.constant.int 9223372036854775807
%int1_5923 = torch.constant.int 1
%5108 = torch.aten.slice.Tensor %5107, %int3_5920, %int0_5921, %int9223372036854775807_5922, %int1_5923 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5108, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5109 = torch_c.to_builtin_tensor %5083 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_5924 = arith.constant 1 : index
%dim_5925 = tensor.dim %5109, %c1_5924 : tensor<4x?x32x128xf16>
%5110 = flow.tensor.bitcast %5109 : tensor<4x?x32x128xf16>{%dim_5925} -> tensor<4x?x32x64xcomplex<f16>>{%dim_5925}
%5111 = torch_c.from_builtin_tensor %5110 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %5111, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%5112 = torch.aten.mul.Tensor %5111, %5108 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %5112, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%5113 = torch_c.to_builtin_tensor %5112 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_5926 = arith.constant 1 : index
%dim_5927 = tensor.dim %5113, %c1_5926 : tensor<4x?x32x64xcomplex<f32>>
%5114 = flow.tensor.bitcast %5113 : tensor<4x?x32x64xcomplex<f32>>{%dim_5927} -> tensor<4x?x32x128xf32>{%dim_5927}
%5115 = torch_c.from_builtin_tensor %5114 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %5115, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_5928 = torch.constant.int 5
%5116 = torch.prims.convert_element_type %5115, %int5_5928 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5116, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_5929 = torch.constant.int 131072
%none_5930 = torch.constant.none
%none_5931 = torch.constant.none
%cpu_5932 = torch.constant.device "cpu"
%false_5933 = torch.constant.bool false
%5117 = torch.aten.arange %int131072_5929, %none_5930, %none_5931, %cpu_5932, %false_5933 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_5934 = torch.constant.int 0
%int128_5935 = torch.constant.int 128
%int2_5936 = torch.constant.int 2
%none_5937 = torch.constant.none
%none_5938 = torch.constant.none
%cpu_5939 = torch.constant.device "cpu"
%false_5940 = torch.constant.bool false
%5118 = torch.aten.arange.start_step %int0_5934, %int128_5935, %int2_5936, %none_5937, %none_5938, %cpu_5939, %false_5940 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_5941 = torch.constant.int 0
%int0_5942 = torch.constant.int 0
%int64_5943 = torch.constant.int 64
%int1_5944 = torch.constant.int 1
%5119 = torch.aten.slice.Tensor %5118, %int0_5941, %int0_5942, %int64_5943, %int1_5944 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_5945 = torch.constant.int 6
%5120 = torch.prims.convert_element_type %5119, %int6_5945 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_5946 = torch.constant.int 128
%5121 = torch.aten.div.Scalar %5120, %int128_5946 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_5947 = torch.constant.float 5.000000e+05
%5122 = torch.aten.pow.Scalar %float5.000000e05_5947, %5121 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5123 = torch.aten.reciprocal %5122 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_5948 = torch.constant.float 1.000000e+00
%5124 = torch.aten.mul.Scalar %5123, %float1.000000e00_5948 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_5949 = torch.constant.int 131072
%int1_5950 = torch.constant.int 1
%5125 = torch.prim.ListConstruct %int131072_5949, %int1_5950 : (!torch.int, !torch.int) -> !torch.list<int>
%5126 = torch.aten.view %5117, %5125 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5127 = torch.aten.mul.Tensor %5126, %5124 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5128 = torch.aten.cos %5127 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5129 = torch.aten.sin %5127 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5130 = torch.aten.complex %5128, %5129 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_5951 = torch.constant.int 1
%5131 = torch.aten.size.int %5074, %int1_5951 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_5952 = torch.constant.int 0
%5132 = torch.aten.add.int %int0_5952, %5131 : !torch.int, !torch.int -> !torch.int
%int0_5953 = torch.constant.int 0
%int0_5954 = torch.constant.int 0
%int1_5955 = torch.constant.int 1
%5133 = torch.aten.slice.Tensor %5130, %int0_5953, %int0_5954, %5132, %int1_5955 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5133, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_5956 = torch.constant.int 1
%int0_5957 = torch.constant.int 0
%int9223372036854775807_5958 = torch.constant.int 9223372036854775807
%int1_5959 = torch.constant.int 1
%5134 = torch.aten.slice.Tensor %5133, %int1_5956, %int0_5957, %int9223372036854775807_5958, %int1_5959 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5134, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_5960 = torch.constant.int 0
%5135 = torch.aten.unsqueeze %5134, %int0_5960 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5135, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_5961 = torch.constant.int 2
%5136 = torch.aten.unsqueeze %5135, %int2_5961 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5136, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_5962 = torch.constant.int 3
%int0_5963 = torch.constant.int 0
%int9223372036854775807_5964 = torch.constant.int 9223372036854775807
%int1_5965 = torch.constant.int 1
%5137 = torch.aten.slice.Tensor %5136, %int3_5962, %int0_5963, %int9223372036854775807_5964, %int1_5965 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5137, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5138 = torch_c.to_builtin_tensor %5085 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_5966 = arith.constant 1 : index
%dim_5967 = tensor.dim %5138, %c1_5966 : tensor<4x?x8x128xf16>
%5139 = flow.tensor.bitcast %5138 : tensor<4x?x8x128xf16>{%dim_5967} -> tensor<4x?x8x64xcomplex<f16>>{%dim_5967}
%5140 = torch_c.from_builtin_tensor %5139 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %5140, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%5141 = torch.aten.mul.Tensor %5140, %5137 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %5141, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%5142 = torch_c.to_builtin_tensor %5141 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_5968 = arith.constant 1 : index
%dim_5969 = tensor.dim %5142, %c1_5968 : tensor<4x?x8x64xcomplex<f32>>
%5143 = flow.tensor.bitcast %5142 : tensor<4x?x8x64xcomplex<f32>>{%dim_5969} -> tensor<4x?x8x128xf32>{%dim_5969}
%5144 = torch_c.from_builtin_tensor %5143 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %5144, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_5970 = torch.constant.int 5
%5145 = torch.prims.convert_element_type %5144, %int5_5970 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5145, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_5971 = torch.constant.int 64
%5146 = torch.aten.mul.Scalar %arg2, %int64_5971 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5146, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int48 = torch.constant.int 48
%int1_5972 = torch.constant.int 1
%5147 = torch.aten.add.Scalar %5146, %int48, %int1_5972 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5147, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_5973 = torch.constant.int 4
%int32_5974 = torch.constant.int 32
%int8_5975 = torch.constant.int 8
%int128_5976 = torch.constant.int 128
%5148 = torch.prim.ListConstruct %int4_5973, %425, %int32_5974, %int8_5975, %int128_5976 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5149 = torch.aten.view %5145, %5148 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5149, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_5977 = torch.constant.int 4
%5150 = torch.aten.mul.int %int4_5977, %425 : !torch.int, !torch.int -> !torch.int
%int32_5978 = torch.constant.int 32
%int8_5979 = torch.constant.int 8
%int128_5980 = torch.constant.int 128
%5151 = torch.prim.ListConstruct %5150, %int32_5978, %int8_5979, %int128_5980 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5152 = torch.aten.view %5149, %5151 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5152, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_5981 = torch.constant.int 4
%5153 = torch.aten.mul.int %int4_5981, %425 : !torch.int, !torch.int -> !torch.int
%5154 = torch.prim.ListConstruct %5153 : (!torch.int) -> !torch.list<int>
%5155 = torch.aten.view %5147, %5154 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5155, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_5982 = torch.constant.int 32
%int2_5983 = torch.constant.int 2
%int32_5984 = torch.constant.int 32
%int8_5985 = torch.constant.int 8
%int128_5986 = torch.constant.int 128
%5156 = torch.prim.ListConstruct %416, %int32_5982, %int2_5983, %int32_5984, %int8_5985, %int128_5986 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5157 = torch.aten.view %4989, %5156 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5157, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_5987 = torch.constant.int 32
%5158 = torch.aten.mul.int %416, %int32_5987 : !torch.int, !torch.int -> !torch.int
%int2_5988 = torch.constant.int 2
%5159 = torch.aten.mul.int %5158, %int2_5988 : !torch.int, !torch.int -> !torch.int
%int32_5989 = torch.constant.int 32
%int8_5990 = torch.constant.int 8
%int128_5991 = torch.constant.int 128
%5160 = torch.prim.ListConstruct %5159, %int32_5989, %int8_5990, %int128_5991 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5161 = torch.aten.view %5157, %5160 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5161, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%5162 = torch.prim.ListConstruct %5155 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_5992 = torch.constant.bool false
%5163 = torch.aten.index_put %5161, %5162, %5152, %false_5992 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5163, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_5993 = torch.constant.int 32
%int2_5994 = torch.constant.int 2
%int32_5995 = torch.constant.int 32
%int8_5996 = torch.constant.int 8
%int128_5997 = torch.constant.int 128
%5164 = torch.prim.ListConstruct %416, %int32_5993, %int2_5994, %int32_5995, %int8_5996, %int128_5997 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5165 = torch.aten.view %5163, %5164 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5165, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_5998 = torch.constant.int 2097152
%5166 = torch.prim.ListConstruct %416, %int2097152_5998 : (!torch.int, !torch.int) -> !torch.list<int>
%5167 = torch.aten.view %5165, %5166 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5167, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_5999 = torch.constant.int 32
%int2_6000 = torch.constant.int 2
%int32_6001 = torch.constant.int 32
%int8_6002 = torch.constant.int 8
%int128_6003 = torch.constant.int 128
%5168 = torch.prim.ListConstruct %416, %int32_5999, %int2_6000, %int32_6001, %int8_6002, %int128_6003 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5169 = torch.aten.view %5167, %5168 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5169, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6004 = torch.constant.int 32
%int8_6005 = torch.constant.int 8
%int128_6006 = torch.constant.int 128
%5170 = torch.prim.ListConstruct %5159, %int32_6004, %int8_6005, %int128_6006 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5171 = torch.aten.view %5169, %5170 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5171, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6007 = torch.constant.int 4
%int32_6008 = torch.constant.int 32
%int8_6009 = torch.constant.int 8
%int128_6010 = torch.constant.int 128
%5172 = torch.prim.ListConstruct %int4_6007, %425, %int32_6008, %int8_6009, %int128_6010 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5173 = torch.aten.view %5087, %5172 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5173, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6011 = torch.constant.int 4
%5174 = torch.aten.mul.int %int4_6011, %425 : !torch.int, !torch.int -> !torch.int
%int32_6012 = torch.constant.int 32
%int8_6013 = torch.constant.int 8
%int128_6014 = torch.constant.int 128
%5175 = torch.prim.ListConstruct %5174, %int32_6012, %int8_6013, %int128_6014 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5176 = torch.aten.view %5173, %5175 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5176, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_6015 = torch.constant.int 1
%int1_6016 = torch.constant.int 1
%5177 = torch.aten.add.Scalar %5147, %int1_6015, %int1_6016 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5177, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6017 = torch.constant.int 4
%5178 = torch.aten.mul.int %int4_6017, %425 : !torch.int, !torch.int -> !torch.int
%5179 = torch.prim.ListConstruct %5178 : (!torch.int) -> !torch.list<int>
%5180 = torch.aten.view %5177, %5179 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5180, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%5181 = torch.prim.ListConstruct %5180 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6018 = torch.constant.bool false
%5182 = torch.aten.index_put %5171, %5181, %5176, %false_6018 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5182, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6019 = torch.constant.int 32
%int2_6020 = torch.constant.int 2
%int32_6021 = torch.constant.int 32
%int8_6022 = torch.constant.int 8
%int128_6023 = torch.constant.int 128
%5183 = torch.prim.ListConstruct %416, %int32_6019, %int2_6020, %int32_6021, %int8_6022, %int128_6023 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5184 = torch.aten.view %5182, %5183 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5184, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6024 = torch.constant.int 2097152
%5185 = torch.prim.ListConstruct %416, %int2097152_6024 : (!torch.int, !torch.int) -> !torch.list<int>
%5186 = torch.aten.view %5184, %5185 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5186, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_6025 = torch.constant.int -2
%5187 = torch.aten.unsqueeze %5145, %int-2_6025 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5187, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_6026 = torch.constant.int 4
%int8_6027 = torch.constant.int 8
%int4_6028 = torch.constant.int 4
%int128_6029 = torch.constant.int 128
%5188 = torch.prim.ListConstruct %int4_6026, %5131, %int8_6027, %int4_6028, %int128_6029 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6030 = torch.constant.bool false
%5189 = torch.aten.expand %5187, %5188, %false_6030 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5189, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6031 = torch.constant.int 0
%5190 = torch.aten.clone %5189, %int0_6031 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5190, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6032 = torch.constant.int 4
%int32_6033 = torch.constant.int 32
%int128_6034 = torch.constant.int 128
%5191 = torch.prim.ListConstruct %int4_6032, %5131, %int32_6033, %int128_6034 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5192 = torch.aten._unsafe_view %5190, %5191 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5192, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_6035 = torch.constant.int -2
%5193 = torch.aten.unsqueeze %5087, %int-2_6035 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5193, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_6036 = torch.constant.int 1
%5194 = torch.aten.size.int %5081, %int1_6036 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_6037 = torch.constant.int 4
%int8_6038 = torch.constant.int 8
%int4_6039 = torch.constant.int 4
%int128_6040 = torch.constant.int 128
%5195 = torch.prim.ListConstruct %int4_6037, %5194, %int8_6038, %int4_6039, %int128_6040 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6041 = torch.constant.bool false
%5196 = torch.aten.expand %5193, %5195, %false_6041 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5196, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6042 = torch.constant.int 0
%5197 = torch.aten.clone %5196, %int0_6042 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5197, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6043 = torch.constant.int 4
%int32_6044 = torch.constant.int 32
%int128_6045 = torch.constant.int 128
%5198 = torch.prim.ListConstruct %int4_6043, %5194, %int32_6044, %int128_6045 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5199 = torch.aten._unsafe_view %5197, %5198 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5199, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_6046 = torch.constant.int 1
%int2_6047 = torch.constant.int 2
%5200 = torch.aten.transpose.int %5116, %int1_6046, %int2_6047 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5200, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6048 = torch.constant.int 1
%int2_6049 = torch.constant.int 2
%5201 = torch.aten.transpose.int %5192, %int1_6048, %int2_6049 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5201, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6050 = torch.constant.int 1
%int2_6051 = torch.constant.int 2
%5202 = torch.aten.transpose.int %5199, %int1_6050, %int2_6051 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5202, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_6052 = torch.constant.float 0.000000e+00
%false_6053 = torch.constant.bool false
%none_6054 = torch.constant.none
%5203:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5200, %5201, %5202, %float0.000000e00_6052, %false_6053, %320, %none_6054) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5203#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6055 = torch.constant.int 1
%int2_6056 = torch.constant.int 2
%5204 = torch.aten.transpose.int %5203#0, %int1_6055, %int2_6056 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5204, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6057 = torch.constant.int 4
%int4096_6058 = torch.constant.int 4096
%5205 = torch.prim.ListConstruct %int4_6057, %5102, %int4096_6058 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5206 = torch.aten.view %5204, %5205 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5206, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6059 = torch.constant.int -2
%int-1_6060 = torch.constant.int -1
%5207 = torch.aten.transpose.int %221, %int-2_6059, %int-1_6060 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6061 = torch.constant.int 4
%5208 = torch.aten.mul.int %int4_6061, %5102 : !torch.int, !torch.int -> !torch.int
%int4096_6062 = torch.constant.int 4096
%5209 = torch.prim.ListConstruct %5208, %int4096_6062 : (!torch.int, !torch.int) -> !torch.list<int>
%5210 = torch.aten.view %5206, %5209 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5210, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5211 = torch.aten.mm %5210, %5207 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5211, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6063 = torch.constant.int 4
%int4096_6064 = torch.constant.int 4096
%5212 = torch.prim.ListConstruct %int4_6063, %5102, %int4096_6064 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5213 = torch.aten.view %5211, %5212 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5213, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6065 = torch.constant.int 1
%5214 = torch.aten.add.Tensor %5051, %5213, %int1_6065 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5214, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6066 = torch.constant.int 6
%5215 = torch.prims.convert_element_type %5214, %int6_6066 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5215, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6067 = torch.constant.int 2
%5216 = torch.aten.pow.Tensor_Scalar %5215, %int2_6067 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5216, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6068 = torch.constant.int -1
%5217 = torch.prim.ListConstruct %int-1_6068 : (!torch.int) -> !torch.list<int>
%true_6069 = torch.constant.bool true
%none_6070 = torch.constant.none
%5218 = torch.aten.mean.dim %5216, %5217, %true_6069, %none_6070 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5218, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6071 = torch.constant.float 9.9999997473787516E-6
%int1_6072 = torch.constant.int 1
%5219 = torch.aten.add.Scalar %5218, %float9.999990e-06_6071, %int1_6072 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5219, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5220 = torch.aten.rsqrt %5219 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5220, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5221 = torch.aten.mul.Tensor %5215, %5220 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5221, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5222 = torch.aten.mul.Tensor %222, %5221 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5222, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6073 = torch.constant.int 5
%5223 = torch.prims.convert_element_type %5222, %int5_6073 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5223, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6074 = torch.constant.int -2
%int-1_6075 = torch.constant.int -1
%5224 = torch.aten.transpose.int %223, %int-2_6074, %int-1_6075 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6076 = torch.constant.int 4
%5225 = torch.aten.mul.int %int4_6076, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6077 = torch.constant.int 4096
%5226 = torch.prim.ListConstruct %5225, %int4096_6077 : (!torch.int, !torch.int) -> !torch.list<int>
%5227 = torch.aten.view %5223, %5226 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5227, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5228 = torch.aten.mm %5227, %5224 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5228, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6078 = torch.constant.int 4
%int14336_6079 = torch.constant.int 14336
%5229 = torch.prim.ListConstruct %int4_6078, %294, %int14336_6079 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5230 = torch.aten.view %5228, %5229 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5230, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5231 = torch.aten.silu %5230 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5231, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6080 = torch.constant.int -2
%int-1_6081 = torch.constant.int -1
%5232 = torch.aten.transpose.int %224, %int-2_6080, %int-1_6081 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6082 = torch.constant.int 4
%5233 = torch.aten.mul.int %int4_6082, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6083 = torch.constant.int 4096
%5234 = torch.prim.ListConstruct %5233, %int4096_6083 : (!torch.int, !torch.int) -> !torch.list<int>
%5235 = torch.aten.view %5223, %5234 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5235, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5236 = torch.aten.mm %5235, %5232 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5236, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6084 = torch.constant.int 4
%int14336_6085 = torch.constant.int 14336
%5237 = torch.prim.ListConstruct %int4_6084, %294, %int14336_6085 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5238 = torch.aten.view %5236, %5237 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5238, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5239 = torch.aten.mul.Tensor %5231, %5238 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5239, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6086 = torch.constant.int -2
%int-1_6087 = torch.constant.int -1
%5240 = torch.aten.transpose.int %225, %int-2_6086, %int-1_6087 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_6088 = torch.constant.int 1
%5241 = torch.aten.size.int %5230, %int1_6088 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_6089 = torch.constant.int 4
%5242 = torch.aten.mul.int %int4_6089, %5241 : !torch.int, !torch.int -> !torch.int
%int14336_6090 = torch.constant.int 14336
%5243 = torch.prim.ListConstruct %5242, %int14336_6090 : (!torch.int, !torch.int) -> !torch.list<int>
%5244 = torch.aten.view %5239, %5243 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5244, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%5245 = torch.aten.mm %5244, %5240 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5245, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6091 = torch.constant.int 4
%int4096_6092 = torch.constant.int 4096
%5246 = torch.prim.ListConstruct %int4_6091, %5241, %int4096_6092 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5247 = torch.aten.view %5245, %5246 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5247, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6093 = torch.constant.int 1
%5248 = torch.aten.add.Tensor %5214, %5247, %int1_6093 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5248, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6094 = torch.constant.int 6
%5249 = torch.prims.convert_element_type %5248, %int6_6094 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5249, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6095 = torch.constant.int 2
%5250 = torch.aten.pow.Tensor_Scalar %5249, %int2_6095 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5250, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6096 = torch.constant.int -1
%5251 = torch.prim.ListConstruct %int-1_6096 : (!torch.int) -> !torch.list<int>
%true_6097 = torch.constant.bool true
%none_6098 = torch.constant.none
%5252 = torch.aten.mean.dim %5250, %5251, %true_6097, %none_6098 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5252, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6099 = torch.constant.float 9.9999997473787516E-6
%int1_6100 = torch.constant.int 1
%5253 = torch.aten.add.Scalar %5252, %float9.999990e-06_6099, %int1_6100 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5253, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5254 = torch.aten.rsqrt %5253 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5254, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5255 = torch.aten.mul.Tensor %5249, %5254 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5255, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5256 = torch.aten.mul.Tensor %226, %5255 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5256, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6101 = torch.constant.int 5
%5257 = torch.prims.convert_element_type %5256, %int5_6101 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5257, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6102 = torch.constant.int -2
%int-1_6103 = torch.constant.int -1
%5258 = torch.aten.transpose.int %227, %int-2_6102, %int-1_6103 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6104 = torch.constant.int 4
%5259 = torch.aten.mul.int %int4_6104, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6105 = torch.constant.int 4096
%5260 = torch.prim.ListConstruct %5259, %int4096_6105 : (!torch.int, !torch.int) -> !torch.list<int>
%5261 = torch.aten.view %5257, %5260 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5261, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5262 = torch.aten.mm %5261, %5258 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5262, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6106 = torch.constant.int 4
%int4096_6107 = torch.constant.int 4096
%5263 = torch.prim.ListConstruct %int4_6106, %294, %int4096_6107 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5264 = torch.aten.view %5262, %5263 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5264, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6108 = torch.constant.int -2
%int-1_6109 = torch.constant.int -1
%5265 = torch.aten.transpose.int %228, %int-2_6108, %int-1_6109 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6110 = torch.constant.int 4
%5266 = torch.aten.mul.int %int4_6110, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6111 = torch.constant.int 4096
%5267 = torch.prim.ListConstruct %5266, %int4096_6111 : (!torch.int, !torch.int) -> !torch.list<int>
%5268 = torch.aten.view %5257, %5267 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5268, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5269 = torch.aten.mm %5268, %5265 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5269, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6112 = torch.constant.int 4
%int1024_6113 = torch.constant.int 1024
%5270 = torch.prim.ListConstruct %int4_6112, %294, %int1024_6113 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5271 = torch.aten.view %5269, %5270 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5271, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_6114 = torch.constant.int -2
%int-1_6115 = torch.constant.int -1
%5272 = torch.aten.transpose.int %229, %int-2_6114, %int-1_6115 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6116 = torch.constant.int 4
%5273 = torch.aten.mul.int %int4_6116, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6117 = torch.constant.int 4096
%5274 = torch.prim.ListConstruct %5273, %int4096_6117 : (!torch.int, !torch.int) -> !torch.list<int>
%5275 = torch.aten.view %5257, %5274 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5275, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5276 = torch.aten.mm %5275, %5272 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5276, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6118 = torch.constant.int 4
%int1024_6119 = torch.constant.int 1024
%5277 = torch.prim.ListConstruct %int4_6118, %294, %int1024_6119 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5278 = torch.aten.view %5276, %5277 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5278, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_6120 = torch.constant.int 4
%int32_6121 = torch.constant.int 32
%int128_6122 = torch.constant.int 128
%5279 = torch.prim.ListConstruct %int4_6120, %294, %int32_6121, %int128_6122 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5280 = torch.aten.view %5264, %5279 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5280, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6123 = torch.constant.int 4
%int8_6124 = torch.constant.int 8
%int128_6125 = torch.constant.int 128
%5281 = torch.prim.ListConstruct %int4_6123, %294, %int8_6124, %int128_6125 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5282 = torch.aten.view %5271, %5281 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5282, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_6126 = torch.constant.int 4
%int8_6127 = torch.constant.int 8
%int128_6128 = torch.constant.int 128
%5283 = torch.prim.ListConstruct %int4_6126, %294, %int8_6127, %int128_6128 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5284 = torch.aten.view %5278, %5283 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5284, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_6129 = torch.constant.int 131072
%none_6130 = torch.constant.none
%none_6131 = torch.constant.none
%cpu_6132 = torch.constant.device "cpu"
%false_6133 = torch.constant.bool false
%5285 = torch.aten.arange %int131072_6129, %none_6130, %none_6131, %cpu_6132, %false_6133 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6134 = torch.constant.int 0
%int128_6135 = torch.constant.int 128
%int2_6136 = torch.constant.int 2
%none_6137 = torch.constant.none
%none_6138 = torch.constant.none
%cpu_6139 = torch.constant.device "cpu"
%false_6140 = torch.constant.bool false
%5286 = torch.aten.arange.start_step %int0_6134, %int128_6135, %int2_6136, %none_6137, %none_6138, %cpu_6139, %false_6140 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6141 = torch.constant.int 0
%int0_6142 = torch.constant.int 0
%int64_6143 = torch.constant.int 64
%int1_6144 = torch.constant.int 1
%5287 = torch.aten.slice.Tensor %5286, %int0_6141, %int0_6142, %int64_6143, %int1_6144 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6145 = torch.constant.int 6
%5288 = torch.prims.convert_element_type %5287, %int6_6145 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6146 = torch.constant.int 128
%5289 = torch.aten.div.Scalar %5288, %int128_6146 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6147 = torch.constant.float 5.000000e+05
%5290 = torch.aten.pow.Scalar %float5.000000e05_6147, %5289 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5291 = torch.aten.reciprocal %5290 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6148 = torch.constant.float 1.000000e+00
%5292 = torch.aten.mul.Scalar %5291, %float1.000000e00_6148 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6149 = torch.constant.int 131072
%int1_6150 = torch.constant.int 1
%5293 = torch.prim.ListConstruct %int131072_6149, %int1_6150 : (!torch.int, !torch.int) -> !torch.list<int>
%5294 = torch.aten.view %5285, %5293 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5295 = torch.aten.mul.Tensor %5294, %5292 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5296 = torch.aten.cos %5295 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5297 = torch.aten.sin %5295 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5298 = torch.aten.complex %5296, %5297 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6151 = torch.constant.int 1
%5299 = torch.aten.size.int %5264, %int1_6151 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_6152 = torch.constant.int 0
%5300 = torch.aten.add.int %int0_6152, %5299 : !torch.int, !torch.int -> !torch.int
%int0_6153 = torch.constant.int 0
%int0_6154 = torch.constant.int 0
%int1_6155 = torch.constant.int 1
%5301 = torch.aten.slice.Tensor %5298, %int0_6153, %int0_6154, %5300, %int1_6155 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5301, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6156 = torch.constant.int 1
%int0_6157 = torch.constant.int 0
%int9223372036854775807_6158 = torch.constant.int 9223372036854775807
%int1_6159 = torch.constant.int 1
%5302 = torch.aten.slice.Tensor %5301, %int1_6156, %int0_6157, %int9223372036854775807_6158, %int1_6159 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5302, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6160 = torch.constant.int 0
%5303 = torch.aten.unsqueeze %5302, %int0_6160 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5303, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6161 = torch.constant.int 2
%5304 = torch.aten.unsqueeze %5303, %int2_6161 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5304, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6162 = torch.constant.int 3
%int0_6163 = torch.constant.int 0
%int9223372036854775807_6164 = torch.constant.int 9223372036854775807
%int1_6165 = torch.constant.int 1
%5305 = torch.aten.slice.Tensor %5304, %int3_6162, %int0_6163, %int9223372036854775807_6164, %int1_6165 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5305, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5306 = torch_c.to_builtin_tensor %5280 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_6166 = arith.constant 1 : index
%dim_6167 = tensor.dim %5306, %c1_6166 : tensor<4x?x32x128xf16>
%5307 = flow.tensor.bitcast %5306 : tensor<4x?x32x128xf16>{%dim_6167} -> tensor<4x?x32x64xcomplex<f16>>{%dim_6167}
%5308 = torch_c.from_builtin_tensor %5307 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %5308, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%5309 = torch.aten.mul.Tensor %5308, %5305 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %5309, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%5310 = torch_c.to_builtin_tensor %5309 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_6168 = arith.constant 1 : index
%dim_6169 = tensor.dim %5310, %c1_6168 : tensor<4x?x32x64xcomplex<f32>>
%5311 = flow.tensor.bitcast %5310 : tensor<4x?x32x64xcomplex<f32>>{%dim_6169} -> tensor<4x?x32x128xf32>{%dim_6169}
%5312 = torch_c.from_builtin_tensor %5311 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %5312, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_6170 = torch.constant.int 5
%5313 = torch.prims.convert_element_type %5312, %int5_6170 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5313, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_6171 = torch.constant.int 131072
%none_6172 = torch.constant.none
%none_6173 = torch.constant.none
%cpu_6174 = torch.constant.device "cpu"
%false_6175 = torch.constant.bool false
%5314 = torch.aten.arange %int131072_6171, %none_6172, %none_6173, %cpu_6174, %false_6175 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6176 = torch.constant.int 0
%int128_6177 = torch.constant.int 128
%int2_6178 = torch.constant.int 2
%none_6179 = torch.constant.none
%none_6180 = torch.constant.none
%cpu_6181 = torch.constant.device "cpu"
%false_6182 = torch.constant.bool false
%5315 = torch.aten.arange.start_step %int0_6176, %int128_6177, %int2_6178, %none_6179, %none_6180, %cpu_6181, %false_6182 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6183 = torch.constant.int 0
%int0_6184 = torch.constant.int 0
%int64_6185 = torch.constant.int 64
%int1_6186 = torch.constant.int 1
%5316 = torch.aten.slice.Tensor %5315, %int0_6183, %int0_6184, %int64_6185, %int1_6186 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6187 = torch.constant.int 6
%5317 = torch.prims.convert_element_type %5316, %int6_6187 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6188 = torch.constant.int 128
%5318 = torch.aten.div.Scalar %5317, %int128_6188 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6189 = torch.constant.float 5.000000e+05
%5319 = torch.aten.pow.Scalar %float5.000000e05_6189, %5318 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5320 = torch.aten.reciprocal %5319 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6190 = torch.constant.float 1.000000e+00
%5321 = torch.aten.mul.Scalar %5320, %float1.000000e00_6190 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6191 = torch.constant.int 131072
%int1_6192 = torch.constant.int 1
%5322 = torch.prim.ListConstruct %int131072_6191, %int1_6192 : (!torch.int, !torch.int) -> !torch.list<int>
%5323 = torch.aten.view %5314, %5322 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5324 = torch.aten.mul.Tensor %5323, %5321 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5325 = torch.aten.cos %5324 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5326 = torch.aten.sin %5324 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5327 = torch.aten.complex %5325, %5326 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6193 = torch.constant.int 1
%5328 = torch.aten.size.int %5271, %int1_6193 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_6194 = torch.constant.int 0
%5329 = torch.aten.add.int %int0_6194, %5328 : !torch.int, !torch.int -> !torch.int
%int0_6195 = torch.constant.int 0
%int0_6196 = torch.constant.int 0
%int1_6197 = torch.constant.int 1
%5330 = torch.aten.slice.Tensor %5327, %int0_6195, %int0_6196, %5329, %int1_6197 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5330, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6198 = torch.constant.int 1
%int0_6199 = torch.constant.int 0
%int9223372036854775807_6200 = torch.constant.int 9223372036854775807
%int1_6201 = torch.constant.int 1
%5331 = torch.aten.slice.Tensor %5330, %int1_6198, %int0_6199, %int9223372036854775807_6200, %int1_6201 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5331, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6202 = torch.constant.int 0
%5332 = torch.aten.unsqueeze %5331, %int0_6202 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5332, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6203 = torch.constant.int 2
%5333 = torch.aten.unsqueeze %5332, %int2_6203 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5333, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6204 = torch.constant.int 3
%int0_6205 = torch.constant.int 0
%int9223372036854775807_6206 = torch.constant.int 9223372036854775807
%int1_6207 = torch.constant.int 1
%5334 = torch.aten.slice.Tensor %5333, %int3_6204, %int0_6205, %int9223372036854775807_6206, %int1_6207 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5334, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5335 = torch_c.to_builtin_tensor %5282 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_6208 = arith.constant 1 : index
%dim_6209 = tensor.dim %5335, %c1_6208 : tensor<4x?x8x128xf16>
%5336 = flow.tensor.bitcast %5335 : tensor<4x?x8x128xf16>{%dim_6209} -> tensor<4x?x8x64xcomplex<f16>>{%dim_6209}
%5337 = torch_c.from_builtin_tensor %5336 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %5337, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%5338 = torch.aten.mul.Tensor %5337, %5334 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %5338, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%5339 = torch_c.to_builtin_tensor %5338 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_6210 = arith.constant 1 : index
%dim_6211 = tensor.dim %5339, %c1_6210 : tensor<4x?x8x64xcomplex<f32>>
%5340 = flow.tensor.bitcast %5339 : tensor<4x?x8x64xcomplex<f32>>{%dim_6211} -> tensor<4x?x8x128xf32>{%dim_6211}
%5341 = torch_c.from_builtin_tensor %5340 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %5341, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_6212 = torch.constant.int 5
%5342 = torch.prims.convert_element_type %5341, %int5_6212 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5342, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_6213 = torch.constant.int 64
%5343 = torch.aten.mul.Scalar %arg2, %int64_6213 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5343, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int50 = torch.constant.int 50
%int1_6214 = torch.constant.int 1
%5344 = torch.aten.add.Scalar %5343, %int50, %int1_6214 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5344, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6215 = torch.constant.int 4
%int32_6216 = torch.constant.int 32
%int8_6217 = torch.constant.int 8
%int128_6218 = torch.constant.int 128
%5345 = torch.prim.ListConstruct %int4_6215, %425, %int32_6216, %int8_6217, %int128_6218 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5346 = torch.aten.view %5342, %5345 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5346, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6219 = torch.constant.int 4
%5347 = torch.aten.mul.int %int4_6219, %425 : !torch.int, !torch.int -> !torch.int
%int32_6220 = torch.constant.int 32
%int8_6221 = torch.constant.int 8
%int128_6222 = torch.constant.int 128
%5348 = torch.prim.ListConstruct %5347, %int32_6220, %int8_6221, %int128_6222 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5349 = torch.aten.view %5346, %5348 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5349, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6223 = torch.constant.int 4
%5350 = torch.aten.mul.int %int4_6223, %425 : !torch.int, !torch.int -> !torch.int
%5351 = torch.prim.ListConstruct %5350 : (!torch.int) -> !torch.list<int>
%5352 = torch.aten.view %5344, %5351 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5352, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_6224 = torch.constant.int 32
%int2_6225 = torch.constant.int 2
%int32_6226 = torch.constant.int 32
%int8_6227 = torch.constant.int 8
%int128_6228 = torch.constant.int 128
%5353 = torch.prim.ListConstruct %416, %int32_6224, %int2_6225, %int32_6226, %int8_6227, %int128_6228 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5354 = torch.aten.view %5186, %5353 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5354, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6229 = torch.constant.int 32
%5355 = torch.aten.mul.int %416, %int32_6229 : !torch.int, !torch.int -> !torch.int
%int2_6230 = torch.constant.int 2
%5356 = torch.aten.mul.int %5355, %int2_6230 : !torch.int, !torch.int -> !torch.int
%int32_6231 = torch.constant.int 32
%int8_6232 = torch.constant.int 8
%int128_6233 = torch.constant.int 128
%5357 = torch.prim.ListConstruct %5356, %int32_6231, %int8_6232, %int128_6233 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5358 = torch.aten.view %5354, %5357 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5358, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%5359 = torch.prim.ListConstruct %5352 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6234 = torch.constant.bool false
%5360 = torch.aten.index_put %5358, %5359, %5349, %false_6234 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5360, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6235 = torch.constant.int 32
%int2_6236 = torch.constant.int 2
%int32_6237 = torch.constant.int 32
%int8_6238 = torch.constant.int 8
%int128_6239 = torch.constant.int 128
%5361 = torch.prim.ListConstruct %416, %int32_6235, %int2_6236, %int32_6237, %int8_6238, %int128_6239 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5362 = torch.aten.view %5360, %5361 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5362, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6240 = torch.constant.int 2097152
%5363 = torch.prim.ListConstruct %416, %int2097152_6240 : (!torch.int, !torch.int) -> !torch.list<int>
%5364 = torch.aten.view %5362, %5363 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5364, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_6241 = torch.constant.int 32
%int2_6242 = torch.constant.int 2
%int32_6243 = torch.constant.int 32
%int8_6244 = torch.constant.int 8
%int128_6245 = torch.constant.int 128
%5365 = torch.prim.ListConstruct %416, %int32_6241, %int2_6242, %int32_6243, %int8_6244, %int128_6245 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5366 = torch.aten.view %5364, %5365 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5366, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6246 = torch.constant.int 32
%int8_6247 = torch.constant.int 8
%int128_6248 = torch.constant.int 128
%5367 = torch.prim.ListConstruct %5356, %int32_6246, %int8_6247, %int128_6248 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5368 = torch.aten.view %5366, %5367 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5368, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6249 = torch.constant.int 4
%int32_6250 = torch.constant.int 32
%int8_6251 = torch.constant.int 8
%int128_6252 = torch.constant.int 128
%5369 = torch.prim.ListConstruct %int4_6249, %425, %int32_6250, %int8_6251, %int128_6252 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5370 = torch.aten.view %5284, %5369 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5370, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6253 = torch.constant.int 4
%5371 = torch.aten.mul.int %int4_6253, %425 : !torch.int, !torch.int -> !torch.int
%int32_6254 = torch.constant.int 32
%int8_6255 = torch.constant.int 8
%int128_6256 = torch.constant.int 128
%5372 = torch.prim.ListConstruct %5371, %int32_6254, %int8_6255, %int128_6256 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5373 = torch.aten.view %5370, %5372 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5373, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_6257 = torch.constant.int 1
%int1_6258 = torch.constant.int 1
%5374 = torch.aten.add.Scalar %5344, %int1_6257, %int1_6258 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5374, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6259 = torch.constant.int 4
%5375 = torch.aten.mul.int %int4_6259, %425 : !torch.int, !torch.int -> !torch.int
%5376 = torch.prim.ListConstruct %5375 : (!torch.int) -> !torch.list<int>
%5377 = torch.aten.view %5374, %5376 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5377, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%5378 = torch.prim.ListConstruct %5377 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6260 = torch.constant.bool false
%5379 = torch.aten.index_put %5368, %5378, %5373, %false_6260 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5379, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6261 = torch.constant.int 32
%int2_6262 = torch.constant.int 2
%int32_6263 = torch.constant.int 32
%int8_6264 = torch.constant.int 8
%int128_6265 = torch.constant.int 128
%5380 = torch.prim.ListConstruct %416, %int32_6261, %int2_6262, %int32_6263, %int8_6264, %int128_6265 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5381 = torch.aten.view %5379, %5380 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5381, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6266 = torch.constant.int 2097152
%5382 = torch.prim.ListConstruct %416, %int2097152_6266 : (!torch.int, !torch.int) -> !torch.list<int>
%5383 = torch.aten.view %5381, %5382 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5383, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_6267 = torch.constant.int -2
%5384 = torch.aten.unsqueeze %5342, %int-2_6267 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5384, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_6268 = torch.constant.int 4
%int8_6269 = torch.constant.int 8
%int4_6270 = torch.constant.int 4
%int128_6271 = torch.constant.int 128
%5385 = torch.prim.ListConstruct %int4_6268, %5328, %int8_6269, %int4_6270, %int128_6271 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6272 = torch.constant.bool false
%5386 = torch.aten.expand %5384, %5385, %false_6272 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5386, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6273 = torch.constant.int 0
%5387 = torch.aten.clone %5386, %int0_6273 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5387, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6274 = torch.constant.int 4
%int32_6275 = torch.constant.int 32
%int128_6276 = torch.constant.int 128
%5388 = torch.prim.ListConstruct %int4_6274, %5328, %int32_6275, %int128_6276 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5389 = torch.aten._unsafe_view %5387, %5388 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5389, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_6277 = torch.constant.int -2
%5390 = torch.aten.unsqueeze %5284, %int-2_6277 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5390, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_6278 = torch.constant.int 1
%5391 = torch.aten.size.int %5278, %int1_6278 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_6279 = torch.constant.int 4
%int8_6280 = torch.constant.int 8
%int4_6281 = torch.constant.int 4
%int128_6282 = torch.constant.int 128
%5392 = torch.prim.ListConstruct %int4_6279, %5391, %int8_6280, %int4_6281, %int128_6282 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6283 = torch.constant.bool false
%5393 = torch.aten.expand %5390, %5392, %false_6283 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5393, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6284 = torch.constant.int 0
%5394 = torch.aten.clone %5393, %int0_6284 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5394, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6285 = torch.constant.int 4
%int32_6286 = torch.constant.int 32
%int128_6287 = torch.constant.int 128
%5395 = torch.prim.ListConstruct %int4_6285, %5391, %int32_6286, %int128_6287 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5396 = torch.aten._unsafe_view %5394, %5395 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5396, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_6288 = torch.constant.int 1
%int2_6289 = torch.constant.int 2
%5397 = torch.aten.transpose.int %5313, %int1_6288, %int2_6289 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5397, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6290 = torch.constant.int 1
%int2_6291 = torch.constant.int 2
%5398 = torch.aten.transpose.int %5389, %int1_6290, %int2_6291 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5398, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6292 = torch.constant.int 1
%int2_6293 = torch.constant.int 2
%5399 = torch.aten.transpose.int %5396, %int1_6292, %int2_6293 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5399, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_6294 = torch.constant.float 0.000000e+00
%false_6295 = torch.constant.bool false
%none_6296 = torch.constant.none
%5400:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5397, %5398, %5399, %float0.000000e00_6294, %false_6295, %320, %none_6296) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5400#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6297 = torch.constant.int 1
%int2_6298 = torch.constant.int 2
%5401 = torch.aten.transpose.int %5400#0, %int1_6297, %int2_6298 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5401, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6299 = torch.constant.int 4
%int4096_6300 = torch.constant.int 4096
%5402 = torch.prim.ListConstruct %int4_6299, %5299, %int4096_6300 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5403 = torch.aten.view %5401, %5402 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5403, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6301 = torch.constant.int -2
%int-1_6302 = torch.constant.int -1
%5404 = torch.aten.transpose.int %230, %int-2_6301, %int-1_6302 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6303 = torch.constant.int 4
%5405 = torch.aten.mul.int %int4_6303, %5299 : !torch.int, !torch.int -> !torch.int
%int4096_6304 = torch.constant.int 4096
%5406 = torch.prim.ListConstruct %5405, %int4096_6304 : (!torch.int, !torch.int) -> !torch.list<int>
%5407 = torch.aten.view %5403, %5406 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5407, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5408 = torch.aten.mm %5407, %5404 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5408, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6305 = torch.constant.int 4
%int4096_6306 = torch.constant.int 4096
%5409 = torch.prim.ListConstruct %int4_6305, %5299, %int4096_6306 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5410 = torch.aten.view %5408, %5409 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5410, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6307 = torch.constant.int 1
%5411 = torch.aten.add.Tensor %5248, %5410, %int1_6307 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5411, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6308 = torch.constant.int 6
%5412 = torch.prims.convert_element_type %5411, %int6_6308 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5412, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6309 = torch.constant.int 2
%5413 = torch.aten.pow.Tensor_Scalar %5412, %int2_6309 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5413, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6310 = torch.constant.int -1
%5414 = torch.prim.ListConstruct %int-1_6310 : (!torch.int) -> !torch.list<int>
%true_6311 = torch.constant.bool true
%none_6312 = torch.constant.none
%5415 = torch.aten.mean.dim %5413, %5414, %true_6311, %none_6312 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5415, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6313 = torch.constant.float 9.9999997473787516E-6
%int1_6314 = torch.constant.int 1
%5416 = torch.aten.add.Scalar %5415, %float9.999990e-06_6313, %int1_6314 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5416, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5417 = torch.aten.rsqrt %5416 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5417, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5418 = torch.aten.mul.Tensor %5412, %5417 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5418, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5419 = torch.aten.mul.Tensor %231, %5418 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5419, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6315 = torch.constant.int 5
%5420 = torch.prims.convert_element_type %5419, %int5_6315 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5420, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6316 = torch.constant.int -2
%int-1_6317 = torch.constant.int -1
%5421 = torch.aten.transpose.int %232, %int-2_6316, %int-1_6317 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6318 = torch.constant.int 4
%5422 = torch.aten.mul.int %int4_6318, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6319 = torch.constant.int 4096
%5423 = torch.prim.ListConstruct %5422, %int4096_6319 : (!torch.int, !torch.int) -> !torch.list<int>
%5424 = torch.aten.view %5420, %5423 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5424, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5425 = torch.aten.mm %5424, %5421 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5425, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6320 = torch.constant.int 4
%int14336_6321 = torch.constant.int 14336
%5426 = torch.prim.ListConstruct %int4_6320, %294, %int14336_6321 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5427 = torch.aten.view %5425, %5426 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5427, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5428 = torch.aten.silu %5427 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5428, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6322 = torch.constant.int -2
%int-1_6323 = torch.constant.int -1
%5429 = torch.aten.transpose.int %233, %int-2_6322, %int-1_6323 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6324 = torch.constant.int 4
%5430 = torch.aten.mul.int %int4_6324, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6325 = torch.constant.int 4096
%5431 = torch.prim.ListConstruct %5430, %int4096_6325 : (!torch.int, !torch.int) -> !torch.list<int>
%5432 = torch.aten.view %5420, %5431 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5432, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5433 = torch.aten.mm %5432, %5429 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5433, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6326 = torch.constant.int 4
%int14336_6327 = torch.constant.int 14336
%5434 = torch.prim.ListConstruct %int4_6326, %294, %int14336_6327 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5435 = torch.aten.view %5433, %5434 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5435, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5436 = torch.aten.mul.Tensor %5428, %5435 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5436, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6328 = torch.constant.int -2
%int-1_6329 = torch.constant.int -1
%5437 = torch.aten.transpose.int %234, %int-2_6328, %int-1_6329 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_6330 = torch.constant.int 1
%5438 = torch.aten.size.int %5427, %int1_6330 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_6331 = torch.constant.int 4
%5439 = torch.aten.mul.int %int4_6331, %5438 : !torch.int, !torch.int -> !torch.int
%int14336_6332 = torch.constant.int 14336
%5440 = torch.prim.ListConstruct %5439, %int14336_6332 : (!torch.int, !torch.int) -> !torch.list<int>
%5441 = torch.aten.view %5436, %5440 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5441, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%5442 = torch.aten.mm %5441, %5437 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5442, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6333 = torch.constant.int 4
%int4096_6334 = torch.constant.int 4096
%5443 = torch.prim.ListConstruct %int4_6333, %5438, %int4096_6334 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5444 = torch.aten.view %5442, %5443 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5444, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6335 = torch.constant.int 1
%5445 = torch.aten.add.Tensor %5411, %5444, %int1_6335 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5445, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6336 = torch.constant.int 6
%5446 = torch.prims.convert_element_type %5445, %int6_6336 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5446, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6337 = torch.constant.int 2
%5447 = torch.aten.pow.Tensor_Scalar %5446, %int2_6337 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5447, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6338 = torch.constant.int -1
%5448 = torch.prim.ListConstruct %int-1_6338 : (!torch.int) -> !torch.list<int>
%true_6339 = torch.constant.bool true
%none_6340 = torch.constant.none
%5449 = torch.aten.mean.dim %5447, %5448, %true_6339, %none_6340 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5449, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6341 = torch.constant.float 9.9999997473787516E-6
%int1_6342 = torch.constant.int 1
%5450 = torch.aten.add.Scalar %5449, %float9.999990e-06_6341, %int1_6342 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5450, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5451 = torch.aten.rsqrt %5450 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5451, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5452 = torch.aten.mul.Tensor %5446, %5451 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5452, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5453 = torch.aten.mul.Tensor %235, %5452 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5453, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6343 = torch.constant.int 5
%5454 = torch.prims.convert_element_type %5453, %int5_6343 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5454, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6344 = torch.constant.int -2
%int-1_6345 = torch.constant.int -1
%5455 = torch.aten.transpose.int %236, %int-2_6344, %int-1_6345 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6346 = torch.constant.int 4
%5456 = torch.aten.mul.int %int4_6346, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6347 = torch.constant.int 4096
%5457 = torch.prim.ListConstruct %5456, %int4096_6347 : (!torch.int, !torch.int) -> !torch.list<int>
%5458 = torch.aten.view %5454, %5457 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5458, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5459 = torch.aten.mm %5458, %5455 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5459, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6348 = torch.constant.int 4
%int4096_6349 = torch.constant.int 4096
%5460 = torch.prim.ListConstruct %int4_6348, %294, %int4096_6349 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5461 = torch.aten.view %5459, %5460 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5461, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6350 = torch.constant.int -2
%int-1_6351 = torch.constant.int -1
%5462 = torch.aten.transpose.int %237, %int-2_6350, %int-1_6351 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6352 = torch.constant.int 4
%5463 = torch.aten.mul.int %int4_6352, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6353 = torch.constant.int 4096
%5464 = torch.prim.ListConstruct %5463, %int4096_6353 : (!torch.int, !torch.int) -> !torch.list<int>
%5465 = torch.aten.view %5454, %5464 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5465, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5466 = torch.aten.mm %5465, %5462 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5466, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6354 = torch.constant.int 4
%int1024_6355 = torch.constant.int 1024
%5467 = torch.prim.ListConstruct %int4_6354, %294, %int1024_6355 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5468 = torch.aten.view %5466, %5467 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5468, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_6356 = torch.constant.int -2
%int-1_6357 = torch.constant.int -1
%5469 = torch.aten.transpose.int %238, %int-2_6356, %int-1_6357 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6358 = torch.constant.int 4
%5470 = torch.aten.mul.int %int4_6358, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6359 = torch.constant.int 4096
%5471 = torch.prim.ListConstruct %5470, %int4096_6359 : (!torch.int, !torch.int) -> !torch.list<int>
%5472 = torch.aten.view %5454, %5471 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5472, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5473 = torch.aten.mm %5472, %5469 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5473, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6360 = torch.constant.int 4
%int1024_6361 = torch.constant.int 1024
%5474 = torch.prim.ListConstruct %int4_6360, %294, %int1024_6361 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5475 = torch.aten.view %5473, %5474 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5475, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_6362 = torch.constant.int 4
%int32_6363 = torch.constant.int 32
%int128_6364 = torch.constant.int 128
%5476 = torch.prim.ListConstruct %int4_6362, %294, %int32_6363, %int128_6364 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5477 = torch.aten.view %5461, %5476 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5477, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6365 = torch.constant.int 4
%int8_6366 = torch.constant.int 8
%int128_6367 = torch.constant.int 128
%5478 = torch.prim.ListConstruct %int4_6365, %294, %int8_6366, %int128_6367 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5479 = torch.aten.view %5468, %5478 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5479, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_6368 = torch.constant.int 4
%int8_6369 = torch.constant.int 8
%int128_6370 = torch.constant.int 128
%5480 = torch.prim.ListConstruct %int4_6368, %294, %int8_6369, %int128_6370 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5481 = torch.aten.view %5475, %5480 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5481, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_6371 = torch.constant.int 131072
%none_6372 = torch.constant.none
%none_6373 = torch.constant.none
%cpu_6374 = torch.constant.device "cpu"
%false_6375 = torch.constant.bool false
%5482 = torch.aten.arange %int131072_6371, %none_6372, %none_6373, %cpu_6374, %false_6375 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6376 = torch.constant.int 0
%int128_6377 = torch.constant.int 128
%int2_6378 = torch.constant.int 2
%none_6379 = torch.constant.none
%none_6380 = torch.constant.none
%cpu_6381 = torch.constant.device "cpu"
%false_6382 = torch.constant.bool false
%5483 = torch.aten.arange.start_step %int0_6376, %int128_6377, %int2_6378, %none_6379, %none_6380, %cpu_6381, %false_6382 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6383 = torch.constant.int 0
%int0_6384 = torch.constant.int 0
%int64_6385 = torch.constant.int 64
%int1_6386 = torch.constant.int 1
%5484 = torch.aten.slice.Tensor %5483, %int0_6383, %int0_6384, %int64_6385, %int1_6386 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6387 = torch.constant.int 6
%5485 = torch.prims.convert_element_type %5484, %int6_6387 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6388 = torch.constant.int 128
%5486 = torch.aten.div.Scalar %5485, %int128_6388 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6389 = torch.constant.float 5.000000e+05
%5487 = torch.aten.pow.Scalar %float5.000000e05_6389, %5486 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5488 = torch.aten.reciprocal %5487 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6390 = torch.constant.float 1.000000e+00
%5489 = torch.aten.mul.Scalar %5488, %float1.000000e00_6390 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6391 = torch.constant.int 131072
%int1_6392 = torch.constant.int 1
%5490 = torch.prim.ListConstruct %int131072_6391, %int1_6392 : (!torch.int, !torch.int) -> !torch.list<int>
%5491 = torch.aten.view %5482, %5490 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5492 = torch.aten.mul.Tensor %5491, %5489 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5493 = torch.aten.cos %5492 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5494 = torch.aten.sin %5492 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5495 = torch.aten.complex %5493, %5494 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6393 = torch.constant.int 1
%5496 = torch.aten.size.int %5461, %int1_6393 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_6394 = torch.constant.int 0
%5497 = torch.aten.add.int %int0_6394, %5496 : !torch.int, !torch.int -> !torch.int
%int0_6395 = torch.constant.int 0
%int0_6396 = torch.constant.int 0
%int1_6397 = torch.constant.int 1
%5498 = torch.aten.slice.Tensor %5495, %int0_6395, %int0_6396, %5497, %int1_6397 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5498, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6398 = torch.constant.int 1
%int0_6399 = torch.constant.int 0
%int9223372036854775807_6400 = torch.constant.int 9223372036854775807
%int1_6401 = torch.constant.int 1
%5499 = torch.aten.slice.Tensor %5498, %int1_6398, %int0_6399, %int9223372036854775807_6400, %int1_6401 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5499, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6402 = torch.constant.int 0
%5500 = torch.aten.unsqueeze %5499, %int0_6402 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5500, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6403 = torch.constant.int 2
%5501 = torch.aten.unsqueeze %5500, %int2_6403 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5501, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6404 = torch.constant.int 3
%int0_6405 = torch.constant.int 0
%int9223372036854775807_6406 = torch.constant.int 9223372036854775807
%int1_6407 = torch.constant.int 1
%5502 = torch.aten.slice.Tensor %5501, %int3_6404, %int0_6405, %int9223372036854775807_6406, %int1_6407 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5502, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5503 = torch_c.to_builtin_tensor %5477 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_6408 = arith.constant 1 : index
%dim_6409 = tensor.dim %5503, %c1_6408 : tensor<4x?x32x128xf16>
%5504 = flow.tensor.bitcast %5503 : tensor<4x?x32x128xf16>{%dim_6409} -> tensor<4x?x32x64xcomplex<f16>>{%dim_6409}
%5505 = torch_c.from_builtin_tensor %5504 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %5505, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%5506 = torch.aten.mul.Tensor %5505, %5502 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %5506, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%5507 = torch_c.to_builtin_tensor %5506 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_6410 = arith.constant 1 : index
%dim_6411 = tensor.dim %5507, %c1_6410 : tensor<4x?x32x64xcomplex<f32>>
%5508 = flow.tensor.bitcast %5507 : tensor<4x?x32x64xcomplex<f32>>{%dim_6411} -> tensor<4x?x32x128xf32>{%dim_6411}
%5509 = torch_c.from_builtin_tensor %5508 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %5509, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_6412 = torch.constant.int 5
%5510 = torch.prims.convert_element_type %5509, %int5_6412 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5510, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_6413 = torch.constant.int 131072
%none_6414 = torch.constant.none
%none_6415 = torch.constant.none
%cpu_6416 = torch.constant.device "cpu"
%false_6417 = torch.constant.bool false
%5511 = torch.aten.arange %int131072_6413, %none_6414, %none_6415, %cpu_6416, %false_6417 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6418 = torch.constant.int 0
%int128_6419 = torch.constant.int 128
%int2_6420 = torch.constant.int 2
%none_6421 = torch.constant.none
%none_6422 = torch.constant.none
%cpu_6423 = torch.constant.device "cpu"
%false_6424 = torch.constant.bool false
%5512 = torch.aten.arange.start_step %int0_6418, %int128_6419, %int2_6420, %none_6421, %none_6422, %cpu_6423, %false_6424 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6425 = torch.constant.int 0
%int0_6426 = torch.constant.int 0
%int64_6427 = torch.constant.int 64
%int1_6428 = torch.constant.int 1
%5513 = torch.aten.slice.Tensor %5512, %int0_6425, %int0_6426, %int64_6427, %int1_6428 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6429 = torch.constant.int 6
%5514 = torch.prims.convert_element_type %5513, %int6_6429 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6430 = torch.constant.int 128
%5515 = torch.aten.div.Scalar %5514, %int128_6430 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6431 = torch.constant.float 5.000000e+05
%5516 = torch.aten.pow.Scalar %float5.000000e05_6431, %5515 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5517 = torch.aten.reciprocal %5516 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6432 = torch.constant.float 1.000000e+00
%5518 = torch.aten.mul.Scalar %5517, %float1.000000e00_6432 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6433 = torch.constant.int 131072
%int1_6434 = torch.constant.int 1
%5519 = torch.prim.ListConstruct %int131072_6433, %int1_6434 : (!torch.int, !torch.int) -> !torch.list<int>
%5520 = torch.aten.view %5511, %5519 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5521 = torch.aten.mul.Tensor %5520, %5518 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5522 = torch.aten.cos %5521 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5523 = torch.aten.sin %5521 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5524 = torch.aten.complex %5522, %5523 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6435 = torch.constant.int 1
%5525 = torch.aten.size.int %5468, %int1_6435 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_6436 = torch.constant.int 0
%5526 = torch.aten.add.int %int0_6436, %5525 : !torch.int, !torch.int -> !torch.int
%int0_6437 = torch.constant.int 0
%int0_6438 = torch.constant.int 0
%int1_6439 = torch.constant.int 1
%5527 = torch.aten.slice.Tensor %5524, %int0_6437, %int0_6438, %5526, %int1_6439 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5527, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6440 = torch.constant.int 1
%int0_6441 = torch.constant.int 0
%int9223372036854775807_6442 = torch.constant.int 9223372036854775807
%int1_6443 = torch.constant.int 1
%5528 = torch.aten.slice.Tensor %5527, %int1_6440, %int0_6441, %int9223372036854775807_6442, %int1_6443 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5528, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6444 = torch.constant.int 0
%5529 = torch.aten.unsqueeze %5528, %int0_6444 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5529, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6445 = torch.constant.int 2
%5530 = torch.aten.unsqueeze %5529, %int2_6445 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5530, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6446 = torch.constant.int 3
%int0_6447 = torch.constant.int 0
%int9223372036854775807_6448 = torch.constant.int 9223372036854775807
%int1_6449 = torch.constant.int 1
%5531 = torch.aten.slice.Tensor %5530, %int3_6446, %int0_6447, %int9223372036854775807_6448, %int1_6449 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5531, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5532 = torch_c.to_builtin_tensor %5479 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_6450 = arith.constant 1 : index
%dim_6451 = tensor.dim %5532, %c1_6450 : tensor<4x?x8x128xf16>
%5533 = flow.tensor.bitcast %5532 : tensor<4x?x8x128xf16>{%dim_6451} -> tensor<4x?x8x64xcomplex<f16>>{%dim_6451}
%5534 = torch_c.from_builtin_tensor %5533 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %5534, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%5535 = torch.aten.mul.Tensor %5534, %5531 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %5535, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%5536 = torch_c.to_builtin_tensor %5535 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_6452 = arith.constant 1 : index
%dim_6453 = tensor.dim %5536, %c1_6452 : tensor<4x?x8x64xcomplex<f32>>
%5537 = flow.tensor.bitcast %5536 : tensor<4x?x8x64xcomplex<f32>>{%dim_6453} -> tensor<4x?x8x128xf32>{%dim_6453}
%5538 = torch_c.from_builtin_tensor %5537 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %5538, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_6454 = torch.constant.int 5
%5539 = torch.prims.convert_element_type %5538, %int5_6454 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5539, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_6455 = torch.constant.int 64
%5540 = torch.aten.mul.Scalar %arg2, %int64_6455 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5540, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int52 = torch.constant.int 52
%int1_6456 = torch.constant.int 1
%5541 = torch.aten.add.Scalar %5540, %int52, %int1_6456 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5541, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6457 = torch.constant.int 4
%int32_6458 = torch.constant.int 32
%int8_6459 = torch.constant.int 8
%int128_6460 = torch.constant.int 128
%5542 = torch.prim.ListConstruct %int4_6457, %425, %int32_6458, %int8_6459, %int128_6460 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5543 = torch.aten.view %5539, %5542 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5543, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6461 = torch.constant.int 4
%5544 = torch.aten.mul.int %int4_6461, %425 : !torch.int, !torch.int -> !torch.int
%int32_6462 = torch.constant.int 32
%int8_6463 = torch.constant.int 8
%int128_6464 = torch.constant.int 128
%5545 = torch.prim.ListConstruct %5544, %int32_6462, %int8_6463, %int128_6464 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5546 = torch.aten.view %5543, %5545 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5546, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6465 = torch.constant.int 4
%5547 = torch.aten.mul.int %int4_6465, %425 : !torch.int, !torch.int -> !torch.int
%5548 = torch.prim.ListConstruct %5547 : (!torch.int) -> !torch.list<int>
%5549 = torch.aten.view %5541, %5548 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5549, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_6466 = torch.constant.int 32
%int2_6467 = torch.constant.int 2
%int32_6468 = torch.constant.int 32
%int8_6469 = torch.constant.int 8
%int128_6470 = torch.constant.int 128
%5550 = torch.prim.ListConstruct %416, %int32_6466, %int2_6467, %int32_6468, %int8_6469, %int128_6470 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5551 = torch.aten.view %5383, %5550 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5551, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6471 = torch.constant.int 32
%5552 = torch.aten.mul.int %416, %int32_6471 : !torch.int, !torch.int -> !torch.int
%int2_6472 = torch.constant.int 2
%5553 = torch.aten.mul.int %5552, %int2_6472 : !torch.int, !torch.int -> !torch.int
%int32_6473 = torch.constant.int 32
%int8_6474 = torch.constant.int 8
%int128_6475 = torch.constant.int 128
%5554 = torch.prim.ListConstruct %5553, %int32_6473, %int8_6474, %int128_6475 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5555 = torch.aten.view %5551, %5554 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5555, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%5556 = torch.prim.ListConstruct %5549 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6476 = torch.constant.bool false
%5557 = torch.aten.index_put %5555, %5556, %5546, %false_6476 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5557, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6477 = torch.constant.int 32
%int2_6478 = torch.constant.int 2
%int32_6479 = torch.constant.int 32
%int8_6480 = torch.constant.int 8
%int128_6481 = torch.constant.int 128
%5558 = torch.prim.ListConstruct %416, %int32_6477, %int2_6478, %int32_6479, %int8_6480, %int128_6481 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5559 = torch.aten.view %5557, %5558 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5559, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6482 = torch.constant.int 2097152
%5560 = torch.prim.ListConstruct %416, %int2097152_6482 : (!torch.int, !torch.int) -> !torch.list<int>
%5561 = torch.aten.view %5559, %5560 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5561, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_6483 = torch.constant.int 32
%int2_6484 = torch.constant.int 2
%int32_6485 = torch.constant.int 32
%int8_6486 = torch.constant.int 8
%int128_6487 = torch.constant.int 128
%5562 = torch.prim.ListConstruct %416, %int32_6483, %int2_6484, %int32_6485, %int8_6486, %int128_6487 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5563 = torch.aten.view %5561, %5562 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5563, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6488 = torch.constant.int 32
%int8_6489 = torch.constant.int 8
%int128_6490 = torch.constant.int 128
%5564 = torch.prim.ListConstruct %5553, %int32_6488, %int8_6489, %int128_6490 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5565 = torch.aten.view %5563, %5564 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5565, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6491 = torch.constant.int 4
%int32_6492 = torch.constant.int 32
%int8_6493 = torch.constant.int 8
%int128_6494 = torch.constant.int 128
%5566 = torch.prim.ListConstruct %int4_6491, %425, %int32_6492, %int8_6493, %int128_6494 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5567 = torch.aten.view %5481, %5566 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5567, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6495 = torch.constant.int 4
%5568 = torch.aten.mul.int %int4_6495, %425 : !torch.int, !torch.int -> !torch.int
%int32_6496 = torch.constant.int 32
%int8_6497 = torch.constant.int 8
%int128_6498 = torch.constant.int 128
%5569 = torch.prim.ListConstruct %5568, %int32_6496, %int8_6497, %int128_6498 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5570 = torch.aten.view %5567, %5569 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5570, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_6499 = torch.constant.int 1
%int1_6500 = torch.constant.int 1
%5571 = torch.aten.add.Scalar %5541, %int1_6499, %int1_6500 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5571, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6501 = torch.constant.int 4
%5572 = torch.aten.mul.int %int4_6501, %425 : !torch.int, !torch.int -> !torch.int
%5573 = torch.prim.ListConstruct %5572 : (!torch.int) -> !torch.list<int>
%5574 = torch.aten.view %5571, %5573 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5574, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%5575 = torch.prim.ListConstruct %5574 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6502 = torch.constant.bool false
%5576 = torch.aten.index_put %5565, %5575, %5570, %false_6502 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5576, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6503 = torch.constant.int 32
%int2_6504 = torch.constant.int 2
%int32_6505 = torch.constant.int 32
%int8_6506 = torch.constant.int 8
%int128_6507 = torch.constant.int 128
%5577 = torch.prim.ListConstruct %416, %int32_6503, %int2_6504, %int32_6505, %int8_6506, %int128_6507 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5578 = torch.aten.view %5576, %5577 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5578, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6508 = torch.constant.int 2097152
%5579 = torch.prim.ListConstruct %416, %int2097152_6508 : (!torch.int, !torch.int) -> !torch.list<int>
%5580 = torch.aten.view %5578, %5579 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5580, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_6509 = torch.constant.int -2
%5581 = torch.aten.unsqueeze %5539, %int-2_6509 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5581, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_6510 = torch.constant.int 4
%int8_6511 = torch.constant.int 8
%int4_6512 = torch.constant.int 4
%int128_6513 = torch.constant.int 128
%5582 = torch.prim.ListConstruct %int4_6510, %5525, %int8_6511, %int4_6512, %int128_6513 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6514 = torch.constant.bool false
%5583 = torch.aten.expand %5581, %5582, %false_6514 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5583, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6515 = torch.constant.int 0
%5584 = torch.aten.clone %5583, %int0_6515 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5584, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6516 = torch.constant.int 4
%int32_6517 = torch.constant.int 32
%int128_6518 = torch.constant.int 128
%5585 = torch.prim.ListConstruct %int4_6516, %5525, %int32_6517, %int128_6518 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5586 = torch.aten._unsafe_view %5584, %5585 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5586, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_6519 = torch.constant.int -2
%5587 = torch.aten.unsqueeze %5481, %int-2_6519 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5587, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_6520 = torch.constant.int 1
%5588 = torch.aten.size.int %5475, %int1_6520 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_6521 = torch.constant.int 4
%int8_6522 = torch.constant.int 8
%int4_6523 = torch.constant.int 4
%int128_6524 = torch.constant.int 128
%5589 = torch.prim.ListConstruct %int4_6521, %5588, %int8_6522, %int4_6523, %int128_6524 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6525 = torch.constant.bool false
%5590 = torch.aten.expand %5587, %5589, %false_6525 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5590, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6526 = torch.constant.int 0
%5591 = torch.aten.clone %5590, %int0_6526 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5591, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6527 = torch.constant.int 4
%int32_6528 = torch.constant.int 32
%int128_6529 = torch.constant.int 128
%5592 = torch.prim.ListConstruct %int4_6527, %5588, %int32_6528, %int128_6529 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5593 = torch.aten._unsafe_view %5591, %5592 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5593, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_6530 = torch.constant.int 1
%int2_6531 = torch.constant.int 2
%5594 = torch.aten.transpose.int %5510, %int1_6530, %int2_6531 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5594, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6532 = torch.constant.int 1
%int2_6533 = torch.constant.int 2
%5595 = torch.aten.transpose.int %5586, %int1_6532, %int2_6533 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5595, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6534 = torch.constant.int 1
%int2_6535 = torch.constant.int 2
%5596 = torch.aten.transpose.int %5593, %int1_6534, %int2_6535 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5596, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_6536 = torch.constant.float 0.000000e+00
%false_6537 = torch.constant.bool false
%none_6538 = torch.constant.none
%5597:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5594, %5595, %5596, %float0.000000e00_6536, %false_6537, %320, %none_6538) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5597#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6539 = torch.constant.int 1
%int2_6540 = torch.constant.int 2
%5598 = torch.aten.transpose.int %5597#0, %int1_6539, %int2_6540 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5598, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6541 = torch.constant.int 4
%int4096_6542 = torch.constant.int 4096
%5599 = torch.prim.ListConstruct %int4_6541, %5496, %int4096_6542 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5600 = torch.aten.view %5598, %5599 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5600, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6543 = torch.constant.int -2
%int-1_6544 = torch.constant.int -1
%5601 = torch.aten.transpose.int %239, %int-2_6543, %int-1_6544 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6545 = torch.constant.int 4
%5602 = torch.aten.mul.int %int4_6545, %5496 : !torch.int, !torch.int -> !torch.int
%int4096_6546 = torch.constant.int 4096
%5603 = torch.prim.ListConstruct %5602, %int4096_6546 : (!torch.int, !torch.int) -> !torch.list<int>
%5604 = torch.aten.view %5600, %5603 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5604, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5605 = torch.aten.mm %5604, %5601 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5605, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6547 = torch.constant.int 4
%int4096_6548 = torch.constant.int 4096
%5606 = torch.prim.ListConstruct %int4_6547, %5496, %int4096_6548 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5607 = torch.aten.view %5605, %5606 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5607, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6549 = torch.constant.int 1
%5608 = torch.aten.add.Tensor %5445, %5607, %int1_6549 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5608, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6550 = torch.constant.int 6
%5609 = torch.prims.convert_element_type %5608, %int6_6550 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5609, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6551 = torch.constant.int 2
%5610 = torch.aten.pow.Tensor_Scalar %5609, %int2_6551 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5610, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6552 = torch.constant.int -1
%5611 = torch.prim.ListConstruct %int-1_6552 : (!torch.int) -> !torch.list<int>
%true_6553 = torch.constant.bool true
%none_6554 = torch.constant.none
%5612 = torch.aten.mean.dim %5610, %5611, %true_6553, %none_6554 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5612, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6555 = torch.constant.float 9.9999997473787516E-6
%int1_6556 = torch.constant.int 1
%5613 = torch.aten.add.Scalar %5612, %float9.999990e-06_6555, %int1_6556 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5613, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5614 = torch.aten.rsqrt %5613 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5614, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5615 = torch.aten.mul.Tensor %5609, %5614 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5615, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5616 = torch.aten.mul.Tensor %240, %5615 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5616, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6557 = torch.constant.int 5
%5617 = torch.prims.convert_element_type %5616, %int5_6557 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5617, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6558 = torch.constant.int -2
%int-1_6559 = torch.constant.int -1
%5618 = torch.aten.transpose.int %241, %int-2_6558, %int-1_6559 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6560 = torch.constant.int 4
%5619 = torch.aten.mul.int %int4_6560, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6561 = torch.constant.int 4096
%5620 = torch.prim.ListConstruct %5619, %int4096_6561 : (!torch.int, !torch.int) -> !torch.list<int>
%5621 = torch.aten.view %5617, %5620 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5621, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5622 = torch.aten.mm %5621, %5618 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5622, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6562 = torch.constant.int 4
%int14336_6563 = torch.constant.int 14336
%5623 = torch.prim.ListConstruct %int4_6562, %294, %int14336_6563 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5624 = torch.aten.view %5622, %5623 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5624, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5625 = torch.aten.silu %5624 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5625, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6564 = torch.constant.int -2
%int-1_6565 = torch.constant.int -1
%5626 = torch.aten.transpose.int %242, %int-2_6564, %int-1_6565 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6566 = torch.constant.int 4
%5627 = torch.aten.mul.int %int4_6566, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6567 = torch.constant.int 4096
%5628 = torch.prim.ListConstruct %5627, %int4096_6567 : (!torch.int, !torch.int) -> !torch.list<int>
%5629 = torch.aten.view %5617, %5628 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5629, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5630 = torch.aten.mm %5629, %5626 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5630, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6568 = torch.constant.int 4
%int14336_6569 = torch.constant.int 14336
%5631 = torch.prim.ListConstruct %int4_6568, %294, %int14336_6569 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5632 = torch.aten.view %5630, %5631 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5632, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5633 = torch.aten.mul.Tensor %5625, %5632 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5633, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6570 = torch.constant.int -2
%int-1_6571 = torch.constant.int -1
%5634 = torch.aten.transpose.int %243, %int-2_6570, %int-1_6571 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_6572 = torch.constant.int 1
%5635 = torch.aten.size.int %5624, %int1_6572 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_6573 = torch.constant.int 4
%5636 = torch.aten.mul.int %int4_6573, %5635 : !torch.int, !torch.int -> !torch.int
%int14336_6574 = torch.constant.int 14336
%5637 = torch.prim.ListConstruct %5636, %int14336_6574 : (!torch.int, !torch.int) -> !torch.list<int>
%5638 = torch.aten.view %5633, %5637 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5638, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%5639 = torch.aten.mm %5638, %5634 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5639, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6575 = torch.constant.int 4
%int4096_6576 = torch.constant.int 4096
%5640 = torch.prim.ListConstruct %int4_6575, %5635, %int4096_6576 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5641 = torch.aten.view %5639, %5640 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5641, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6577 = torch.constant.int 1
%5642 = torch.aten.add.Tensor %5608, %5641, %int1_6577 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5642, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6578 = torch.constant.int 6
%5643 = torch.prims.convert_element_type %5642, %int6_6578 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5643, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6579 = torch.constant.int 2
%5644 = torch.aten.pow.Tensor_Scalar %5643, %int2_6579 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5644, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6580 = torch.constant.int -1
%5645 = torch.prim.ListConstruct %int-1_6580 : (!torch.int) -> !torch.list<int>
%true_6581 = torch.constant.bool true
%none_6582 = torch.constant.none
%5646 = torch.aten.mean.dim %5644, %5645, %true_6581, %none_6582 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5646, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6583 = torch.constant.float 9.9999997473787516E-6
%int1_6584 = torch.constant.int 1
%5647 = torch.aten.add.Scalar %5646, %float9.999990e-06_6583, %int1_6584 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5647, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5648 = torch.aten.rsqrt %5647 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5648, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5649 = torch.aten.mul.Tensor %5643, %5648 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5649, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5650 = torch.aten.mul.Tensor %244, %5649 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5650, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6585 = torch.constant.int 5
%5651 = torch.prims.convert_element_type %5650, %int5_6585 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5651, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6586 = torch.constant.int -2
%int-1_6587 = torch.constant.int -1
%5652 = torch.aten.transpose.int %245, %int-2_6586, %int-1_6587 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6588 = torch.constant.int 4
%5653 = torch.aten.mul.int %int4_6588, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6589 = torch.constant.int 4096
%5654 = torch.prim.ListConstruct %5653, %int4096_6589 : (!torch.int, !torch.int) -> !torch.list<int>
%5655 = torch.aten.view %5651, %5654 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5655, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5656 = torch.aten.mm %5655, %5652 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5656, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6590 = torch.constant.int 4
%int4096_6591 = torch.constant.int 4096
%5657 = torch.prim.ListConstruct %int4_6590, %294, %int4096_6591 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5658 = torch.aten.view %5656, %5657 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5658, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6592 = torch.constant.int -2
%int-1_6593 = torch.constant.int -1
%5659 = torch.aten.transpose.int %246, %int-2_6592, %int-1_6593 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6594 = torch.constant.int 4
%5660 = torch.aten.mul.int %int4_6594, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6595 = torch.constant.int 4096
%5661 = torch.prim.ListConstruct %5660, %int4096_6595 : (!torch.int, !torch.int) -> !torch.list<int>
%5662 = torch.aten.view %5651, %5661 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5662, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5663 = torch.aten.mm %5662, %5659 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5663, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6596 = torch.constant.int 4
%int1024_6597 = torch.constant.int 1024
%5664 = torch.prim.ListConstruct %int4_6596, %294, %int1024_6597 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5665 = torch.aten.view %5663, %5664 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5665, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_6598 = torch.constant.int -2
%int-1_6599 = torch.constant.int -1
%5666 = torch.aten.transpose.int %247, %int-2_6598, %int-1_6599 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6600 = torch.constant.int 4
%5667 = torch.aten.mul.int %int4_6600, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6601 = torch.constant.int 4096
%5668 = torch.prim.ListConstruct %5667, %int4096_6601 : (!torch.int, !torch.int) -> !torch.list<int>
%5669 = torch.aten.view %5651, %5668 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5669, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5670 = torch.aten.mm %5669, %5666 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5670, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6602 = torch.constant.int 4
%int1024_6603 = torch.constant.int 1024
%5671 = torch.prim.ListConstruct %int4_6602, %294, %int1024_6603 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5672 = torch.aten.view %5670, %5671 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5672, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_6604 = torch.constant.int 4
%int32_6605 = torch.constant.int 32
%int128_6606 = torch.constant.int 128
%5673 = torch.prim.ListConstruct %int4_6604, %294, %int32_6605, %int128_6606 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5674 = torch.aten.view %5658, %5673 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5674, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6607 = torch.constant.int 4
%int8_6608 = torch.constant.int 8
%int128_6609 = torch.constant.int 128
%5675 = torch.prim.ListConstruct %int4_6607, %294, %int8_6608, %int128_6609 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5676 = torch.aten.view %5665, %5675 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5676, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_6610 = torch.constant.int 4
%int8_6611 = torch.constant.int 8
%int128_6612 = torch.constant.int 128
%5677 = torch.prim.ListConstruct %int4_6610, %294, %int8_6611, %int128_6612 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5678 = torch.aten.view %5672, %5677 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5678, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_6613 = torch.constant.int 131072
%none_6614 = torch.constant.none
%none_6615 = torch.constant.none
%cpu_6616 = torch.constant.device "cpu"
%false_6617 = torch.constant.bool false
%5679 = torch.aten.arange %int131072_6613, %none_6614, %none_6615, %cpu_6616, %false_6617 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6618 = torch.constant.int 0
%int128_6619 = torch.constant.int 128
%int2_6620 = torch.constant.int 2
%none_6621 = torch.constant.none
%none_6622 = torch.constant.none
%cpu_6623 = torch.constant.device "cpu"
%false_6624 = torch.constant.bool false
%5680 = torch.aten.arange.start_step %int0_6618, %int128_6619, %int2_6620, %none_6621, %none_6622, %cpu_6623, %false_6624 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6625 = torch.constant.int 0
%int0_6626 = torch.constant.int 0
%int64_6627 = torch.constant.int 64
%int1_6628 = torch.constant.int 1
%5681 = torch.aten.slice.Tensor %5680, %int0_6625, %int0_6626, %int64_6627, %int1_6628 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6629 = torch.constant.int 6
%5682 = torch.prims.convert_element_type %5681, %int6_6629 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6630 = torch.constant.int 128
%5683 = torch.aten.div.Scalar %5682, %int128_6630 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6631 = torch.constant.float 5.000000e+05
%5684 = torch.aten.pow.Scalar %float5.000000e05_6631, %5683 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5685 = torch.aten.reciprocal %5684 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6632 = torch.constant.float 1.000000e+00
%5686 = torch.aten.mul.Scalar %5685, %float1.000000e00_6632 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6633 = torch.constant.int 131072
%int1_6634 = torch.constant.int 1
%5687 = torch.prim.ListConstruct %int131072_6633, %int1_6634 : (!torch.int, !torch.int) -> !torch.list<int>
%5688 = torch.aten.view %5679, %5687 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5689 = torch.aten.mul.Tensor %5688, %5686 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5690 = torch.aten.cos %5689 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5691 = torch.aten.sin %5689 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5692 = torch.aten.complex %5690, %5691 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6635 = torch.constant.int 1
%5693 = torch.aten.size.int %5658, %int1_6635 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_6636 = torch.constant.int 0
%5694 = torch.aten.add.int %int0_6636, %5693 : !torch.int, !torch.int -> !torch.int
%int0_6637 = torch.constant.int 0
%int0_6638 = torch.constant.int 0
%int1_6639 = torch.constant.int 1
%5695 = torch.aten.slice.Tensor %5692, %int0_6637, %int0_6638, %5694, %int1_6639 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5695, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6640 = torch.constant.int 1
%int0_6641 = torch.constant.int 0
%int9223372036854775807_6642 = torch.constant.int 9223372036854775807
%int1_6643 = torch.constant.int 1
%5696 = torch.aten.slice.Tensor %5695, %int1_6640, %int0_6641, %int9223372036854775807_6642, %int1_6643 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5696, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6644 = torch.constant.int 0
%5697 = torch.aten.unsqueeze %5696, %int0_6644 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5697, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6645 = torch.constant.int 2
%5698 = torch.aten.unsqueeze %5697, %int2_6645 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5698, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6646 = torch.constant.int 3
%int0_6647 = torch.constant.int 0
%int9223372036854775807_6648 = torch.constant.int 9223372036854775807
%int1_6649 = torch.constant.int 1
%5699 = torch.aten.slice.Tensor %5698, %int3_6646, %int0_6647, %int9223372036854775807_6648, %int1_6649 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5699, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5700 = torch_c.to_builtin_tensor %5674 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_6650 = arith.constant 1 : index
%dim_6651 = tensor.dim %5700, %c1_6650 : tensor<4x?x32x128xf16>
%5701 = flow.tensor.bitcast %5700 : tensor<4x?x32x128xf16>{%dim_6651} -> tensor<4x?x32x64xcomplex<f16>>{%dim_6651}
%5702 = torch_c.from_builtin_tensor %5701 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %5702, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%5703 = torch.aten.mul.Tensor %5702, %5699 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %5703, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%5704 = torch_c.to_builtin_tensor %5703 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_6652 = arith.constant 1 : index
%dim_6653 = tensor.dim %5704, %c1_6652 : tensor<4x?x32x64xcomplex<f32>>
%5705 = flow.tensor.bitcast %5704 : tensor<4x?x32x64xcomplex<f32>>{%dim_6653} -> tensor<4x?x32x128xf32>{%dim_6653}
%5706 = torch_c.from_builtin_tensor %5705 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %5706, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_6654 = torch.constant.int 5
%5707 = torch.prims.convert_element_type %5706, %int5_6654 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5707, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_6655 = torch.constant.int 131072
%none_6656 = torch.constant.none
%none_6657 = torch.constant.none
%cpu_6658 = torch.constant.device "cpu"
%false_6659 = torch.constant.bool false
%5708 = torch.aten.arange %int131072_6655, %none_6656, %none_6657, %cpu_6658, %false_6659 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6660 = torch.constant.int 0
%int128_6661 = torch.constant.int 128
%int2_6662 = torch.constant.int 2
%none_6663 = torch.constant.none
%none_6664 = torch.constant.none
%cpu_6665 = torch.constant.device "cpu"
%false_6666 = torch.constant.bool false
%5709 = torch.aten.arange.start_step %int0_6660, %int128_6661, %int2_6662, %none_6663, %none_6664, %cpu_6665, %false_6666 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6667 = torch.constant.int 0
%int0_6668 = torch.constant.int 0
%int64_6669 = torch.constant.int 64
%int1_6670 = torch.constant.int 1
%5710 = torch.aten.slice.Tensor %5709, %int0_6667, %int0_6668, %int64_6669, %int1_6670 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6671 = torch.constant.int 6
%5711 = torch.prims.convert_element_type %5710, %int6_6671 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6672 = torch.constant.int 128
%5712 = torch.aten.div.Scalar %5711, %int128_6672 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6673 = torch.constant.float 5.000000e+05
%5713 = torch.aten.pow.Scalar %float5.000000e05_6673, %5712 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5714 = torch.aten.reciprocal %5713 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6674 = torch.constant.float 1.000000e+00
%5715 = torch.aten.mul.Scalar %5714, %float1.000000e00_6674 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6675 = torch.constant.int 131072
%int1_6676 = torch.constant.int 1
%5716 = torch.prim.ListConstruct %int131072_6675, %int1_6676 : (!torch.int, !torch.int) -> !torch.list<int>
%5717 = torch.aten.view %5708, %5716 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5718 = torch.aten.mul.Tensor %5717, %5715 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5719 = torch.aten.cos %5718 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5720 = torch.aten.sin %5718 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5721 = torch.aten.complex %5719, %5720 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6677 = torch.constant.int 1
%5722 = torch.aten.size.int %5665, %int1_6677 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_6678 = torch.constant.int 0
%5723 = torch.aten.add.int %int0_6678, %5722 : !torch.int, !torch.int -> !torch.int
%int0_6679 = torch.constant.int 0
%int0_6680 = torch.constant.int 0
%int1_6681 = torch.constant.int 1
%5724 = torch.aten.slice.Tensor %5721, %int0_6679, %int0_6680, %5723, %int1_6681 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5724, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6682 = torch.constant.int 1
%int0_6683 = torch.constant.int 0
%int9223372036854775807_6684 = torch.constant.int 9223372036854775807
%int1_6685 = torch.constant.int 1
%5725 = torch.aten.slice.Tensor %5724, %int1_6682, %int0_6683, %int9223372036854775807_6684, %int1_6685 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5725, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6686 = torch.constant.int 0
%5726 = torch.aten.unsqueeze %5725, %int0_6686 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5726, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6687 = torch.constant.int 2
%5727 = torch.aten.unsqueeze %5726, %int2_6687 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5727, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6688 = torch.constant.int 3
%int0_6689 = torch.constant.int 0
%int9223372036854775807_6690 = torch.constant.int 9223372036854775807
%int1_6691 = torch.constant.int 1
%5728 = torch.aten.slice.Tensor %5727, %int3_6688, %int0_6689, %int9223372036854775807_6690, %int1_6691 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5728, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5729 = torch_c.to_builtin_tensor %5676 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_6692 = arith.constant 1 : index
%dim_6693 = tensor.dim %5729, %c1_6692 : tensor<4x?x8x128xf16>
%5730 = flow.tensor.bitcast %5729 : tensor<4x?x8x128xf16>{%dim_6693} -> tensor<4x?x8x64xcomplex<f16>>{%dim_6693}
%5731 = torch_c.from_builtin_tensor %5730 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %5731, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%5732 = torch.aten.mul.Tensor %5731, %5728 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %5732, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%5733 = torch_c.to_builtin_tensor %5732 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_6694 = arith.constant 1 : index
%dim_6695 = tensor.dim %5733, %c1_6694 : tensor<4x?x8x64xcomplex<f32>>
%5734 = flow.tensor.bitcast %5733 : tensor<4x?x8x64xcomplex<f32>>{%dim_6695} -> tensor<4x?x8x128xf32>{%dim_6695}
%5735 = torch_c.from_builtin_tensor %5734 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %5735, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_6696 = torch.constant.int 5
%5736 = torch.prims.convert_element_type %5735, %int5_6696 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5736, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_6697 = torch.constant.int 64
%5737 = torch.aten.mul.Scalar %arg2, %int64_6697 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5737, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int54 = torch.constant.int 54
%int1_6698 = torch.constant.int 1
%5738 = torch.aten.add.Scalar %5737, %int54, %int1_6698 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5738, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6699 = torch.constant.int 4
%int32_6700 = torch.constant.int 32
%int8_6701 = torch.constant.int 8
%int128_6702 = torch.constant.int 128
%5739 = torch.prim.ListConstruct %int4_6699, %425, %int32_6700, %int8_6701, %int128_6702 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5740 = torch.aten.view %5736, %5739 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5740, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6703 = torch.constant.int 4
%5741 = torch.aten.mul.int %int4_6703, %425 : !torch.int, !torch.int -> !torch.int
%int32_6704 = torch.constant.int 32
%int8_6705 = torch.constant.int 8
%int128_6706 = torch.constant.int 128
%5742 = torch.prim.ListConstruct %5741, %int32_6704, %int8_6705, %int128_6706 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5743 = torch.aten.view %5740, %5742 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5743, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6707 = torch.constant.int 4
%5744 = torch.aten.mul.int %int4_6707, %425 : !torch.int, !torch.int -> !torch.int
%5745 = torch.prim.ListConstruct %5744 : (!torch.int) -> !torch.list<int>
%5746 = torch.aten.view %5738, %5745 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5746, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_6708 = torch.constant.int 32
%int2_6709 = torch.constant.int 2
%int32_6710 = torch.constant.int 32
%int8_6711 = torch.constant.int 8
%int128_6712 = torch.constant.int 128
%5747 = torch.prim.ListConstruct %416, %int32_6708, %int2_6709, %int32_6710, %int8_6711, %int128_6712 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5748 = torch.aten.view %5580, %5747 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5748, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6713 = torch.constant.int 32
%5749 = torch.aten.mul.int %416, %int32_6713 : !torch.int, !torch.int -> !torch.int
%int2_6714 = torch.constant.int 2
%5750 = torch.aten.mul.int %5749, %int2_6714 : !torch.int, !torch.int -> !torch.int
%int32_6715 = torch.constant.int 32
%int8_6716 = torch.constant.int 8
%int128_6717 = torch.constant.int 128
%5751 = torch.prim.ListConstruct %5750, %int32_6715, %int8_6716, %int128_6717 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5752 = torch.aten.view %5748, %5751 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5752, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%5753 = torch.prim.ListConstruct %5746 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6718 = torch.constant.bool false
%5754 = torch.aten.index_put %5752, %5753, %5743, %false_6718 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5754, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6719 = torch.constant.int 32
%int2_6720 = torch.constant.int 2
%int32_6721 = torch.constant.int 32
%int8_6722 = torch.constant.int 8
%int128_6723 = torch.constant.int 128
%5755 = torch.prim.ListConstruct %416, %int32_6719, %int2_6720, %int32_6721, %int8_6722, %int128_6723 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5756 = torch.aten.view %5754, %5755 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5756, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6724 = torch.constant.int 2097152
%5757 = torch.prim.ListConstruct %416, %int2097152_6724 : (!torch.int, !torch.int) -> !torch.list<int>
%5758 = torch.aten.view %5756, %5757 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5758, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_6725 = torch.constant.int 32
%int2_6726 = torch.constant.int 2
%int32_6727 = torch.constant.int 32
%int8_6728 = torch.constant.int 8
%int128_6729 = torch.constant.int 128
%5759 = torch.prim.ListConstruct %416, %int32_6725, %int2_6726, %int32_6727, %int8_6728, %int128_6729 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5760 = torch.aten.view %5758, %5759 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5760, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6730 = torch.constant.int 32
%int8_6731 = torch.constant.int 8
%int128_6732 = torch.constant.int 128
%5761 = torch.prim.ListConstruct %5750, %int32_6730, %int8_6731, %int128_6732 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5762 = torch.aten.view %5760, %5761 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5762, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6733 = torch.constant.int 4
%int32_6734 = torch.constant.int 32
%int8_6735 = torch.constant.int 8
%int128_6736 = torch.constant.int 128
%5763 = torch.prim.ListConstruct %int4_6733, %425, %int32_6734, %int8_6735, %int128_6736 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5764 = torch.aten.view %5678, %5763 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5764, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6737 = torch.constant.int 4
%5765 = torch.aten.mul.int %int4_6737, %425 : !torch.int, !torch.int -> !torch.int
%int32_6738 = torch.constant.int 32
%int8_6739 = torch.constant.int 8
%int128_6740 = torch.constant.int 128
%5766 = torch.prim.ListConstruct %5765, %int32_6738, %int8_6739, %int128_6740 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5767 = torch.aten.view %5764, %5766 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5767, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_6741 = torch.constant.int 1
%int1_6742 = torch.constant.int 1
%5768 = torch.aten.add.Scalar %5738, %int1_6741, %int1_6742 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5768, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6743 = torch.constant.int 4
%5769 = torch.aten.mul.int %int4_6743, %425 : !torch.int, !torch.int -> !torch.int
%5770 = torch.prim.ListConstruct %5769 : (!torch.int) -> !torch.list<int>
%5771 = torch.aten.view %5768, %5770 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5771, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%5772 = torch.prim.ListConstruct %5771 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6744 = torch.constant.bool false
%5773 = torch.aten.index_put %5762, %5772, %5767, %false_6744 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5773, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6745 = torch.constant.int 32
%int2_6746 = torch.constant.int 2
%int32_6747 = torch.constant.int 32
%int8_6748 = torch.constant.int 8
%int128_6749 = torch.constant.int 128
%5774 = torch.prim.ListConstruct %416, %int32_6745, %int2_6746, %int32_6747, %int8_6748, %int128_6749 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5775 = torch.aten.view %5773, %5774 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5775, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6750 = torch.constant.int 2097152
%5776 = torch.prim.ListConstruct %416, %int2097152_6750 : (!torch.int, !torch.int) -> !torch.list<int>
%5777 = torch.aten.view %5775, %5776 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5777, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_6751 = torch.constant.int -2
%5778 = torch.aten.unsqueeze %5736, %int-2_6751 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5778, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_6752 = torch.constant.int 4
%int8_6753 = torch.constant.int 8
%int4_6754 = torch.constant.int 4
%int128_6755 = torch.constant.int 128
%5779 = torch.prim.ListConstruct %int4_6752, %5722, %int8_6753, %int4_6754, %int128_6755 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6756 = torch.constant.bool false
%5780 = torch.aten.expand %5778, %5779, %false_6756 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5780, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6757 = torch.constant.int 0
%5781 = torch.aten.clone %5780, %int0_6757 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5781, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6758 = torch.constant.int 4
%int32_6759 = torch.constant.int 32
%int128_6760 = torch.constant.int 128
%5782 = torch.prim.ListConstruct %int4_6758, %5722, %int32_6759, %int128_6760 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5783 = torch.aten._unsafe_view %5781, %5782 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5783, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_6761 = torch.constant.int -2
%5784 = torch.aten.unsqueeze %5678, %int-2_6761 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5784, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_6762 = torch.constant.int 1
%5785 = torch.aten.size.int %5672, %int1_6762 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_6763 = torch.constant.int 4
%int8_6764 = torch.constant.int 8
%int4_6765 = torch.constant.int 4
%int128_6766 = torch.constant.int 128
%5786 = torch.prim.ListConstruct %int4_6763, %5785, %int8_6764, %int4_6765, %int128_6766 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6767 = torch.constant.bool false
%5787 = torch.aten.expand %5784, %5786, %false_6767 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5787, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6768 = torch.constant.int 0
%5788 = torch.aten.clone %5787, %int0_6768 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5788, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_6769 = torch.constant.int 4
%int32_6770 = torch.constant.int 32
%int128_6771 = torch.constant.int 128
%5789 = torch.prim.ListConstruct %int4_6769, %5785, %int32_6770, %int128_6771 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5790 = torch.aten._unsafe_view %5788, %5789 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5790, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_6772 = torch.constant.int 1
%int2_6773 = torch.constant.int 2
%5791 = torch.aten.transpose.int %5707, %int1_6772, %int2_6773 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5791, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6774 = torch.constant.int 1
%int2_6775 = torch.constant.int 2
%5792 = torch.aten.transpose.int %5783, %int1_6774, %int2_6775 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5792, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6776 = torch.constant.int 1
%int2_6777 = torch.constant.int 2
%5793 = torch.aten.transpose.int %5790, %int1_6776, %int2_6777 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5793, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_6778 = torch.constant.float 0.000000e+00
%false_6779 = torch.constant.bool false
%none_6780 = torch.constant.none
%5794:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5791, %5792, %5793, %float0.000000e00_6778, %false_6779, %320, %none_6780) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5794#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_6781 = torch.constant.int 1
%int2_6782 = torch.constant.int 2
%5795 = torch.aten.transpose.int %5794#0, %int1_6781, %int2_6782 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5795, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6783 = torch.constant.int 4
%int4096_6784 = torch.constant.int 4096
%5796 = torch.prim.ListConstruct %int4_6783, %5693, %int4096_6784 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5797 = torch.aten.view %5795, %5796 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5797, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6785 = torch.constant.int -2
%int-1_6786 = torch.constant.int -1
%5798 = torch.aten.transpose.int %248, %int-2_6785, %int-1_6786 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6787 = torch.constant.int 4
%5799 = torch.aten.mul.int %int4_6787, %5693 : !torch.int, !torch.int -> !torch.int
%int4096_6788 = torch.constant.int 4096
%5800 = torch.prim.ListConstruct %5799, %int4096_6788 : (!torch.int, !torch.int) -> !torch.list<int>
%5801 = torch.aten.view %5797, %5800 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5801, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5802 = torch.aten.mm %5801, %5798 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5802, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6789 = torch.constant.int 4
%int4096_6790 = torch.constant.int 4096
%5803 = torch.prim.ListConstruct %int4_6789, %5693, %int4096_6790 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5804 = torch.aten.view %5802, %5803 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5804, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6791 = torch.constant.int 1
%5805 = torch.aten.add.Tensor %5642, %5804, %int1_6791 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5805, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6792 = torch.constant.int 6
%5806 = torch.prims.convert_element_type %5805, %int6_6792 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5806, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6793 = torch.constant.int 2
%5807 = torch.aten.pow.Tensor_Scalar %5806, %int2_6793 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5807, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6794 = torch.constant.int -1
%5808 = torch.prim.ListConstruct %int-1_6794 : (!torch.int) -> !torch.list<int>
%true_6795 = torch.constant.bool true
%none_6796 = torch.constant.none
%5809 = torch.aten.mean.dim %5807, %5808, %true_6795, %none_6796 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5809, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6797 = torch.constant.float 9.9999997473787516E-6
%int1_6798 = torch.constant.int 1
%5810 = torch.aten.add.Scalar %5809, %float9.999990e-06_6797, %int1_6798 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5810, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5811 = torch.aten.rsqrt %5810 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5811, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5812 = torch.aten.mul.Tensor %5806, %5811 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5812, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5813 = torch.aten.mul.Tensor %249, %5812 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5813, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6799 = torch.constant.int 5
%5814 = torch.prims.convert_element_type %5813, %int5_6799 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5814, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6800 = torch.constant.int -2
%int-1_6801 = torch.constant.int -1
%5815 = torch.aten.transpose.int %250, %int-2_6800, %int-1_6801 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6802 = torch.constant.int 4
%5816 = torch.aten.mul.int %int4_6802, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6803 = torch.constant.int 4096
%5817 = torch.prim.ListConstruct %5816, %int4096_6803 : (!torch.int, !torch.int) -> !torch.list<int>
%5818 = torch.aten.view %5814, %5817 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5818, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5819 = torch.aten.mm %5818, %5815 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5819, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6804 = torch.constant.int 4
%int14336_6805 = torch.constant.int 14336
%5820 = torch.prim.ListConstruct %int4_6804, %294, %int14336_6805 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5821 = torch.aten.view %5819, %5820 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5821, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5822 = torch.aten.silu %5821 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5822, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6806 = torch.constant.int -2
%int-1_6807 = torch.constant.int -1
%5823 = torch.aten.transpose.int %251, %int-2_6806, %int-1_6807 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_6808 = torch.constant.int 4
%5824 = torch.aten.mul.int %int4_6808, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6809 = torch.constant.int 4096
%5825 = torch.prim.ListConstruct %5824, %int4096_6809 : (!torch.int, !torch.int) -> !torch.list<int>
%5826 = torch.aten.view %5814, %5825 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5826, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5827 = torch.aten.mm %5826, %5823 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5827, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_6810 = torch.constant.int 4
%int14336_6811 = torch.constant.int 14336
%5828 = torch.prim.ListConstruct %int4_6810, %294, %int14336_6811 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5829 = torch.aten.view %5827, %5828 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5829, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%5830 = torch.aten.mul.Tensor %5822, %5829 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %5830, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_6812 = torch.constant.int -2
%int-1_6813 = torch.constant.int -1
%5831 = torch.aten.transpose.int %252, %int-2_6812, %int-1_6813 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_6814 = torch.constant.int 1
%5832 = torch.aten.size.int %5821, %int1_6814 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_6815 = torch.constant.int 4
%5833 = torch.aten.mul.int %int4_6815, %5832 : !torch.int, !torch.int -> !torch.int
%int14336_6816 = torch.constant.int 14336
%5834 = torch.prim.ListConstruct %5833, %int14336_6816 : (!torch.int, !torch.int) -> !torch.list<int>
%5835 = torch.aten.view %5830, %5834 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %5835, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%5836 = torch.aten.mm %5835, %5831 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5836, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6817 = torch.constant.int 4
%int4096_6818 = torch.constant.int 4096
%5837 = torch.prim.ListConstruct %int4_6817, %5832, %int4096_6818 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5838 = torch.aten.view %5836, %5837 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5838, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_6819 = torch.constant.int 1
%5839 = torch.aten.add.Tensor %5805, %5838, %int1_6819 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5839, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_6820 = torch.constant.int 6
%5840 = torch.prims.convert_element_type %5839, %int6_6820 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5840, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_6821 = torch.constant.int 2
%5841 = torch.aten.pow.Tensor_Scalar %5840, %int2_6821 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5841, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_6822 = torch.constant.int -1
%5842 = torch.prim.ListConstruct %int-1_6822 : (!torch.int) -> !torch.list<int>
%true_6823 = torch.constant.bool true
%none_6824 = torch.constant.none
%5843 = torch.aten.mean.dim %5841, %5842, %true_6823, %none_6824 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5843, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_6825 = torch.constant.float 9.9999997473787516E-6
%int1_6826 = torch.constant.int 1
%5844 = torch.aten.add.Scalar %5843, %float9.999990e-06_6825, %int1_6826 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5844, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5845 = torch.aten.rsqrt %5844 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %5845, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%5846 = torch.aten.mul.Tensor %5840, %5845 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5846, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%5847 = torch.aten.mul.Tensor %253, %5846 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %5847, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_6827 = torch.constant.int 5
%5848 = torch.prims.convert_element_type %5847, %int5_6827 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5848, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6828 = torch.constant.int -2
%int-1_6829 = torch.constant.int -1
%5849 = torch.aten.transpose.int %254, %int-2_6828, %int-1_6829 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_6830 = torch.constant.int 4
%5850 = torch.aten.mul.int %int4_6830, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6831 = torch.constant.int 4096
%5851 = torch.prim.ListConstruct %5850, %int4096_6831 : (!torch.int, !torch.int) -> !torch.list<int>
%5852 = torch.aten.view %5848, %5851 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5852, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5853 = torch.aten.mm %5852, %5849 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5853, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_6832 = torch.constant.int 4
%int4096_6833 = torch.constant.int 4096
%5854 = torch.prim.ListConstruct %int4_6832, %294, %int4096_6833 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5855 = torch.aten.view %5853, %5854 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5855, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_6834 = torch.constant.int -2
%int-1_6835 = torch.constant.int -1
%5856 = torch.aten.transpose.int %255, %int-2_6834, %int-1_6835 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6836 = torch.constant.int 4
%5857 = torch.aten.mul.int %int4_6836, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6837 = torch.constant.int 4096
%5858 = torch.prim.ListConstruct %5857, %int4096_6837 : (!torch.int, !torch.int) -> !torch.list<int>
%5859 = torch.aten.view %5848, %5858 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5859, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5860 = torch.aten.mm %5859, %5856 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5860, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6838 = torch.constant.int 4
%int1024_6839 = torch.constant.int 1024
%5861 = torch.prim.ListConstruct %int4_6838, %294, %int1024_6839 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5862 = torch.aten.view %5860, %5861 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5862, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_6840 = torch.constant.int -2
%int-1_6841 = torch.constant.int -1
%5863 = torch.aten.transpose.int %256, %int-2_6840, %int-1_6841 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_6842 = torch.constant.int 4
%5864 = torch.aten.mul.int %int4_6842, %294 : !torch.int, !torch.int -> !torch.int
%int4096_6843 = torch.constant.int 4096
%5865 = torch.prim.ListConstruct %5864, %int4096_6843 : (!torch.int, !torch.int) -> !torch.list<int>
%5866 = torch.aten.view %5848, %5865 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5866, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5867 = torch.aten.mm %5866, %5863 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %5867, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_6844 = torch.constant.int 4
%int1024_6845 = torch.constant.int 1024
%5868 = torch.prim.ListConstruct %int4_6844, %294, %int1024_6845 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5869 = torch.aten.view %5867, %5868 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %5869, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_6846 = torch.constant.int 4
%int32_6847 = torch.constant.int 32
%int128_6848 = torch.constant.int 128
%5870 = torch.prim.ListConstruct %int4_6846, %294, %int32_6847, %int128_6848 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5871 = torch.aten.view %5855, %5870 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5871, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_6849 = torch.constant.int 4
%int8_6850 = torch.constant.int 8
%int128_6851 = torch.constant.int 128
%5872 = torch.prim.ListConstruct %int4_6849, %294, %int8_6850, %int128_6851 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5873 = torch.aten.view %5862, %5872 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5873, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_6852 = torch.constant.int 4
%int8_6853 = torch.constant.int 8
%int128_6854 = torch.constant.int 128
%5874 = torch.prim.ListConstruct %int4_6852, %294, %int8_6853, %int128_6854 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5875 = torch.aten.view %5869, %5874 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5875, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_6855 = torch.constant.int 131072
%none_6856 = torch.constant.none
%none_6857 = torch.constant.none
%cpu_6858 = torch.constant.device "cpu"
%false_6859 = torch.constant.bool false
%5876 = torch.aten.arange %int131072_6855, %none_6856, %none_6857, %cpu_6858, %false_6859 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6860 = torch.constant.int 0
%int128_6861 = torch.constant.int 128
%int2_6862 = torch.constant.int 2
%none_6863 = torch.constant.none
%none_6864 = torch.constant.none
%cpu_6865 = torch.constant.device "cpu"
%false_6866 = torch.constant.bool false
%5877 = torch.aten.arange.start_step %int0_6860, %int128_6861, %int2_6862, %none_6863, %none_6864, %cpu_6865, %false_6866 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6867 = torch.constant.int 0
%int0_6868 = torch.constant.int 0
%int64_6869 = torch.constant.int 64
%int1_6870 = torch.constant.int 1
%5878 = torch.aten.slice.Tensor %5877, %int0_6867, %int0_6868, %int64_6869, %int1_6870 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6871 = torch.constant.int 6
%5879 = torch.prims.convert_element_type %5878, %int6_6871 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6872 = torch.constant.int 128
%5880 = torch.aten.div.Scalar %5879, %int128_6872 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6873 = torch.constant.float 5.000000e+05
%5881 = torch.aten.pow.Scalar %float5.000000e05_6873, %5880 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5882 = torch.aten.reciprocal %5881 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6874 = torch.constant.float 1.000000e+00
%5883 = torch.aten.mul.Scalar %5882, %float1.000000e00_6874 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6875 = torch.constant.int 131072
%int1_6876 = torch.constant.int 1
%5884 = torch.prim.ListConstruct %int131072_6875, %int1_6876 : (!torch.int, !torch.int) -> !torch.list<int>
%5885 = torch.aten.view %5876, %5884 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5886 = torch.aten.mul.Tensor %5885, %5883 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5887 = torch.aten.cos %5886 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5888 = torch.aten.sin %5886 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5889 = torch.aten.complex %5887, %5888 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6877 = torch.constant.int 1
%5890 = torch.aten.size.int %5855, %int1_6877 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_6878 = torch.constant.int 0
%5891 = torch.aten.add.int %int0_6878, %5890 : !torch.int, !torch.int -> !torch.int
%int0_6879 = torch.constant.int 0
%int0_6880 = torch.constant.int 0
%int1_6881 = torch.constant.int 1
%5892 = torch.aten.slice.Tensor %5889, %int0_6879, %int0_6880, %5891, %int1_6881 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5892, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6882 = torch.constant.int 1
%int0_6883 = torch.constant.int 0
%int9223372036854775807_6884 = torch.constant.int 9223372036854775807
%int1_6885 = torch.constant.int 1
%5893 = torch.aten.slice.Tensor %5892, %int1_6882, %int0_6883, %int9223372036854775807_6884, %int1_6885 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5893, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6886 = torch.constant.int 0
%5894 = torch.aten.unsqueeze %5893, %int0_6886 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5894, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6887 = torch.constant.int 2
%5895 = torch.aten.unsqueeze %5894, %int2_6887 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5895, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6888 = torch.constant.int 3
%int0_6889 = torch.constant.int 0
%int9223372036854775807_6890 = torch.constant.int 9223372036854775807
%int1_6891 = torch.constant.int 1
%5896 = torch.aten.slice.Tensor %5895, %int3_6888, %int0_6889, %int9223372036854775807_6890, %int1_6891 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5896, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5897 = torch_c.to_builtin_tensor %5871 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_6892 = arith.constant 1 : index
%dim_6893 = tensor.dim %5897, %c1_6892 : tensor<4x?x32x128xf16>
%5898 = flow.tensor.bitcast %5897 : tensor<4x?x32x128xf16>{%dim_6893} -> tensor<4x?x32x64xcomplex<f16>>{%dim_6893}
%5899 = torch_c.from_builtin_tensor %5898 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %5899, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%5900 = torch.aten.mul.Tensor %5899, %5896 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %5900, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%5901 = torch_c.to_builtin_tensor %5900 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_6894 = arith.constant 1 : index
%dim_6895 = tensor.dim %5901, %c1_6894 : tensor<4x?x32x64xcomplex<f32>>
%5902 = flow.tensor.bitcast %5901 : tensor<4x?x32x64xcomplex<f32>>{%dim_6895} -> tensor<4x?x32x128xf32>{%dim_6895}
%5903 = torch_c.from_builtin_tensor %5902 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %5903, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_6896 = torch.constant.int 5
%5904 = torch.prims.convert_element_type %5903, %int5_6896 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5904, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_6897 = torch.constant.int 131072
%none_6898 = torch.constant.none
%none_6899 = torch.constant.none
%cpu_6900 = torch.constant.device "cpu"
%false_6901 = torch.constant.bool false
%5905 = torch.aten.arange %int131072_6897, %none_6898, %none_6899, %cpu_6900, %false_6901 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_6902 = torch.constant.int 0
%int128_6903 = torch.constant.int 128
%int2_6904 = torch.constant.int 2
%none_6905 = torch.constant.none
%none_6906 = torch.constant.none
%cpu_6907 = torch.constant.device "cpu"
%false_6908 = torch.constant.bool false
%5906 = torch.aten.arange.start_step %int0_6902, %int128_6903, %int2_6904, %none_6905, %none_6906, %cpu_6907, %false_6908 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_6909 = torch.constant.int 0
%int0_6910 = torch.constant.int 0
%int64_6911 = torch.constant.int 64
%int1_6912 = torch.constant.int 1
%5907 = torch.aten.slice.Tensor %5906, %int0_6909, %int0_6910, %int64_6911, %int1_6912 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_6913 = torch.constant.int 6
%5908 = torch.prims.convert_element_type %5907, %int6_6913 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_6914 = torch.constant.int 128
%5909 = torch.aten.div.Scalar %5908, %int128_6914 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_6915 = torch.constant.float 5.000000e+05
%5910 = torch.aten.pow.Scalar %float5.000000e05_6915, %5909 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%5911 = torch.aten.reciprocal %5910 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_6916 = torch.constant.float 1.000000e+00
%5912 = torch.aten.mul.Scalar %5911, %float1.000000e00_6916 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_6917 = torch.constant.int 131072
%int1_6918 = torch.constant.int 1
%5913 = torch.prim.ListConstruct %int131072_6917, %int1_6918 : (!torch.int, !torch.int) -> !torch.list<int>
%5914 = torch.aten.view %5905, %5913 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%5915 = torch.aten.mul.Tensor %5914, %5912 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%5916 = torch.aten.cos %5915 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5917 = torch.aten.sin %5915 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%5918 = torch.aten.complex %5916, %5917 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_6919 = torch.constant.int 1
%5919 = torch.aten.size.int %5862, %int1_6919 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_6920 = torch.constant.int 0
%5920 = torch.aten.add.int %int0_6920, %5919 : !torch.int, !torch.int -> !torch.int
%int0_6921 = torch.constant.int 0
%int0_6922 = torch.constant.int 0
%int1_6923 = torch.constant.int 1
%5921 = torch.aten.slice.Tensor %5918, %int0_6921, %int0_6922, %5920, %int1_6923 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5921, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_6924 = torch.constant.int 1
%int0_6925 = torch.constant.int 0
%int9223372036854775807_6926 = torch.constant.int 9223372036854775807
%int1_6927 = torch.constant.int 1
%5922 = torch.aten.slice.Tensor %5921, %int1_6924, %int0_6925, %int9223372036854775807_6926, %int1_6927 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %5922, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_6928 = torch.constant.int 0
%5923 = torch.aten.unsqueeze %5922, %int0_6928 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %5923, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_6929 = torch.constant.int 2
%5924 = torch.aten.unsqueeze %5923, %int2_6929 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5924, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_6930 = torch.constant.int 3
%int0_6931 = torch.constant.int 0
%int9223372036854775807_6932 = torch.constant.int 9223372036854775807
%int1_6933 = torch.constant.int 1
%5925 = torch.aten.slice.Tensor %5924, %int3_6930, %int0_6931, %int9223372036854775807_6932, %int1_6933 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %5925, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%5926 = torch_c.to_builtin_tensor %5873 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_6934 = arith.constant 1 : index
%dim_6935 = tensor.dim %5926, %c1_6934 : tensor<4x?x8x128xf16>
%5927 = flow.tensor.bitcast %5926 : tensor<4x?x8x128xf16>{%dim_6935} -> tensor<4x?x8x64xcomplex<f16>>{%dim_6935}
%5928 = torch_c.from_builtin_tensor %5927 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %5928, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%5929 = torch.aten.mul.Tensor %5928, %5925 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %5929, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%5930 = torch_c.to_builtin_tensor %5929 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_6936 = arith.constant 1 : index
%dim_6937 = tensor.dim %5930, %c1_6936 : tensor<4x?x8x64xcomplex<f32>>
%5931 = flow.tensor.bitcast %5930 : tensor<4x?x8x64xcomplex<f32>>{%dim_6937} -> tensor<4x?x8x128xf32>{%dim_6937}
%5932 = torch_c.from_builtin_tensor %5931 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %5932, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_6938 = torch.constant.int 5
%5933 = torch.prims.convert_element_type %5932, %int5_6938 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %5933, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_6939 = torch.constant.int 64
%5934 = torch.aten.mul.Scalar %arg2, %int64_6939 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5934, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int56 = torch.constant.int 56
%int1_6940 = torch.constant.int 1
%5935 = torch.aten.add.Scalar %5934, %int56, %int1_6940 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5935, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6941 = torch.constant.int 4
%int32_6942 = torch.constant.int 32
%int8_6943 = torch.constant.int 8
%int128_6944 = torch.constant.int 128
%5936 = torch.prim.ListConstruct %int4_6941, %425, %int32_6942, %int8_6943, %int128_6944 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5937 = torch.aten.view %5933, %5936 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5937, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6945 = torch.constant.int 4
%5938 = torch.aten.mul.int %int4_6945, %425 : !torch.int, !torch.int -> !torch.int
%int32_6946 = torch.constant.int 32
%int8_6947 = torch.constant.int 8
%int128_6948 = torch.constant.int 128
%5939 = torch.prim.ListConstruct %5938, %int32_6946, %int8_6947, %int128_6948 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5940 = torch.aten.view %5937, %5939 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5940, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6949 = torch.constant.int 4
%5941 = torch.aten.mul.int %int4_6949, %425 : !torch.int, !torch.int -> !torch.int
%5942 = torch.prim.ListConstruct %5941 : (!torch.int) -> !torch.list<int>
%5943 = torch.aten.view %5935, %5942 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5943, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_6950 = torch.constant.int 32
%int2_6951 = torch.constant.int 2
%int32_6952 = torch.constant.int 32
%int8_6953 = torch.constant.int 8
%int128_6954 = torch.constant.int 128
%5944 = torch.prim.ListConstruct %416, %int32_6950, %int2_6951, %int32_6952, %int8_6953, %int128_6954 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5945 = torch.aten.view %5777, %5944 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5945, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6955 = torch.constant.int 32
%5946 = torch.aten.mul.int %416, %int32_6955 : !torch.int, !torch.int -> !torch.int
%int2_6956 = torch.constant.int 2
%5947 = torch.aten.mul.int %5946, %int2_6956 : !torch.int, !torch.int -> !torch.int
%int32_6957 = torch.constant.int 32
%int8_6958 = torch.constant.int 8
%int128_6959 = torch.constant.int 128
%5948 = torch.prim.ListConstruct %5947, %int32_6957, %int8_6958, %int128_6959 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5949 = torch.aten.view %5945, %5948 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5949, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%5950 = torch.prim.ListConstruct %5943 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6960 = torch.constant.bool false
%5951 = torch.aten.index_put %5949, %5950, %5940, %false_6960 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5951, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6961 = torch.constant.int 32
%int2_6962 = torch.constant.int 2
%int32_6963 = torch.constant.int 32
%int8_6964 = torch.constant.int 8
%int128_6965 = torch.constant.int 128
%5952 = torch.prim.ListConstruct %416, %int32_6961, %int2_6962, %int32_6963, %int8_6964, %int128_6965 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5953 = torch.aten.view %5951, %5952 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5953, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6966 = torch.constant.int 2097152
%5954 = torch.prim.ListConstruct %416, %int2097152_6966 : (!torch.int, !torch.int) -> !torch.list<int>
%5955 = torch.aten.view %5953, %5954 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5955, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_6967 = torch.constant.int 32
%int2_6968 = torch.constant.int 2
%int32_6969 = torch.constant.int 32
%int8_6970 = torch.constant.int 8
%int128_6971 = torch.constant.int 128
%5956 = torch.prim.ListConstruct %416, %int32_6967, %int2_6968, %int32_6969, %int8_6970, %int128_6971 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5957 = torch.aten.view %5955, %5956 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5957, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_6972 = torch.constant.int 32
%int8_6973 = torch.constant.int 8
%int128_6974 = torch.constant.int 128
%5958 = torch.prim.ListConstruct %5947, %int32_6972, %int8_6973, %int128_6974 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5959 = torch.aten.view %5957, %5958 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5959, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_6975 = torch.constant.int 4
%int32_6976 = torch.constant.int 32
%int8_6977 = torch.constant.int 8
%int128_6978 = torch.constant.int 128
%5960 = torch.prim.ListConstruct %int4_6975, %425, %int32_6976, %int8_6977, %int128_6978 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5961 = torch.aten.view %5875, %5960 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %5961, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_6979 = torch.constant.int 4
%5962 = torch.aten.mul.int %int4_6979, %425 : !torch.int, !torch.int -> !torch.int
%int32_6980 = torch.constant.int 32
%int8_6981 = torch.constant.int 8
%int128_6982 = torch.constant.int 128
%5963 = torch.prim.ListConstruct %5962, %int32_6980, %int8_6981, %int128_6982 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5964 = torch.aten.view %5961, %5963 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5964, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_6983 = torch.constant.int 1
%int1_6984 = torch.constant.int 1
%5965 = torch.aten.add.Scalar %5935, %int1_6983, %int1_6984 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %5965, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_6985 = torch.constant.int 4
%5966 = torch.aten.mul.int %int4_6985, %425 : !torch.int, !torch.int -> !torch.int
%5967 = torch.prim.ListConstruct %5966 : (!torch.int) -> !torch.list<int>
%5968 = torch.aten.view %5965, %5967 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %5968, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%5969 = torch.prim.ListConstruct %5968 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_6986 = torch.constant.bool false
%5970 = torch.aten.index_put %5959, %5969, %5964, %false_6986 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %5970, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_6987 = torch.constant.int 32
%int2_6988 = torch.constant.int 2
%int32_6989 = torch.constant.int 32
%int8_6990 = torch.constant.int 8
%int128_6991 = torch.constant.int 128
%5971 = torch.prim.ListConstruct %416, %int32_6987, %int2_6988, %int32_6989, %int8_6990, %int128_6991 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5972 = torch.aten.view %5970, %5971 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %5972, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_6992 = torch.constant.int 2097152
%5973 = torch.prim.ListConstruct %416, %int2097152_6992 : (!torch.int, !torch.int) -> !torch.list<int>
%5974 = torch.aten.view %5972, %5973 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %5974, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_6993 = torch.constant.int -2
%5975 = torch.aten.unsqueeze %5933, %int-2_6993 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5975, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_6994 = torch.constant.int 4
%int8_6995 = torch.constant.int 8
%int4_6996 = torch.constant.int 4
%int128_6997 = torch.constant.int 128
%5976 = torch.prim.ListConstruct %int4_6994, %5919, %int8_6995, %int4_6996, %int128_6997 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_6998 = torch.constant.bool false
%5977 = torch.aten.expand %5975, %5976, %false_6998 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5977, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_6999 = torch.constant.int 0
%5978 = torch.aten.clone %5977, %int0_6999 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5978, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7000 = torch.constant.int 4
%int32_7001 = torch.constant.int 32
%int128_7002 = torch.constant.int 128
%5979 = torch.prim.ListConstruct %int4_7000, %5919, %int32_7001, %int128_7002 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5980 = torch.aten._unsafe_view %5978, %5979 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5980, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_7003 = torch.constant.int -2
%5981 = torch.aten.unsqueeze %5875, %int-2_7003 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %5981, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_7004 = torch.constant.int 1
%5982 = torch.aten.size.int %5869, %int1_7004 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_7005 = torch.constant.int 4
%int8_7006 = torch.constant.int 8
%int4_7007 = torch.constant.int 4
%int128_7008 = torch.constant.int 128
%5983 = torch.prim.ListConstruct %int4_7005, %5982, %int8_7006, %int4_7007, %int128_7008 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7009 = torch.constant.bool false
%5984 = torch.aten.expand %5981, %5983, %false_7009 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5984, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7010 = torch.constant.int 0
%5985 = torch.aten.clone %5984, %int0_7010 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %5985, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7011 = torch.constant.int 4
%int32_7012 = torch.constant.int 32
%int128_7013 = torch.constant.int 128
%5986 = torch.prim.ListConstruct %int4_7011, %5982, %int32_7012, %int128_7013 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5987 = torch.aten._unsafe_view %5985, %5986 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5987, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_7014 = torch.constant.int 1
%int2_7015 = torch.constant.int 2
%5988 = torch.aten.transpose.int %5904, %int1_7014, %int2_7015 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5988, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7016 = torch.constant.int 1
%int2_7017 = torch.constant.int 2
%5989 = torch.aten.transpose.int %5980, %int1_7016, %int2_7017 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5989, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7018 = torch.constant.int 1
%int2_7019 = torch.constant.int 2
%5990 = torch.aten.transpose.int %5987, %int1_7018, %int2_7019 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %5990, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_7020 = torch.constant.float 0.000000e+00
%false_7021 = torch.constant.bool false
%none_7022 = torch.constant.none
%5991:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%5988, %5989, %5990, %float0.000000e00_7020, %false_7021, %320, %none_7022) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %5991#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7023 = torch.constant.int 1
%int2_7024 = torch.constant.int 2
%5992 = torch.aten.transpose.int %5991#0, %int1_7023, %int2_7024 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %5992, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7025 = torch.constant.int 4
%int4096_7026 = torch.constant.int 4096
%5993 = torch.prim.ListConstruct %int4_7025, %5890, %int4096_7026 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%5994 = torch.aten.view %5992, %5993 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %5994, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7027 = torch.constant.int -2
%int-1_7028 = torch.constant.int -1
%5995 = torch.aten.transpose.int %257, %int-2_7027, %int-1_7028 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7029 = torch.constant.int 4
%5996 = torch.aten.mul.int %int4_7029, %5890 : !torch.int, !torch.int -> !torch.int
%int4096_7030 = torch.constant.int 4096
%5997 = torch.prim.ListConstruct %5996, %int4096_7030 : (!torch.int, !torch.int) -> !torch.list<int>
%5998 = torch.aten.view %5994, %5997 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5998, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%5999 = torch.aten.mm %5998, %5995 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %5999, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7031 = torch.constant.int 4
%int4096_7032 = torch.constant.int 4096
%6000 = torch.prim.ListConstruct %int4_7031, %5890, %int4096_7032 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6001 = torch.aten.view %5999, %6000 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6001, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7033 = torch.constant.int 1
%6002 = torch.aten.add.Tensor %5839, %6001, %int1_7033 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6002, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7034 = torch.constant.int 6
%6003 = torch.prims.convert_element_type %6002, %int6_7034 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6003, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7035 = torch.constant.int 2
%6004 = torch.aten.pow.Tensor_Scalar %6003, %int2_7035 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6004, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7036 = torch.constant.int -1
%6005 = torch.prim.ListConstruct %int-1_7036 : (!torch.int) -> !torch.list<int>
%true_7037 = torch.constant.bool true
%none_7038 = torch.constant.none
%6006 = torch.aten.mean.dim %6004, %6005, %true_7037, %none_7038 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6006, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7039 = torch.constant.float 9.9999997473787516E-6
%int1_7040 = torch.constant.int 1
%6007 = torch.aten.add.Scalar %6006, %float9.999990e-06_7039, %int1_7040 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6007, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6008 = torch.aten.rsqrt %6007 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6008, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6009 = torch.aten.mul.Tensor %6003, %6008 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6009, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6010 = torch.aten.mul.Tensor %258, %6009 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6010, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7041 = torch.constant.int 5
%6011 = torch.prims.convert_element_type %6010, %int5_7041 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6011, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7042 = torch.constant.int -2
%int-1_7043 = torch.constant.int -1
%6012 = torch.aten.transpose.int %259, %int-2_7042, %int-1_7043 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7044 = torch.constant.int 4
%6013 = torch.aten.mul.int %int4_7044, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7045 = torch.constant.int 4096
%6014 = torch.prim.ListConstruct %6013, %int4096_7045 : (!torch.int, !torch.int) -> !torch.list<int>
%6015 = torch.aten.view %6011, %6014 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6015, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6016 = torch.aten.mm %6015, %6012 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6016, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7046 = torch.constant.int 4
%int14336_7047 = torch.constant.int 14336
%6017 = torch.prim.ListConstruct %int4_7046, %294, %int14336_7047 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6018 = torch.aten.view %6016, %6017 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6018, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6019 = torch.aten.silu %6018 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6019, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7048 = torch.constant.int -2
%int-1_7049 = torch.constant.int -1
%6020 = torch.aten.transpose.int %260, %int-2_7048, %int-1_7049 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7050 = torch.constant.int 4
%6021 = torch.aten.mul.int %int4_7050, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7051 = torch.constant.int 4096
%6022 = torch.prim.ListConstruct %6021, %int4096_7051 : (!torch.int, !torch.int) -> !torch.list<int>
%6023 = torch.aten.view %6011, %6022 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6023, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6024 = torch.aten.mm %6023, %6020 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6024, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7052 = torch.constant.int 4
%int14336_7053 = torch.constant.int 14336
%6025 = torch.prim.ListConstruct %int4_7052, %294, %int14336_7053 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6026 = torch.aten.view %6024, %6025 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6026, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6027 = torch.aten.mul.Tensor %6019, %6026 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6027, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7054 = torch.constant.int -2
%int-1_7055 = torch.constant.int -1
%6028 = torch.aten.transpose.int %261, %int-2_7054, %int-1_7055 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_7056 = torch.constant.int 1
%6029 = torch.aten.size.int %6018, %int1_7056 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_7057 = torch.constant.int 4
%6030 = torch.aten.mul.int %int4_7057, %6029 : !torch.int, !torch.int -> !torch.int
%int14336_7058 = torch.constant.int 14336
%6031 = torch.prim.ListConstruct %6030, %int14336_7058 : (!torch.int, !torch.int) -> !torch.list<int>
%6032 = torch.aten.view %6027, %6031 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6032, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%6033 = torch.aten.mm %6032, %6028 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6033, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7059 = torch.constant.int 4
%int4096_7060 = torch.constant.int 4096
%6034 = torch.prim.ListConstruct %int4_7059, %6029, %int4096_7060 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6035 = torch.aten.view %6033, %6034 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6035, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7061 = torch.constant.int 1
%6036 = torch.aten.add.Tensor %6002, %6035, %int1_7061 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6036, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7062 = torch.constant.int 6
%6037 = torch.prims.convert_element_type %6036, %int6_7062 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6037, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7063 = torch.constant.int 2
%6038 = torch.aten.pow.Tensor_Scalar %6037, %int2_7063 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6038, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7064 = torch.constant.int -1
%6039 = torch.prim.ListConstruct %int-1_7064 : (!torch.int) -> !torch.list<int>
%true_7065 = torch.constant.bool true
%none_7066 = torch.constant.none
%6040 = torch.aten.mean.dim %6038, %6039, %true_7065, %none_7066 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6040, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7067 = torch.constant.float 9.9999997473787516E-6
%int1_7068 = torch.constant.int 1
%6041 = torch.aten.add.Scalar %6040, %float9.999990e-06_7067, %int1_7068 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6041, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6042 = torch.aten.rsqrt %6041 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6042, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6043 = torch.aten.mul.Tensor %6037, %6042 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6043, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6044 = torch.aten.mul.Tensor %262, %6043 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6044, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7069 = torch.constant.int 5
%6045 = torch.prims.convert_element_type %6044, %int5_7069 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6045, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7070 = torch.constant.int -2
%int-1_7071 = torch.constant.int -1
%6046 = torch.aten.transpose.int %263, %int-2_7070, %int-1_7071 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7072 = torch.constant.int 4
%6047 = torch.aten.mul.int %int4_7072, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7073 = torch.constant.int 4096
%6048 = torch.prim.ListConstruct %6047, %int4096_7073 : (!torch.int, !torch.int) -> !torch.list<int>
%6049 = torch.aten.view %6045, %6048 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6049, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6050 = torch.aten.mm %6049, %6046 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6050, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7074 = torch.constant.int 4
%int4096_7075 = torch.constant.int 4096
%6051 = torch.prim.ListConstruct %int4_7074, %294, %int4096_7075 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6052 = torch.aten.view %6050, %6051 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6052, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7076 = torch.constant.int -2
%int-1_7077 = torch.constant.int -1
%6053 = torch.aten.transpose.int %264, %int-2_7076, %int-1_7077 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7078 = torch.constant.int 4
%6054 = torch.aten.mul.int %int4_7078, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7079 = torch.constant.int 4096
%6055 = torch.prim.ListConstruct %6054, %int4096_7079 : (!torch.int, !torch.int) -> !torch.list<int>
%6056 = torch.aten.view %6045, %6055 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6056, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6057 = torch.aten.mm %6056, %6053 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6057, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7080 = torch.constant.int 4
%int1024_7081 = torch.constant.int 1024
%6058 = torch.prim.ListConstruct %int4_7080, %294, %int1024_7081 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6059 = torch.aten.view %6057, %6058 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6059, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_7082 = torch.constant.int -2
%int-1_7083 = torch.constant.int -1
%6060 = torch.aten.transpose.int %265, %int-2_7082, %int-1_7083 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7084 = torch.constant.int 4
%6061 = torch.aten.mul.int %int4_7084, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7085 = torch.constant.int 4096
%6062 = torch.prim.ListConstruct %6061, %int4096_7085 : (!torch.int, !torch.int) -> !torch.list<int>
%6063 = torch.aten.view %6045, %6062 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6063, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6064 = torch.aten.mm %6063, %6060 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6064, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7086 = torch.constant.int 4
%int1024_7087 = torch.constant.int 1024
%6065 = torch.prim.ListConstruct %int4_7086, %294, %int1024_7087 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6066 = torch.aten.view %6064, %6065 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6066, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_7088 = torch.constant.int 4
%int32_7089 = torch.constant.int 32
%int128_7090 = torch.constant.int 128
%6067 = torch.prim.ListConstruct %int4_7088, %294, %int32_7089, %int128_7090 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6068 = torch.aten.view %6052, %6067 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6068, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7091 = torch.constant.int 4
%int8_7092 = torch.constant.int 8
%int128_7093 = torch.constant.int 128
%6069 = torch.prim.ListConstruct %int4_7091, %294, %int8_7092, %int128_7093 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6070 = torch.aten.view %6059, %6069 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6070, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_7094 = torch.constant.int 4
%int8_7095 = torch.constant.int 8
%int128_7096 = torch.constant.int 128
%6071 = torch.prim.ListConstruct %int4_7094, %294, %int8_7095, %int128_7096 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6072 = torch.aten.view %6066, %6071 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6072, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_7097 = torch.constant.int 131072
%none_7098 = torch.constant.none
%none_7099 = torch.constant.none
%cpu_7100 = torch.constant.device "cpu"
%false_7101 = torch.constant.bool false
%6073 = torch.aten.arange %int131072_7097, %none_7098, %none_7099, %cpu_7100, %false_7101 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7102 = torch.constant.int 0
%int128_7103 = torch.constant.int 128
%int2_7104 = torch.constant.int 2
%none_7105 = torch.constant.none
%none_7106 = torch.constant.none
%cpu_7107 = torch.constant.device "cpu"
%false_7108 = torch.constant.bool false
%6074 = torch.aten.arange.start_step %int0_7102, %int128_7103, %int2_7104, %none_7105, %none_7106, %cpu_7107, %false_7108 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7109 = torch.constant.int 0
%int0_7110 = torch.constant.int 0
%int64_7111 = torch.constant.int 64
%int1_7112 = torch.constant.int 1
%6075 = torch.aten.slice.Tensor %6074, %int0_7109, %int0_7110, %int64_7111, %int1_7112 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7113 = torch.constant.int 6
%6076 = torch.prims.convert_element_type %6075, %int6_7113 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7114 = torch.constant.int 128
%6077 = torch.aten.div.Scalar %6076, %int128_7114 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7115 = torch.constant.float 5.000000e+05
%6078 = torch.aten.pow.Scalar %float5.000000e05_7115, %6077 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6079 = torch.aten.reciprocal %6078 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7116 = torch.constant.float 1.000000e+00
%6080 = torch.aten.mul.Scalar %6079, %float1.000000e00_7116 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7117 = torch.constant.int 131072
%int1_7118 = torch.constant.int 1
%6081 = torch.prim.ListConstruct %int131072_7117, %int1_7118 : (!torch.int, !torch.int) -> !torch.list<int>
%6082 = torch.aten.view %6073, %6081 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6083 = torch.aten.mul.Tensor %6082, %6080 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6084 = torch.aten.cos %6083 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6085 = torch.aten.sin %6083 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6086 = torch.aten.complex %6084, %6085 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7119 = torch.constant.int 1
%6087 = torch.aten.size.int %6052, %int1_7119 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_7120 = torch.constant.int 0
%6088 = torch.aten.add.int %int0_7120, %6087 : !torch.int, !torch.int -> !torch.int
%int0_7121 = torch.constant.int 0
%int0_7122 = torch.constant.int 0
%int1_7123 = torch.constant.int 1
%6089 = torch.aten.slice.Tensor %6086, %int0_7121, %int0_7122, %6088, %int1_7123 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6089, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7124 = torch.constant.int 1
%int0_7125 = torch.constant.int 0
%int9223372036854775807_7126 = torch.constant.int 9223372036854775807
%int1_7127 = torch.constant.int 1
%6090 = torch.aten.slice.Tensor %6089, %int1_7124, %int0_7125, %int9223372036854775807_7126, %int1_7127 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6090, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7128 = torch.constant.int 0
%6091 = torch.aten.unsqueeze %6090, %int0_7128 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6091, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7129 = torch.constant.int 2
%6092 = torch.aten.unsqueeze %6091, %int2_7129 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6092, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7130 = torch.constant.int 3
%int0_7131 = torch.constant.int 0
%int9223372036854775807_7132 = torch.constant.int 9223372036854775807
%int1_7133 = torch.constant.int 1
%6093 = torch.aten.slice.Tensor %6092, %int3_7130, %int0_7131, %int9223372036854775807_7132, %int1_7133 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6093, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6094 = torch_c.to_builtin_tensor %6068 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_7134 = arith.constant 1 : index
%dim_7135 = tensor.dim %6094, %c1_7134 : tensor<4x?x32x128xf16>
%6095 = flow.tensor.bitcast %6094 : tensor<4x?x32x128xf16>{%dim_7135} -> tensor<4x?x32x64xcomplex<f16>>{%dim_7135}
%6096 = torch_c.from_builtin_tensor %6095 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %6096, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%6097 = torch.aten.mul.Tensor %6096, %6093 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %6097, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%6098 = torch_c.to_builtin_tensor %6097 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_7136 = arith.constant 1 : index
%dim_7137 = tensor.dim %6098, %c1_7136 : tensor<4x?x32x64xcomplex<f32>>
%6099 = flow.tensor.bitcast %6098 : tensor<4x?x32x64xcomplex<f32>>{%dim_7137} -> tensor<4x?x32x128xf32>{%dim_7137}
%6100 = torch_c.from_builtin_tensor %6099 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %6100, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_7138 = torch.constant.int 5
%6101 = torch.prims.convert_element_type %6100, %int5_7138 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6101, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_7139 = torch.constant.int 131072
%none_7140 = torch.constant.none
%none_7141 = torch.constant.none
%cpu_7142 = torch.constant.device "cpu"
%false_7143 = torch.constant.bool false
%6102 = torch.aten.arange %int131072_7139, %none_7140, %none_7141, %cpu_7142, %false_7143 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7144 = torch.constant.int 0
%int128_7145 = torch.constant.int 128
%int2_7146 = torch.constant.int 2
%none_7147 = torch.constant.none
%none_7148 = torch.constant.none
%cpu_7149 = torch.constant.device "cpu"
%false_7150 = torch.constant.bool false
%6103 = torch.aten.arange.start_step %int0_7144, %int128_7145, %int2_7146, %none_7147, %none_7148, %cpu_7149, %false_7150 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7151 = torch.constant.int 0
%int0_7152 = torch.constant.int 0
%int64_7153 = torch.constant.int 64
%int1_7154 = torch.constant.int 1
%6104 = torch.aten.slice.Tensor %6103, %int0_7151, %int0_7152, %int64_7153, %int1_7154 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7155 = torch.constant.int 6
%6105 = torch.prims.convert_element_type %6104, %int6_7155 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7156 = torch.constant.int 128
%6106 = torch.aten.div.Scalar %6105, %int128_7156 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7157 = torch.constant.float 5.000000e+05
%6107 = torch.aten.pow.Scalar %float5.000000e05_7157, %6106 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6108 = torch.aten.reciprocal %6107 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7158 = torch.constant.float 1.000000e+00
%6109 = torch.aten.mul.Scalar %6108, %float1.000000e00_7158 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7159 = torch.constant.int 131072
%int1_7160 = torch.constant.int 1
%6110 = torch.prim.ListConstruct %int131072_7159, %int1_7160 : (!torch.int, !torch.int) -> !torch.list<int>
%6111 = torch.aten.view %6102, %6110 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6112 = torch.aten.mul.Tensor %6111, %6109 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6113 = torch.aten.cos %6112 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6114 = torch.aten.sin %6112 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6115 = torch.aten.complex %6113, %6114 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7161 = torch.constant.int 1
%6116 = torch.aten.size.int %6059, %int1_7161 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_7162 = torch.constant.int 0
%6117 = torch.aten.add.int %int0_7162, %6116 : !torch.int, !torch.int -> !torch.int
%int0_7163 = torch.constant.int 0
%int0_7164 = torch.constant.int 0
%int1_7165 = torch.constant.int 1
%6118 = torch.aten.slice.Tensor %6115, %int0_7163, %int0_7164, %6117, %int1_7165 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6118, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7166 = torch.constant.int 1
%int0_7167 = torch.constant.int 0
%int9223372036854775807_7168 = torch.constant.int 9223372036854775807
%int1_7169 = torch.constant.int 1
%6119 = torch.aten.slice.Tensor %6118, %int1_7166, %int0_7167, %int9223372036854775807_7168, %int1_7169 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6119, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7170 = torch.constant.int 0
%6120 = torch.aten.unsqueeze %6119, %int0_7170 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6120, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7171 = torch.constant.int 2
%6121 = torch.aten.unsqueeze %6120, %int2_7171 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6121, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7172 = torch.constant.int 3
%int0_7173 = torch.constant.int 0
%int9223372036854775807_7174 = torch.constant.int 9223372036854775807
%int1_7175 = torch.constant.int 1
%6122 = torch.aten.slice.Tensor %6121, %int3_7172, %int0_7173, %int9223372036854775807_7174, %int1_7175 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6122, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6123 = torch_c.to_builtin_tensor %6070 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_7176 = arith.constant 1 : index
%dim_7177 = tensor.dim %6123, %c1_7176 : tensor<4x?x8x128xf16>
%6124 = flow.tensor.bitcast %6123 : tensor<4x?x8x128xf16>{%dim_7177} -> tensor<4x?x8x64xcomplex<f16>>{%dim_7177}
%6125 = torch_c.from_builtin_tensor %6124 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %6125, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%6126 = torch.aten.mul.Tensor %6125, %6122 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %6126, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%6127 = torch_c.to_builtin_tensor %6126 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_7178 = arith.constant 1 : index
%dim_7179 = tensor.dim %6127, %c1_7178 : tensor<4x?x8x64xcomplex<f32>>
%6128 = flow.tensor.bitcast %6127 : tensor<4x?x8x64xcomplex<f32>>{%dim_7179} -> tensor<4x?x8x128xf32>{%dim_7179}
%6129 = torch_c.from_builtin_tensor %6128 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %6129, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_7180 = torch.constant.int 5
%6130 = torch.prims.convert_element_type %6129, %int5_7180 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6130, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_7181 = torch.constant.int 64
%6131 = torch.aten.mul.Scalar %arg2, %int64_7181 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6131, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int58 = torch.constant.int 58
%int1_7182 = torch.constant.int 1
%6132 = torch.aten.add.Scalar %6131, %int58, %int1_7182 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6132, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7183 = torch.constant.int 4
%int32_7184 = torch.constant.int 32
%int8_7185 = torch.constant.int 8
%int128_7186 = torch.constant.int 128
%6133 = torch.prim.ListConstruct %int4_7183, %425, %int32_7184, %int8_7185, %int128_7186 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6134 = torch.aten.view %6130, %6133 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6134, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7187 = torch.constant.int 4
%6135 = torch.aten.mul.int %int4_7187, %425 : !torch.int, !torch.int -> !torch.int
%int32_7188 = torch.constant.int 32
%int8_7189 = torch.constant.int 8
%int128_7190 = torch.constant.int 128
%6136 = torch.prim.ListConstruct %6135, %int32_7188, %int8_7189, %int128_7190 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6137 = torch.aten.view %6134, %6136 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6137, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7191 = torch.constant.int 4
%6138 = torch.aten.mul.int %int4_7191, %425 : !torch.int, !torch.int -> !torch.int
%6139 = torch.prim.ListConstruct %6138 : (!torch.int) -> !torch.list<int>
%6140 = torch.aten.view %6132, %6139 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6140, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_7192 = torch.constant.int 32
%int2_7193 = torch.constant.int 2
%int32_7194 = torch.constant.int 32
%int8_7195 = torch.constant.int 8
%int128_7196 = torch.constant.int 128
%6141 = torch.prim.ListConstruct %416, %int32_7192, %int2_7193, %int32_7194, %int8_7195, %int128_7196 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6142 = torch.aten.view %5974, %6141 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6142, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7197 = torch.constant.int 32
%6143 = torch.aten.mul.int %416, %int32_7197 : !torch.int, !torch.int -> !torch.int
%int2_7198 = torch.constant.int 2
%6144 = torch.aten.mul.int %6143, %int2_7198 : !torch.int, !torch.int -> !torch.int
%int32_7199 = torch.constant.int 32
%int8_7200 = torch.constant.int 8
%int128_7201 = torch.constant.int 128
%6145 = torch.prim.ListConstruct %6144, %int32_7199, %int8_7200, %int128_7201 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6146 = torch.aten.view %6142, %6145 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6146, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%6147 = torch.prim.ListConstruct %6140 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7202 = torch.constant.bool false
%6148 = torch.aten.index_put %6146, %6147, %6137, %false_7202 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6148, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7203 = torch.constant.int 32
%int2_7204 = torch.constant.int 2
%int32_7205 = torch.constant.int 32
%int8_7206 = torch.constant.int 8
%int128_7207 = torch.constant.int 128
%6149 = torch.prim.ListConstruct %416, %int32_7203, %int2_7204, %int32_7205, %int8_7206, %int128_7207 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6150 = torch.aten.view %6148, %6149 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6150, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7208 = torch.constant.int 2097152
%6151 = torch.prim.ListConstruct %416, %int2097152_7208 : (!torch.int, !torch.int) -> !torch.list<int>
%6152 = torch.aten.view %6150, %6151 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %6152, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_7209 = torch.constant.int 32
%int2_7210 = torch.constant.int 2
%int32_7211 = torch.constant.int 32
%int8_7212 = torch.constant.int 8
%int128_7213 = torch.constant.int 128
%6153 = torch.prim.ListConstruct %416, %int32_7209, %int2_7210, %int32_7211, %int8_7212, %int128_7213 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6154 = torch.aten.view %6152, %6153 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6154, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7214 = torch.constant.int 32
%int8_7215 = torch.constant.int 8
%int128_7216 = torch.constant.int 128
%6155 = torch.prim.ListConstruct %6144, %int32_7214, %int8_7215, %int128_7216 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6156 = torch.aten.view %6154, %6155 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6156, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7217 = torch.constant.int 4
%int32_7218 = torch.constant.int 32
%int8_7219 = torch.constant.int 8
%int128_7220 = torch.constant.int 128
%6157 = torch.prim.ListConstruct %int4_7217, %425, %int32_7218, %int8_7219, %int128_7220 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6158 = torch.aten.view %6072, %6157 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6158, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7221 = torch.constant.int 4
%6159 = torch.aten.mul.int %int4_7221, %425 : !torch.int, !torch.int -> !torch.int
%int32_7222 = torch.constant.int 32
%int8_7223 = torch.constant.int 8
%int128_7224 = torch.constant.int 128
%6160 = torch.prim.ListConstruct %6159, %int32_7222, %int8_7223, %int128_7224 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6161 = torch.aten.view %6158, %6160 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6161, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_7225 = torch.constant.int 1
%int1_7226 = torch.constant.int 1
%6162 = torch.aten.add.Scalar %6132, %int1_7225, %int1_7226 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6162, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7227 = torch.constant.int 4
%6163 = torch.aten.mul.int %int4_7227, %425 : !torch.int, !torch.int -> !torch.int
%6164 = torch.prim.ListConstruct %6163 : (!torch.int) -> !torch.list<int>
%6165 = torch.aten.view %6162, %6164 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6165, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%6166 = torch.prim.ListConstruct %6165 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7228 = torch.constant.bool false
%6167 = torch.aten.index_put %6156, %6166, %6161, %false_7228 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6167, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7229 = torch.constant.int 32
%int2_7230 = torch.constant.int 2
%int32_7231 = torch.constant.int 32
%int8_7232 = torch.constant.int 8
%int128_7233 = torch.constant.int 128
%6168 = torch.prim.ListConstruct %416, %int32_7229, %int2_7230, %int32_7231, %int8_7232, %int128_7233 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6169 = torch.aten.view %6167, %6168 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6169, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7234 = torch.constant.int 2097152
%6170 = torch.prim.ListConstruct %416, %int2097152_7234 : (!torch.int, !torch.int) -> !torch.list<int>
%6171 = torch.aten.view %6169, %6170 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %6171, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_7235 = torch.constant.int -2
%6172 = torch.aten.unsqueeze %6130, %int-2_7235 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6172, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_7236 = torch.constant.int 4
%int8_7237 = torch.constant.int 8
%int4_7238 = torch.constant.int 4
%int128_7239 = torch.constant.int 128
%6173 = torch.prim.ListConstruct %int4_7236, %6116, %int8_7237, %int4_7238, %int128_7239 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7240 = torch.constant.bool false
%6174 = torch.aten.expand %6172, %6173, %false_7240 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6174, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7241 = torch.constant.int 0
%6175 = torch.aten.clone %6174, %int0_7241 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6175, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7242 = torch.constant.int 4
%int32_7243 = torch.constant.int 32
%int128_7244 = torch.constant.int 128
%6176 = torch.prim.ListConstruct %int4_7242, %6116, %int32_7243, %int128_7244 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6177 = torch.aten._unsafe_view %6175, %6176 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6177, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_7245 = torch.constant.int -2
%6178 = torch.aten.unsqueeze %6072, %int-2_7245 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6178, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_7246 = torch.constant.int 1
%6179 = torch.aten.size.int %6066, %int1_7246 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_7247 = torch.constant.int 4
%int8_7248 = torch.constant.int 8
%int4_7249 = torch.constant.int 4
%int128_7250 = torch.constant.int 128
%6180 = torch.prim.ListConstruct %int4_7247, %6179, %int8_7248, %int4_7249, %int128_7250 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7251 = torch.constant.bool false
%6181 = torch.aten.expand %6178, %6180, %false_7251 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6181, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7252 = torch.constant.int 0
%6182 = torch.aten.clone %6181, %int0_7252 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6182, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7253 = torch.constant.int 4
%int32_7254 = torch.constant.int 32
%int128_7255 = torch.constant.int 128
%6183 = torch.prim.ListConstruct %int4_7253, %6179, %int32_7254, %int128_7255 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6184 = torch.aten._unsafe_view %6182, %6183 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6184, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_7256 = torch.constant.int 1
%int2_7257 = torch.constant.int 2
%6185 = torch.aten.transpose.int %6101, %int1_7256, %int2_7257 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6185, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7258 = torch.constant.int 1
%int2_7259 = torch.constant.int 2
%6186 = torch.aten.transpose.int %6177, %int1_7258, %int2_7259 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6186, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7260 = torch.constant.int 1
%int2_7261 = torch.constant.int 2
%6187 = torch.aten.transpose.int %6184, %int1_7260, %int2_7261 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6187, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_7262 = torch.constant.float 0.000000e+00
%false_7263 = torch.constant.bool false
%none_7264 = torch.constant.none
%6188:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%6185, %6186, %6187, %float0.000000e00_7262, %false_7263, %320, %none_7264) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %6188#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7265 = torch.constant.int 1
%int2_7266 = torch.constant.int 2
%6189 = torch.aten.transpose.int %6188#0, %int1_7265, %int2_7266 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6189, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7267 = torch.constant.int 4
%int4096_7268 = torch.constant.int 4096
%6190 = torch.prim.ListConstruct %int4_7267, %6087, %int4096_7268 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6191 = torch.aten.view %6189, %6190 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6191, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7269 = torch.constant.int -2
%int-1_7270 = torch.constant.int -1
%6192 = torch.aten.transpose.int %266, %int-2_7269, %int-1_7270 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7271 = torch.constant.int 4
%6193 = torch.aten.mul.int %int4_7271, %6087 : !torch.int, !torch.int -> !torch.int
%int4096_7272 = torch.constant.int 4096
%6194 = torch.prim.ListConstruct %6193, %int4096_7272 : (!torch.int, !torch.int) -> !torch.list<int>
%6195 = torch.aten.view %6191, %6194 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6195, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6196 = torch.aten.mm %6195, %6192 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6196, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7273 = torch.constant.int 4
%int4096_7274 = torch.constant.int 4096
%6197 = torch.prim.ListConstruct %int4_7273, %6087, %int4096_7274 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6198 = torch.aten.view %6196, %6197 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6198, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7275 = torch.constant.int 1
%6199 = torch.aten.add.Tensor %6036, %6198, %int1_7275 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6199, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7276 = torch.constant.int 6
%6200 = torch.prims.convert_element_type %6199, %int6_7276 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6200, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7277 = torch.constant.int 2
%6201 = torch.aten.pow.Tensor_Scalar %6200, %int2_7277 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6201, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7278 = torch.constant.int -1
%6202 = torch.prim.ListConstruct %int-1_7278 : (!torch.int) -> !torch.list<int>
%true_7279 = torch.constant.bool true
%none_7280 = torch.constant.none
%6203 = torch.aten.mean.dim %6201, %6202, %true_7279, %none_7280 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6203, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7281 = torch.constant.float 9.9999997473787516E-6
%int1_7282 = torch.constant.int 1
%6204 = torch.aten.add.Scalar %6203, %float9.999990e-06_7281, %int1_7282 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6204, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6205 = torch.aten.rsqrt %6204 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6205, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6206 = torch.aten.mul.Tensor %6200, %6205 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6206, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6207 = torch.aten.mul.Tensor %267, %6206 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6207, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7283 = torch.constant.int 5
%6208 = torch.prims.convert_element_type %6207, %int5_7283 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6208, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7284 = torch.constant.int -2
%int-1_7285 = torch.constant.int -1
%6209 = torch.aten.transpose.int %268, %int-2_7284, %int-1_7285 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7286 = torch.constant.int 4
%6210 = torch.aten.mul.int %int4_7286, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7287 = torch.constant.int 4096
%6211 = torch.prim.ListConstruct %6210, %int4096_7287 : (!torch.int, !torch.int) -> !torch.list<int>
%6212 = torch.aten.view %6208, %6211 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6212, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6213 = torch.aten.mm %6212, %6209 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6213, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7288 = torch.constant.int 4
%int14336_7289 = torch.constant.int 14336
%6214 = torch.prim.ListConstruct %int4_7288, %294, %int14336_7289 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6215 = torch.aten.view %6213, %6214 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6215, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6216 = torch.aten.silu %6215 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6216, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7290 = torch.constant.int -2
%int-1_7291 = torch.constant.int -1
%6217 = torch.aten.transpose.int %269, %int-2_7290, %int-1_7291 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7292 = torch.constant.int 4
%6218 = torch.aten.mul.int %int4_7292, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7293 = torch.constant.int 4096
%6219 = torch.prim.ListConstruct %6218, %int4096_7293 : (!torch.int, !torch.int) -> !torch.list<int>
%6220 = torch.aten.view %6208, %6219 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6220, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6221 = torch.aten.mm %6220, %6217 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6221, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7294 = torch.constant.int 4
%int14336_7295 = torch.constant.int 14336
%6222 = torch.prim.ListConstruct %int4_7294, %294, %int14336_7295 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6223 = torch.aten.view %6221, %6222 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6223, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6224 = torch.aten.mul.Tensor %6216, %6223 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6224, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7296 = torch.constant.int -2
%int-1_7297 = torch.constant.int -1
%6225 = torch.aten.transpose.int %270, %int-2_7296, %int-1_7297 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_7298 = torch.constant.int 1
%6226 = torch.aten.size.int %6215, %int1_7298 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_7299 = torch.constant.int 4
%6227 = torch.aten.mul.int %int4_7299, %6226 : !torch.int, !torch.int -> !torch.int
%int14336_7300 = torch.constant.int 14336
%6228 = torch.prim.ListConstruct %6227, %int14336_7300 : (!torch.int, !torch.int) -> !torch.list<int>
%6229 = torch.aten.view %6224, %6228 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6229, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%6230 = torch.aten.mm %6229, %6225 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6230, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7301 = torch.constant.int 4
%int4096_7302 = torch.constant.int 4096
%6231 = torch.prim.ListConstruct %int4_7301, %6226, %int4096_7302 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6232 = torch.aten.view %6230, %6231 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6232, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7303 = torch.constant.int 1
%6233 = torch.aten.add.Tensor %6199, %6232, %int1_7303 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6233, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7304 = torch.constant.int 6
%6234 = torch.prims.convert_element_type %6233, %int6_7304 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6234, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7305 = torch.constant.int 2
%6235 = torch.aten.pow.Tensor_Scalar %6234, %int2_7305 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6235, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7306 = torch.constant.int -1
%6236 = torch.prim.ListConstruct %int-1_7306 : (!torch.int) -> !torch.list<int>
%true_7307 = torch.constant.bool true
%none_7308 = torch.constant.none
%6237 = torch.aten.mean.dim %6235, %6236, %true_7307, %none_7308 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6237, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7309 = torch.constant.float 9.9999997473787516E-6
%int1_7310 = torch.constant.int 1
%6238 = torch.aten.add.Scalar %6237, %float9.999990e-06_7309, %int1_7310 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6238, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6239 = torch.aten.rsqrt %6238 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6239, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6240 = torch.aten.mul.Tensor %6234, %6239 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6240, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6241 = torch.aten.mul.Tensor %271, %6240 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6241, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7311 = torch.constant.int 5
%6242 = torch.prims.convert_element_type %6241, %int5_7311 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6242, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7312 = torch.constant.int -2
%int-1_7313 = torch.constant.int -1
%6243 = torch.aten.transpose.int %272, %int-2_7312, %int-1_7313 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7314 = torch.constant.int 4
%6244 = torch.aten.mul.int %int4_7314, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7315 = torch.constant.int 4096
%6245 = torch.prim.ListConstruct %6244, %int4096_7315 : (!torch.int, !torch.int) -> !torch.list<int>
%6246 = torch.aten.view %6242, %6245 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6246, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6247 = torch.aten.mm %6246, %6243 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6247, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7316 = torch.constant.int 4
%int4096_7317 = torch.constant.int 4096
%6248 = torch.prim.ListConstruct %int4_7316, %294, %int4096_7317 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6249 = torch.aten.view %6247, %6248 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6249, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7318 = torch.constant.int -2
%int-1_7319 = torch.constant.int -1
%6250 = torch.aten.transpose.int %273, %int-2_7318, %int-1_7319 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7320 = torch.constant.int 4
%6251 = torch.aten.mul.int %int4_7320, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7321 = torch.constant.int 4096
%6252 = torch.prim.ListConstruct %6251, %int4096_7321 : (!torch.int, !torch.int) -> !torch.list<int>
%6253 = torch.aten.view %6242, %6252 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6253, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6254 = torch.aten.mm %6253, %6250 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6254, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7322 = torch.constant.int 4
%int1024_7323 = torch.constant.int 1024
%6255 = torch.prim.ListConstruct %int4_7322, %294, %int1024_7323 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6256 = torch.aten.view %6254, %6255 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6256, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_7324 = torch.constant.int -2
%int-1_7325 = torch.constant.int -1
%6257 = torch.aten.transpose.int %274, %int-2_7324, %int-1_7325 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7326 = torch.constant.int 4
%6258 = torch.aten.mul.int %int4_7326, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7327 = torch.constant.int 4096
%6259 = torch.prim.ListConstruct %6258, %int4096_7327 : (!torch.int, !torch.int) -> !torch.list<int>
%6260 = torch.aten.view %6242, %6259 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6260, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6261 = torch.aten.mm %6260, %6257 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6261, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7328 = torch.constant.int 4
%int1024_7329 = torch.constant.int 1024
%6262 = torch.prim.ListConstruct %int4_7328, %294, %int1024_7329 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6263 = torch.aten.view %6261, %6262 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6263, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_7330 = torch.constant.int 4
%int32_7331 = torch.constant.int 32
%int128_7332 = torch.constant.int 128
%6264 = torch.prim.ListConstruct %int4_7330, %294, %int32_7331, %int128_7332 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6265 = torch.aten.view %6249, %6264 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6265, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7333 = torch.constant.int 4
%int8_7334 = torch.constant.int 8
%int128_7335 = torch.constant.int 128
%6266 = torch.prim.ListConstruct %int4_7333, %294, %int8_7334, %int128_7335 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6267 = torch.aten.view %6256, %6266 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6267, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_7336 = torch.constant.int 4
%int8_7337 = torch.constant.int 8
%int128_7338 = torch.constant.int 128
%6268 = torch.prim.ListConstruct %int4_7336, %294, %int8_7337, %int128_7338 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6269 = torch.aten.view %6263, %6268 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6269, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_7339 = torch.constant.int 131072
%none_7340 = torch.constant.none
%none_7341 = torch.constant.none
%cpu_7342 = torch.constant.device "cpu"
%false_7343 = torch.constant.bool false
%6270 = torch.aten.arange %int131072_7339, %none_7340, %none_7341, %cpu_7342, %false_7343 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7344 = torch.constant.int 0
%int128_7345 = torch.constant.int 128
%int2_7346 = torch.constant.int 2
%none_7347 = torch.constant.none
%none_7348 = torch.constant.none
%cpu_7349 = torch.constant.device "cpu"
%false_7350 = torch.constant.bool false
%6271 = torch.aten.arange.start_step %int0_7344, %int128_7345, %int2_7346, %none_7347, %none_7348, %cpu_7349, %false_7350 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7351 = torch.constant.int 0
%int0_7352 = torch.constant.int 0
%int64_7353 = torch.constant.int 64
%int1_7354 = torch.constant.int 1
%6272 = torch.aten.slice.Tensor %6271, %int0_7351, %int0_7352, %int64_7353, %int1_7354 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7355 = torch.constant.int 6
%6273 = torch.prims.convert_element_type %6272, %int6_7355 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7356 = torch.constant.int 128
%6274 = torch.aten.div.Scalar %6273, %int128_7356 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7357 = torch.constant.float 5.000000e+05
%6275 = torch.aten.pow.Scalar %float5.000000e05_7357, %6274 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6276 = torch.aten.reciprocal %6275 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7358 = torch.constant.float 1.000000e+00
%6277 = torch.aten.mul.Scalar %6276, %float1.000000e00_7358 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7359 = torch.constant.int 131072
%int1_7360 = torch.constant.int 1
%6278 = torch.prim.ListConstruct %int131072_7359, %int1_7360 : (!torch.int, !torch.int) -> !torch.list<int>
%6279 = torch.aten.view %6270, %6278 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6280 = torch.aten.mul.Tensor %6279, %6277 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6281 = torch.aten.cos %6280 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6282 = torch.aten.sin %6280 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6283 = torch.aten.complex %6281, %6282 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7361 = torch.constant.int 1
%6284 = torch.aten.size.int %6249, %int1_7361 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_7362 = torch.constant.int 0
%6285 = torch.aten.add.int %int0_7362, %6284 : !torch.int, !torch.int -> !torch.int
%int0_7363 = torch.constant.int 0
%int0_7364 = torch.constant.int 0
%int1_7365 = torch.constant.int 1
%6286 = torch.aten.slice.Tensor %6283, %int0_7363, %int0_7364, %6285, %int1_7365 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6286, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7366 = torch.constant.int 1
%int0_7367 = torch.constant.int 0
%int9223372036854775807_7368 = torch.constant.int 9223372036854775807
%int1_7369 = torch.constant.int 1
%6287 = torch.aten.slice.Tensor %6286, %int1_7366, %int0_7367, %int9223372036854775807_7368, %int1_7369 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6287, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7370 = torch.constant.int 0
%6288 = torch.aten.unsqueeze %6287, %int0_7370 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6288, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7371 = torch.constant.int 2
%6289 = torch.aten.unsqueeze %6288, %int2_7371 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6289, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7372 = torch.constant.int 3
%int0_7373 = torch.constant.int 0
%int9223372036854775807_7374 = torch.constant.int 9223372036854775807
%int1_7375 = torch.constant.int 1
%6290 = torch.aten.slice.Tensor %6289, %int3_7372, %int0_7373, %int9223372036854775807_7374, %int1_7375 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6290, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6291 = torch_c.to_builtin_tensor %6265 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_7376 = arith.constant 1 : index
%dim_7377 = tensor.dim %6291, %c1_7376 : tensor<4x?x32x128xf16>
%6292 = flow.tensor.bitcast %6291 : tensor<4x?x32x128xf16>{%dim_7377} -> tensor<4x?x32x64xcomplex<f16>>{%dim_7377}
%6293 = torch_c.from_builtin_tensor %6292 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %6293, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%6294 = torch.aten.mul.Tensor %6293, %6290 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %6294, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%6295 = torch_c.to_builtin_tensor %6294 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_7378 = arith.constant 1 : index
%dim_7379 = tensor.dim %6295, %c1_7378 : tensor<4x?x32x64xcomplex<f32>>
%6296 = flow.tensor.bitcast %6295 : tensor<4x?x32x64xcomplex<f32>>{%dim_7379} -> tensor<4x?x32x128xf32>{%dim_7379}
%6297 = torch_c.from_builtin_tensor %6296 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %6297, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_7380 = torch.constant.int 5
%6298 = torch.prims.convert_element_type %6297, %int5_7380 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6298, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_7381 = torch.constant.int 131072
%none_7382 = torch.constant.none
%none_7383 = torch.constant.none
%cpu_7384 = torch.constant.device "cpu"
%false_7385 = torch.constant.bool false
%6299 = torch.aten.arange %int131072_7381, %none_7382, %none_7383, %cpu_7384, %false_7385 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7386 = torch.constant.int 0
%int128_7387 = torch.constant.int 128
%int2_7388 = torch.constant.int 2
%none_7389 = torch.constant.none
%none_7390 = torch.constant.none
%cpu_7391 = torch.constant.device "cpu"
%false_7392 = torch.constant.bool false
%6300 = torch.aten.arange.start_step %int0_7386, %int128_7387, %int2_7388, %none_7389, %none_7390, %cpu_7391, %false_7392 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7393 = torch.constant.int 0
%int0_7394 = torch.constant.int 0
%int64_7395 = torch.constant.int 64
%int1_7396 = torch.constant.int 1
%6301 = torch.aten.slice.Tensor %6300, %int0_7393, %int0_7394, %int64_7395, %int1_7396 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7397 = torch.constant.int 6
%6302 = torch.prims.convert_element_type %6301, %int6_7397 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7398 = torch.constant.int 128
%6303 = torch.aten.div.Scalar %6302, %int128_7398 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7399 = torch.constant.float 5.000000e+05
%6304 = torch.aten.pow.Scalar %float5.000000e05_7399, %6303 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6305 = torch.aten.reciprocal %6304 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7400 = torch.constant.float 1.000000e+00
%6306 = torch.aten.mul.Scalar %6305, %float1.000000e00_7400 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7401 = torch.constant.int 131072
%int1_7402 = torch.constant.int 1
%6307 = torch.prim.ListConstruct %int131072_7401, %int1_7402 : (!torch.int, !torch.int) -> !torch.list<int>
%6308 = torch.aten.view %6299, %6307 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6309 = torch.aten.mul.Tensor %6308, %6306 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6310 = torch.aten.cos %6309 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6311 = torch.aten.sin %6309 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6312 = torch.aten.complex %6310, %6311 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7403 = torch.constant.int 1
%6313 = torch.aten.size.int %6256, %int1_7403 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_7404 = torch.constant.int 0
%6314 = torch.aten.add.int %int0_7404, %6313 : !torch.int, !torch.int -> !torch.int
%int0_7405 = torch.constant.int 0
%int0_7406 = torch.constant.int 0
%int1_7407 = torch.constant.int 1
%6315 = torch.aten.slice.Tensor %6312, %int0_7405, %int0_7406, %6314, %int1_7407 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6315, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7408 = torch.constant.int 1
%int0_7409 = torch.constant.int 0
%int9223372036854775807_7410 = torch.constant.int 9223372036854775807
%int1_7411 = torch.constant.int 1
%6316 = torch.aten.slice.Tensor %6315, %int1_7408, %int0_7409, %int9223372036854775807_7410, %int1_7411 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6316, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7412 = torch.constant.int 0
%6317 = torch.aten.unsqueeze %6316, %int0_7412 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6317, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7413 = torch.constant.int 2
%6318 = torch.aten.unsqueeze %6317, %int2_7413 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6318, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7414 = torch.constant.int 3
%int0_7415 = torch.constant.int 0
%int9223372036854775807_7416 = torch.constant.int 9223372036854775807
%int1_7417 = torch.constant.int 1
%6319 = torch.aten.slice.Tensor %6318, %int3_7414, %int0_7415, %int9223372036854775807_7416, %int1_7417 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6319, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6320 = torch_c.to_builtin_tensor %6267 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_7418 = arith.constant 1 : index
%dim_7419 = tensor.dim %6320, %c1_7418 : tensor<4x?x8x128xf16>
%6321 = flow.tensor.bitcast %6320 : tensor<4x?x8x128xf16>{%dim_7419} -> tensor<4x?x8x64xcomplex<f16>>{%dim_7419}
%6322 = torch_c.from_builtin_tensor %6321 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %6322, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%6323 = torch.aten.mul.Tensor %6322, %6319 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %6323, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%6324 = torch_c.to_builtin_tensor %6323 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_7420 = arith.constant 1 : index
%dim_7421 = tensor.dim %6324, %c1_7420 : tensor<4x?x8x64xcomplex<f32>>
%6325 = flow.tensor.bitcast %6324 : tensor<4x?x8x64xcomplex<f32>>{%dim_7421} -> tensor<4x?x8x128xf32>{%dim_7421}
%6326 = torch_c.from_builtin_tensor %6325 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %6326, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_7422 = torch.constant.int 5
%6327 = torch.prims.convert_element_type %6326, %int5_7422 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6327, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_7423 = torch.constant.int 64
%6328 = torch.aten.mul.Scalar %arg2, %int64_7423 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6328, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int60 = torch.constant.int 60
%int1_7424 = torch.constant.int 1
%6329 = torch.aten.add.Scalar %6328, %int60, %int1_7424 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6329, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7425 = torch.constant.int 4
%int32_7426 = torch.constant.int 32
%int8_7427 = torch.constant.int 8
%int128_7428 = torch.constant.int 128
%6330 = torch.prim.ListConstruct %int4_7425, %425, %int32_7426, %int8_7427, %int128_7428 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6331 = torch.aten.view %6327, %6330 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6331, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7429 = torch.constant.int 4
%6332 = torch.aten.mul.int %int4_7429, %425 : !torch.int, !torch.int -> !torch.int
%int32_7430 = torch.constant.int 32
%int8_7431 = torch.constant.int 8
%int128_7432 = torch.constant.int 128
%6333 = torch.prim.ListConstruct %6332, %int32_7430, %int8_7431, %int128_7432 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6334 = torch.aten.view %6331, %6333 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6334, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7433 = torch.constant.int 4
%6335 = torch.aten.mul.int %int4_7433, %425 : !torch.int, !torch.int -> !torch.int
%6336 = torch.prim.ListConstruct %6335 : (!torch.int) -> !torch.list<int>
%6337 = torch.aten.view %6329, %6336 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6337, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_7434 = torch.constant.int 32
%int2_7435 = torch.constant.int 2
%int32_7436 = torch.constant.int 32
%int8_7437 = torch.constant.int 8
%int128_7438 = torch.constant.int 128
%6338 = torch.prim.ListConstruct %416, %int32_7434, %int2_7435, %int32_7436, %int8_7437, %int128_7438 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6339 = torch.aten.view %6171, %6338 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6339, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7439 = torch.constant.int 32
%6340 = torch.aten.mul.int %416, %int32_7439 : !torch.int, !torch.int -> !torch.int
%int2_7440 = torch.constant.int 2
%6341 = torch.aten.mul.int %6340, %int2_7440 : !torch.int, !torch.int -> !torch.int
%int32_7441 = torch.constant.int 32
%int8_7442 = torch.constant.int 8
%int128_7443 = torch.constant.int 128
%6342 = torch.prim.ListConstruct %6341, %int32_7441, %int8_7442, %int128_7443 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6343 = torch.aten.view %6339, %6342 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6343, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%6344 = torch.prim.ListConstruct %6337 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7444 = torch.constant.bool false
%6345 = torch.aten.index_put %6343, %6344, %6334, %false_7444 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6345, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7445 = torch.constant.int 32
%int2_7446 = torch.constant.int 2
%int32_7447 = torch.constant.int 32
%int8_7448 = torch.constant.int 8
%int128_7449 = torch.constant.int 128
%6346 = torch.prim.ListConstruct %416, %int32_7445, %int2_7446, %int32_7447, %int8_7448, %int128_7449 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6347 = torch.aten.view %6345, %6346 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6347, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7450 = torch.constant.int 2097152
%6348 = torch.prim.ListConstruct %416, %int2097152_7450 : (!torch.int, !torch.int) -> !torch.list<int>
%6349 = torch.aten.view %6347, %6348 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %6349, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_7451 = torch.constant.int 32
%int2_7452 = torch.constant.int 2
%int32_7453 = torch.constant.int 32
%int8_7454 = torch.constant.int 8
%int128_7455 = torch.constant.int 128
%6350 = torch.prim.ListConstruct %416, %int32_7451, %int2_7452, %int32_7453, %int8_7454, %int128_7455 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6351 = torch.aten.view %6349, %6350 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6351, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7456 = torch.constant.int 32
%int8_7457 = torch.constant.int 8
%int128_7458 = torch.constant.int 128
%6352 = torch.prim.ListConstruct %6341, %int32_7456, %int8_7457, %int128_7458 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6353 = torch.aten.view %6351, %6352 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6353, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7459 = torch.constant.int 4
%int32_7460 = torch.constant.int 32
%int8_7461 = torch.constant.int 8
%int128_7462 = torch.constant.int 128
%6354 = torch.prim.ListConstruct %int4_7459, %425, %int32_7460, %int8_7461, %int128_7462 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6355 = torch.aten.view %6269, %6354 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6355, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7463 = torch.constant.int 4
%6356 = torch.aten.mul.int %int4_7463, %425 : !torch.int, !torch.int -> !torch.int
%int32_7464 = torch.constant.int 32
%int8_7465 = torch.constant.int 8
%int128_7466 = torch.constant.int 128
%6357 = torch.prim.ListConstruct %6356, %int32_7464, %int8_7465, %int128_7466 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6358 = torch.aten.view %6355, %6357 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6358, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_7467 = torch.constant.int 1
%int1_7468 = torch.constant.int 1
%6359 = torch.aten.add.Scalar %6329, %int1_7467, %int1_7468 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6359, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7469 = torch.constant.int 4
%6360 = torch.aten.mul.int %int4_7469, %425 : !torch.int, !torch.int -> !torch.int
%6361 = torch.prim.ListConstruct %6360 : (!torch.int) -> !torch.list<int>
%6362 = torch.aten.view %6359, %6361 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6362, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%6363 = torch.prim.ListConstruct %6362 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7470 = torch.constant.bool false
%6364 = torch.aten.index_put %6353, %6363, %6358, %false_7470 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6364, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7471 = torch.constant.int 32
%int2_7472 = torch.constant.int 2
%int32_7473 = torch.constant.int 32
%int8_7474 = torch.constant.int 8
%int128_7475 = torch.constant.int 128
%6365 = torch.prim.ListConstruct %416, %int32_7471, %int2_7472, %int32_7473, %int8_7474, %int128_7475 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6366 = torch.aten.view %6364, %6365 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6366, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7476 = torch.constant.int 2097152
%6367 = torch.prim.ListConstruct %416, %int2097152_7476 : (!torch.int, !torch.int) -> !torch.list<int>
%6368 = torch.aten.view %6366, %6367 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %6368, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_7477 = torch.constant.int -2
%6369 = torch.aten.unsqueeze %6327, %int-2_7477 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6369, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_7478 = torch.constant.int 4
%int8_7479 = torch.constant.int 8
%int4_7480 = torch.constant.int 4
%int128_7481 = torch.constant.int 128
%6370 = torch.prim.ListConstruct %int4_7478, %6313, %int8_7479, %int4_7480, %int128_7481 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7482 = torch.constant.bool false
%6371 = torch.aten.expand %6369, %6370, %false_7482 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6371, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7483 = torch.constant.int 0
%6372 = torch.aten.clone %6371, %int0_7483 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6372, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7484 = torch.constant.int 4
%int32_7485 = torch.constant.int 32
%int128_7486 = torch.constant.int 128
%6373 = torch.prim.ListConstruct %int4_7484, %6313, %int32_7485, %int128_7486 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6374 = torch.aten._unsafe_view %6372, %6373 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6374, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_7487 = torch.constant.int -2
%6375 = torch.aten.unsqueeze %6269, %int-2_7487 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6375, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_7488 = torch.constant.int 1
%6376 = torch.aten.size.int %6263, %int1_7488 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_7489 = torch.constant.int 4
%int8_7490 = torch.constant.int 8
%int4_7491 = torch.constant.int 4
%int128_7492 = torch.constant.int 128
%6377 = torch.prim.ListConstruct %int4_7489, %6376, %int8_7490, %int4_7491, %int128_7492 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7493 = torch.constant.bool false
%6378 = torch.aten.expand %6375, %6377, %false_7493 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6378, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7494 = torch.constant.int 0
%6379 = torch.aten.clone %6378, %int0_7494 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6379, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7495 = torch.constant.int 4
%int32_7496 = torch.constant.int 32
%int128_7497 = torch.constant.int 128
%6380 = torch.prim.ListConstruct %int4_7495, %6376, %int32_7496, %int128_7497 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6381 = torch.aten._unsafe_view %6379, %6380 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6381, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_7498 = torch.constant.int 1
%int2_7499 = torch.constant.int 2
%6382 = torch.aten.transpose.int %6298, %int1_7498, %int2_7499 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6382, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7500 = torch.constant.int 1
%int2_7501 = torch.constant.int 2
%6383 = torch.aten.transpose.int %6374, %int1_7500, %int2_7501 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6383, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7502 = torch.constant.int 1
%int2_7503 = torch.constant.int 2
%6384 = torch.aten.transpose.int %6381, %int1_7502, %int2_7503 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6384, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_7504 = torch.constant.float 0.000000e+00
%false_7505 = torch.constant.bool false
%none_7506 = torch.constant.none
%6385:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%6382, %6383, %6384, %float0.000000e00_7504, %false_7505, %320, %none_7506) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %6385#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7507 = torch.constant.int 1
%int2_7508 = torch.constant.int 2
%6386 = torch.aten.transpose.int %6385#0, %int1_7507, %int2_7508 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6386, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7509 = torch.constant.int 4
%int4096_7510 = torch.constant.int 4096
%6387 = torch.prim.ListConstruct %int4_7509, %6284, %int4096_7510 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6388 = torch.aten.view %6386, %6387 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6388, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7511 = torch.constant.int -2
%int-1_7512 = torch.constant.int -1
%6389 = torch.aten.transpose.int %275, %int-2_7511, %int-1_7512 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7513 = torch.constant.int 4
%6390 = torch.aten.mul.int %int4_7513, %6284 : !torch.int, !torch.int -> !torch.int
%int4096_7514 = torch.constant.int 4096
%6391 = torch.prim.ListConstruct %6390, %int4096_7514 : (!torch.int, !torch.int) -> !torch.list<int>
%6392 = torch.aten.view %6388, %6391 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6392, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6393 = torch.aten.mm %6392, %6389 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6393, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7515 = torch.constant.int 4
%int4096_7516 = torch.constant.int 4096
%6394 = torch.prim.ListConstruct %int4_7515, %6284, %int4096_7516 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6395 = torch.aten.view %6393, %6394 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6395, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7517 = torch.constant.int 1
%6396 = torch.aten.add.Tensor %6233, %6395, %int1_7517 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6396, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7518 = torch.constant.int 6
%6397 = torch.prims.convert_element_type %6396, %int6_7518 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6397, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7519 = torch.constant.int 2
%6398 = torch.aten.pow.Tensor_Scalar %6397, %int2_7519 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6398, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7520 = torch.constant.int -1
%6399 = torch.prim.ListConstruct %int-1_7520 : (!torch.int) -> !torch.list<int>
%true_7521 = torch.constant.bool true
%none_7522 = torch.constant.none
%6400 = torch.aten.mean.dim %6398, %6399, %true_7521, %none_7522 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6400, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7523 = torch.constant.float 9.9999997473787516E-6
%int1_7524 = torch.constant.int 1
%6401 = torch.aten.add.Scalar %6400, %float9.999990e-06_7523, %int1_7524 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6401, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6402 = torch.aten.rsqrt %6401 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6402, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6403 = torch.aten.mul.Tensor %6397, %6402 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6403, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6404 = torch.aten.mul.Tensor %276, %6403 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6404, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7525 = torch.constant.int 5
%6405 = torch.prims.convert_element_type %6404, %int5_7525 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6405, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7526 = torch.constant.int -2
%int-1_7527 = torch.constant.int -1
%6406 = torch.aten.transpose.int %277, %int-2_7526, %int-1_7527 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7528 = torch.constant.int 4
%6407 = torch.aten.mul.int %int4_7528, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7529 = torch.constant.int 4096
%6408 = torch.prim.ListConstruct %6407, %int4096_7529 : (!torch.int, !torch.int) -> !torch.list<int>
%6409 = torch.aten.view %6405, %6408 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6409, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6410 = torch.aten.mm %6409, %6406 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6410, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7530 = torch.constant.int 4
%int14336_7531 = torch.constant.int 14336
%6411 = torch.prim.ListConstruct %int4_7530, %294, %int14336_7531 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6412 = torch.aten.view %6410, %6411 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6412, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6413 = torch.aten.silu %6412 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6413, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7532 = torch.constant.int -2
%int-1_7533 = torch.constant.int -1
%6414 = torch.aten.transpose.int %278, %int-2_7532, %int-1_7533 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7534 = torch.constant.int 4
%6415 = torch.aten.mul.int %int4_7534, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7535 = torch.constant.int 4096
%6416 = torch.prim.ListConstruct %6415, %int4096_7535 : (!torch.int, !torch.int) -> !torch.list<int>
%6417 = torch.aten.view %6405, %6416 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6417, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6418 = torch.aten.mm %6417, %6414 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6418, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7536 = torch.constant.int 4
%int14336_7537 = torch.constant.int 14336
%6419 = torch.prim.ListConstruct %int4_7536, %294, %int14336_7537 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6420 = torch.aten.view %6418, %6419 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6420, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6421 = torch.aten.mul.Tensor %6413, %6420 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6421, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7538 = torch.constant.int -2
%int-1_7539 = torch.constant.int -1
%6422 = torch.aten.transpose.int %279, %int-2_7538, %int-1_7539 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_7540 = torch.constant.int 1
%6423 = torch.aten.size.int %6412, %int1_7540 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_7541 = torch.constant.int 4
%6424 = torch.aten.mul.int %int4_7541, %6423 : !torch.int, !torch.int -> !torch.int
%int14336_7542 = torch.constant.int 14336
%6425 = torch.prim.ListConstruct %6424, %int14336_7542 : (!torch.int, !torch.int) -> !torch.list<int>
%6426 = torch.aten.view %6421, %6425 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6426, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%6427 = torch.aten.mm %6426, %6422 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6427, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7543 = torch.constant.int 4
%int4096_7544 = torch.constant.int 4096
%6428 = torch.prim.ListConstruct %int4_7543, %6423, %int4096_7544 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6429 = torch.aten.view %6427, %6428 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6429, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7545 = torch.constant.int 1
%6430 = torch.aten.add.Tensor %6396, %6429, %int1_7545 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6430, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7546 = torch.constant.int 6
%6431 = torch.prims.convert_element_type %6430, %int6_7546 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6431, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7547 = torch.constant.int 2
%6432 = torch.aten.pow.Tensor_Scalar %6431, %int2_7547 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6432, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7548 = torch.constant.int -1
%6433 = torch.prim.ListConstruct %int-1_7548 : (!torch.int) -> !torch.list<int>
%true_7549 = torch.constant.bool true
%none_7550 = torch.constant.none
%6434 = torch.aten.mean.dim %6432, %6433, %true_7549, %none_7550 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6434, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7551 = torch.constant.float 9.9999997473787516E-6
%int1_7552 = torch.constant.int 1
%6435 = torch.aten.add.Scalar %6434, %float9.999990e-06_7551, %int1_7552 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6435, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6436 = torch.aten.rsqrt %6435 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6436, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6437 = torch.aten.mul.Tensor %6431, %6436 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6437, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6438 = torch.aten.mul.Tensor %280, %6437 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6438, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7553 = torch.constant.int 5
%6439 = torch.prims.convert_element_type %6438, %int5_7553 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6439, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7554 = torch.constant.int -2
%int-1_7555 = torch.constant.int -1
%6440 = torch.aten.transpose.int %281, %int-2_7554, %int-1_7555 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7556 = torch.constant.int 4
%6441 = torch.aten.mul.int %int4_7556, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7557 = torch.constant.int 4096
%6442 = torch.prim.ListConstruct %6441, %int4096_7557 : (!torch.int, !torch.int) -> !torch.list<int>
%6443 = torch.aten.view %6439, %6442 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6443, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6444 = torch.aten.mm %6443, %6440 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6444, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7558 = torch.constant.int 4
%int4096_7559 = torch.constant.int 4096
%6445 = torch.prim.ListConstruct %int4_7558, %294, %int4096_7559 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6446 = torch.aten.view %6444, %6445 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6446, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7560 = torch.constant.int -2
%int-1_7561 = torch.constant.int -1
%6447 = torch.aten.transpose.int %282, %int-2_7560, %int-1_7561 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7562 = torch.constant.int 4
%6448 = torch.aten.mul.int %int4_7562, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7563 = torch.constant.int 4096
%6449 = torch.prim.ListConstruct %6448, %int4096_7563 : (!torch.int, !torch.int) -> !torch.list<int>
%6450 = torch.aten.view %6439, %6449 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6450, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6451 = torch.aten.mm %6450, %6447 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6451, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7564 = torch.constant.int 4
%int1024_7565 = torch.constant.int 1024
%6452 = torch.prim.ListConstruct %int4_7564, %294, %int1024_7565 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6453 = torch.aten.view %6451, %6452 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6453, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int-2_7566 = torch.constant.int -2
%int-1_7567 = torch.constant.int -1
%6454 = torch.aten.transpose.int %283, %int-2_7566, %int-1_7567 : !torch.vtensor<[1024,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,1024],f16>
%int4_7568 = torch.constant.int 4
%6455 = torch.aten.mul.int %int4_7568, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7569 = torch.constant.int 4096
%6456 = torch.prim.ListConstruct %6455, %int4096_7569 : (!torch.int, !torch.int) -> !torch.list<int>
%6457 = torch.aten.view %6439, %6456 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6457, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6458 = torch.aten.mm %6457, %6454 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,1024],f16> -> !torch.vtensor<[?,1024],f16>
torch.bind_symbolic_shape %6458, [%292], affine_map<()[s0] -> (s0 * 128, 1024)> : !torch.vtensor<[?,1024],f16>
%int4_7570 = torch.constant.int 4
%int1024_7571 = torch.constant.int 1024
%6459 = torch.prim.ListConstruct %int4_7570, %294, %int1024_7571 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6460 = torch.aten.view %6458, %6459 : !torch.vtensor<[?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,1024],f16>
torch.bind_symbolic_shape %6460, [%292], affine_map<()[s0] -> (4, s0 * 32, 1024)> : !torch.vtensor<[4,?,1024],f16>
%int4_7572 = torch.constant.int 4
%int32_7573 = torch.constant.int 32
%int128_7574 = torch.constant.int 128
%6461 = torch.prim.ListConstruct %int4_7572, %294, %int32_7573, %int128_7574 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6462 = torch.aten.view %6446, %6461 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6462, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7575 = torch.constant.int 4
%int8_7576 = torch.constant.int 8
%int128_7577 = torch.constant.int 128
%6463 = torch.prim.ListConstruct %int4_7575, %294, %int8_7576, %int128_7577 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6464 = torch.aten.view %6453, %6463 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6464, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int4_7578 = torch.constant.int 4
%int8_7579 = torch.constant.int 8
%int128_7580 = torch.constant.int 128
%6465 = torch.prim.ListConstruct %int4_7578, %294, %int8_7579, %int128_7580 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6466 = torch.aten.view %6460, %6465 : !torch.vtensor<[4,?,1024],f16>, !torch.list<int> -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6466, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int131072_7581 = torch.constant.int 131072
%none_7582 = torch.constant.none
%none_7583 = torch.constant.none
%cpu_7584 = torch.constant.device "cpu"
%false_7585 = torch.constant.bool false
%6467 = torch.aten.arange %int131072_7581, %none_7582, %none_7583, %cpu_7584, %false_7585 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7586 = torch.constant.int 0
%int128_7587 = torch.constant.int 128
%int2_7588 = torch.constant.int 2
%none_7589 = torch.constant.none
%none_7590 = torch.constant.none
%cpu_7591 = torch.constant.device "cpu"
%false_7592 = torch.constant.bool false
%6468 = torch.aten.arange.start_step %int0_7586, %int128_7587, %int2_7588, %none_7589, %none_7590, %cpu_7591, %false_7592 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7593 = torch.constant.int 0
%int0_7594 = torch.constant.int 0
%int64_7595 = torch.constant.int 64
%int1_7596 = torch.constant.int 1
%6469 = torch.aten.slice.Tensor %6468, %int0_7593, %int0_7594, %int64_7595, %int1_7596 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7597 = torch.constant.int 6
%6470 = torch.prims.convert_element_type %6469, %int6_7597 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7598 = torch.constant.int 128
%6471 = torch.aten.div.Scalar %6470, %int128_7598 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7599 = torch.constant.float 5.000000e+05
%6472 = torch.aten.pow.Scalar %float5.000000e05_7599, %6471 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6473 = torch.aten.reciprocal %6472 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7600 = torch.constant.float 1.000000e+00
%6474 = torch.aten.mul.Scalar %6473, %float1.000000e00_7600 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7601 = torch.constant.int 131072
%int1_7602 = torch.constant.int 1
%6475 = torch.prim.ListConstruct %int131072_7601, %int1_7602 : (!torch.int, !torch.int) -> !torch.list<int>
%6476 = torch.aten.view %6467, %6475 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6477 = torch.aten.mul.Tensor %6476, %6474 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6478 = torch.aten.cos %6477 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6479 = torch.aten.sin %6477 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6480 = torch.aten.complex %6478, %6479 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7603 = torch.constant.int 1
%6481 = torch.aten.size.int %6446, %int1_7603 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.int
%int0_7604 = torch.constant.int 0
%6482 = torch.aten.add.int %int0_7604, %6481 : !torch.int, !torch.int -> !torch.int
%int0_7605 = torch.constant.int 0
%int0_7606 = torch.constant.int 0
%int1_7607 = torch.constant.int 1
%6483 = torch.aten.slice.Tensor %6480, %int0_7605, %int0_7606, %6482, %int1_7607 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6483, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7608 = torch.constant.int 1
%int0_7609 = torch.constant.int 0
%int9223372036854775807_7610 = torch.constant.int 9223372036854775807
%int1_7611 = torch.constant.int 1
%6484 = torch.aten.slice.Tensor %6483, %int1_7608, %int0_7609, %int9223372036854775807_7610, %int1_7611 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6484, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7612 = torch.constant.int 0
%6485 = torch.aten.unsqueeze %6484, %int0_7612 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6485, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7613 = torch.constant.int 2
%6486 = torch.aten.unsqueeze %6485, %int2_7613 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6486, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7614 = torch.constant.int 3
%int0_7615 = torch.constant.int 0
%int9223372036854775807_7616 = torch.constant.int 9223372036854775807
%int1_7617 = torch.constant.int 1
%6487 = torch.aten.slice.Tensor %6486, %int3_7614, %int0_7615, %int9223372036854775807_7616, %int1_7617 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6487, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6488 = torch_c.to_builtin_tensor %6462 : !torch.vtensor<[4,?,32,128],f16> -> tensor<4x?x32x128xf16>
%c1_7618 = arith.constant 1 : index
%dim_7619 = tensor.dim %6488, %c1_7618 : tensor<4x?x32x128xf16>
%6489 = flow.tensor.bitcast %6488 : tensor<4x?x32x128xf16>{%dim_7619} -> tensor<4x?x32x64xcomplex<f16>>{%dim_7619}
%6490 = torch_c.from_builtin_tensor %6489 : tensor<4x?x32x64xcomplex<f16>> -> !torch.vtensor<[4,?,32,64],complex<f16>>
torch.bind_symbolic_shape %6490, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f16>>
%6491 = torch.aten.mul.Tensor %6490, %6487 : !torch.vtensor<[4,?,32,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,32,64],complex<f32>>
torch.bind_symbolic_shape %6491, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 64)> : !torch.vtensor<[4,?,32,64],complex<f32>>
%6492 = torch_c.to_builtin_tensor %6491 : !torch.vtensor<[4,?,32,64],complex<f32>> -> tensor<4x?x32x64xcomplex<f32>>
%c1_7620 = arith.constant 1 : index
%dim_7621 = tensor.dim %6492, %c1_7620 : tensor<4x?x32x64xcomplex<f32>>
%6493 = flow.tensor.bitcast %6492 : tensor<4x?x32x64xcomplex<f32>>{%dim_7621} -> tensor<4x?x32x128xf32>{%dim_7621}
%6494 = torch_c.from_builtin_tensor %6493 : tensor<4x?x32x128xf32> -> !torch.vtensor<[4,?,32,128],f32>
torch.bind_symbolic_shape %6494, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f32>
%int5_7622 = torch.constant.int 5
%6495 = torch.prims.convert_element_type %6494, %int5_7622 : !torch.vtensor<[4,?,32,128],f32>, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6495, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int131072_7623 = torch.constant.int 131072
%none_7624 = torch.constant.none
%none_7625 = torch.constant.none
%cpu_7626 = torch.constant.device "cpu"
%false_7627 = torch.constant.bool false
%6496 = torch.aten.arange %int131072_7623, %none_7624, %none_7625, %cpu_7626, %false_7627 : !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[131072],si64>
%int0_7628 = torch.constant.int 0
%int128_7629 = torch.constant.int 128
%int2_7630 = torch.constant.int 2
%none_7631 = torch.constant.none
%none_7632 = torch.constant.none
%cpu_7633 = torch.constant.device "cpu"
%false_7634 = torch.constant.bool false
%6497 = torch.aten.arange.start_step %int0_7628, %int128_7629, %int2_7630, %none_7631, %none_7632, %cpu_7633, %false_7634 : !torch.int, !torch.int, !torch.int, !torch.none, !torch.none, !torch.Device, !torch.bool -> !torch.vtensor<[64],si64>
%int0_7635 = torch.constant.int 0
%int0_7636 = torch.constant.int 0
%int64_7637 = torch.constant.int 64
%int1_7638 = torch.constant.int 1
%6498 = torch.aten.slice.Tensor %6497, %int0_7635, %int0_7636, %int64_7637, %int1_7638 : !torch.vtensor<[64],si64>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[64],si64>
%int6_7639 = torch.constant.int 6
%6499 = torch.prims.convert_element_type %6498, %int6_7639 : !torch.vtensor<[64],si64>, !torch.int -> !torch.vtensor<[64],f32>
%int128_7640 = torch.constant.int 128
%6500 = torch.aten.div.Scalar %6499, %int128_7640 : !torch.vtensor<[64],f32>, !torch.int -> !torch.vtensor<[64],f32>
%float5.000000e05_7641 = torch.constant.float 5.000000e+05
%6501 = torch.aten.pow.Scalar %float5.000000e05_7641, %6500 : !torch.float, !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%6502 = torch.aten.reciprocal %6501 : !torch.vtensor<[64],f32> -> !torch.vtensor<[64],f32>
%float1.000000e00_7642 = torch.constant.float 1.000000e+00
%6503 = torch.aten.mul.Scalar %6502, %float1.000000e00_7642 : !torch.vtensor<[64],f32>, !torch.float -> !torch.vtensor<[64],f32>
%int131072_7643 = torch.constant.int 131072
%int1_7644 = torch.constant.int 1
%6504 = torch.prim.ListConstruct %int131072_7643, %int1_7644 : (!torch.int, !torch.int) -> !torch.list<int>
%6505 = torch.aten.view %6496, %6504 : !torch.vtensor<[131072],si64>, !torch.list<int> -> !torch.vtensor<[131072,1],si64>
%6506 = torch.aten.mul.Tensor %6505, %6503 : !torch.vtensor<[131072,1],si64>, !torch.vtensor<[64],f32> -> !torch.vtensor<[131072,64],f32>
%6507 = torch.aten.cos %6506 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6508 = torch.aten.sin %6506 : !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],f32>
%6509 = torch.aten.complex %6507, %6508 : !torch.vtensor<[131072,64],f32>, !torch.vtensor<[131072,64],f32> -> !torch.vtensor<[131072,64],complex<f32>>
%int1_7645 = torch.constant.int 1
%6510 = torch.aten.size.int %6453, %int1_7645 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int0_7646 = torch.constant.int 0
%6511 = torch.aten.add.int %int0_7646, %6510 : !torch.int, !torch.int -> !torch.int
%int0_7647 = torch.constant.int 0
%int0_7648 = torch.constant.int 0
%int1_7649 = torch.constant.int 1
%6512 = torch.aten.slice.Tensor %6509, %int0_7647, %int0_7648, %6511, %int1_7649 : !torch.vtensor<[131072,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6512, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int1_7650 = torch.constant.int 1
%int0_7651 = torch.constant.int 0
%int9223372036854775807_7652 = torch.constant.int 9223372036854775807
%int1_7653 = torch.constant.int 1
%6513 = torch.aten.slice.Tensor %6512, %int1_7650, %int0_7651, %int9223372036854775807_7652, %int1_7653 : !torch.vtensor<[?,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[?,64],complex<f32>>
torch.bind_symbolic_shape %6513, [%292], affine_map<()[s0] -> (s0 * 32, 64)> : !torch.vtensor<[?,64],complex<f32>>
%int0_7654 = torch.constant.int 0
%6514 = torch.aten.unsqueeze %6513, %int0_7654 : !torch.vtensor<[?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,64],complex<f32>>
torch.bind_symbolic_shape %6514, [%292], affine_map<()[s0] -> (1, s0 * 32, 64)> : !torch.vtensor<[1,?,64],complex<f32>>
%int2_7655 = torch.constant.int 2
%6515 = torch.aten.unsqueeze %6514, %int2_7655 : !torch.vtensor<[1,?,64],complex<f32>>, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6515, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%int3_7656 = torch.constant.int 3
%int0_7657 = torch.constant.int 0
%int9223372036854775807_7658 = torch.constant.int 9223372036854775807
%int1_7659 = torch.constant.int 1
%6516 = torch.aten.slice.Tensor %6515, %int3_7656, %int0_7657, %int9223372036854775807_7658, %int1_7659 : !torch.vtensor<[1,?,1,64],complex<f32>>, !torch.int, !torch.int, !torch.int, !torch.int -> !torch.vtensor<[1,?,1,64],complex<f32>>
torch.bind_symbolic_shape %6516, [%292], affine_map<()[s0] -> (1, s0 * 32, 1, 64)> : !torch.vtensor<[1,?,1,64],complex<f32>>
%6517 = torch_c.to_builtin_tensor %6464 : !torch.vtensor<[4,?,8,128],f16> -> tensor<4x?x8x128xf16>
%c1_7660 = arith.constant 1 : index
%dim_7661 = tensor.dim %6517, %c1_7660 : tensor<4x?x8x128xf16>
%6518 = flow.tensor.bitcast %6517 : tensor<4x?x8x128xf16>{%dim_7661} -> tensor<4x?x8x64xcomplex<f16>>{%dim_7661}
%6519 = torch_c.from_builtin_tensor %6518 : tensor<4x?x8x64xcomplex<f16>> -> !torch.vtensor<[4,?,8,64],complex<f16>>
torch.bind_symbolic_shape %6519, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f16>>
%6520 = torch.aten.mul.Tensor %6519, %6516 : !torch.vtensor<[4,?,8,64],complex<f16>>, !torch.vtensor<[1,?,1,64],complex<f32>> -> !torch.vtensor<[4,?,8,64],complex<f32>>
torch.bind_symbolic_shape %6520, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 64)> : !torch.vtensor<[4,?,8,64],complex<f32>>
%6521 = torch_c.to_builtin_tensor %6520 : !torch.vtensor<[4,?,8,64],complex<f32>> -> tensor<4x?x8x64xcomplex<f32>>
%c1_7662 = arith.constant 1 : index
%dim_7663 = tensor.dim %6521, %c1_7662 : tensor<4x?x8x64xcomplex<f32>>
%6522 = flow.tensor.bitcast %6521 : tensor<4x?x8x64xcomplex<f32>>{%dim_7663} -> tensor<4x?x8x128xf32>{%dim_7663}
%6523 = torch_c.from_builtin_tensor %6522 : tensor<4x?x8x128xf32> -> !torch.vtensor<[4,?,8,128],f32>
torch.bind_symbolic_shape %6523, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f32>
%int5_7664 = torch.constant.int 5
%6524 = torch.prims.convert_element_type %6523, %int5_7664 : !torch.vtensor<[4,?,8,128],f32>, !torch.int -> !torch.vtensor<[4,?,8,128],f16>
torch.bind_symbolic_shape %6524, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 128)> : !torch.vtensor<[4,?,8,128],f16>
%int64_7665 = torch.constant.int 64
%6525 = torch.aten.mul.Scalar %arg2, %int64_7665 : !torch.vtensor<[4,?],si64>, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6525, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int62 = torch.constant.int 62
%int1_7666 = torch.constant.int 1
%6526 = torch.aten.add.Scalar %6525, %int62, %int1_7666 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6526, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7667 = torch.constant.int 4
%int32_7668 = torch.constant.int 32
%int8_7669 = torch.constant.int 8
%int128_7670 = torch.constant.int 128
%6527 = torch.prim.ListConstruct %int4_7667, %425, %int32_7668, %int8_7669, %int128_7670 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6528 = torch.aten.view %6524, %6527 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6528, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7671 = torch.constant.int 4
%6529 = torch.aten.mul.int %int4_7671, %425 : !torch.int, !torch.int -> !torch.int
%int32_7672 = torch.constant.int 32
%int8_7673 = torch.constant.int 8
%int128_7674 = torch.constant.int 128
%6530 = torch.prim.ListConstruct %6529, %int32_7672, %int8_7673, %int128_7674 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6531 = torch.aten.view %6528, %6530 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6531, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7675 = torch.constant.int 4
%6532 = torch.aten.mul.int %int4_7675, %425 : !torch.int, !torch.int -> !torch.int
%6533 = torch.prim.ListConstruct %6532 : (!torch.int) -> !torch.list<int>
%6534 = torch.aten.view %6526, %6533 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6534, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%int32_7676 = torch.constant.int 32
%int2_7677 = torch.constant.int 2
%int32_7678 = torch.constant.int 32
%int8_7679 = torch.constant.int 8
%int128_7680 = torch.constant.int 128
%6535 = torch.prim.ListConstruct %416, %int32_7676, %int2_7677, %int32_7678, %int8_7679, %int128_7680 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6536 = torch.aten.view %6368, %6535 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6536, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7681 = torch.constant.int 32
%6537 = torch.aten.mul.int %416, %int32_7681 : !torch.int, !torch.int -> !torch.int
%int2_7682 = torch.constant.int 2
%6538 = torch.aten.mul.int %6537, %int2_7682 : !torch.int, !torch.int -> !torch.int
%int32_7683 = torch.constant.int 32
%int8_7684 = torch.constant.int 8
%int128_7685 = torch.constant.int 128
%6539 = torch.prim.ListConstruct %6538, %int32_7683, %int8_7684, %int128_7685 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6540 = torch.aten.view %6536, %6539 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6540, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%6541 = torch.prim.ListConstruct %6534 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7686 = torch.constant.bool false
%6542 = torch.aten.index_put %6540, %6541, %6531, %false_7686 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6542, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7687 = torch.constant.int 32
%int2_7688 = torch.constant.int 2
%int32_7689 = torch.constant.int 32
%int8_7690 = torch.constant.int 8
%int128_7691 = torch.constant.int 128
%6543 = torch.prim.ListConstruct %416, %int32_7687, %int2_7688, %int32_7689, %int8_7690, %int128_7691 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6544 = torch.aten.view %6542, %6543 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6544, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7692 = torch.constant.int 2097152
%6545 = torch.prim.ListConstruct %416, %int2097152_7692 : (!torch.int, !torch.int) -> !torch.list<int>
%6546 = torch.aten.view %6544, %6545 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.bind_symbolic_shape %6546, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int32_7693 = torch.constant.int 32
%int2_7694 = torch.constant.int 2
%int32_7695 = torch.constant.int 32
%int8_7696 = torch.constant.int 8
%int128_7697 = torch.constant.int 128
%6547 = torch.prim.ListConstruct %416, %int32_7693, %int2_7694, %int32_7695, %int8_7696, %int128_7697 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6548 = torch.aten.view %6546, %6547 : !torch.vtensor<[?,2097152],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6548, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int32_7698 = torch.constant.int 32
%int8_7699 = torch.constant.int 8
%int128_7700 = torch.constant.int 128
%6549 = torch.prim.ListConstruct %6538, %int32_7698, %int8_7699, %int128_7700 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6550 = torch.aten.view %6548, %6549 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6550, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int4_7701 = torch.constant.int 4
%int32_7702 = torch.constant.int 32
%int8_7703 = torch.constant.int 8
%int128_7704 = torch.constant.int 128
%6551 = torch.prim.ListConstruct %int4_7701, %425, %int32_7702, %int8_7703, %int128_7704 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6552 = torch.aten.view %6466, %6551 : !torch.vtensor<[4,?,8,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,8,128],f16>
torch.bind_symbolic_shape %6552, [%292], affine_map<()[s0] -> (4, s0, 32, 8, 128)> : !torch.vtensor<[4,?,32,8,128],f16>
%int4_7705 = torch.constant.int 4
%6553 = torch.aten.mul.int %int4_7705, %425 : !torch.int, !torch.int -> !torch.int
%int32_7706 = torch.constant.int 32
%int8_7707 = torch.constant.int 8
%int128_7708 = torch.constant.int 128
%6554 = torch.prim.ListConstruct %6553, %int32_7706, %int8_7707, %int128_7708 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6555 = torch.aten.view %6552, %6554 : !torch.vtensor<[4,?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6555, [%292], affine_map<()[s0] -> (s0 * 4, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int1_7709 = torch.constant.int 1
%int1_7710 = torch.constant.int 1
%6556 = torch.aten.add.Scalar %6526, %int1_7709, %int1_7710 : !torch.vtensor<[4,?],si64>, !torch.int, !torch.int -> !torch.vtensor<[4,?],si64>
torch.bind_symbolic_shape %6556, [%292], affine_map<()[s0] -> (4, s0)> : !torch.vtensor<[4,?],si64>
%int4_7711 = torch.constant.int 4
%6557 = torch.aten.mul.int %int4_7711, %425 : !torch.int, !torch.int -> !torch.int
%6558 = torch.prim.ListConstruct %6557 : (!torch.int) -> !torch.list<int>
%6559 = torch.aten.view %6556, %6558 : !torch.vtensor<[4,?],si64>, !torch.list<int> -> !torch.vtensor<[?],si64>
torch.bind_symbolic_shape %6559, [%292], affine_map<()[s0] -> (s0 * 4)> : !torch.vtensor<[?],si64>
%6560 = torch.prim.ListConstruct %6559 : (!torch.vtensor<[?],si64>) -> !torch.list<optional<vtensor>>
%false_7712 = torch.constant.bool false
%6561 = torch.aten.index_put %6550, %6560, %6555, %false_7712 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<optional<vtensor>>, !torch.vtensor<[?,32,8,128],f16>, !torch.bool -> !torch.vtensor<[?,32,8,128],f16>
torch.bind_symbolic_shape %6561, [%293], affine_map<()[s0] -> (s0 * 64, 32, 8, 128)> : !torch.vtensor<[?,32,8,128],f16>
%int32_7713 = torch.constant.int 32
%int2_7714 = torch.constant.int 2
%int32_7715 = torch.constant.int 32
%int8_7716 = torch.constant.int 8
%int128_7717 = torch.constant.int 128
%6562 = torch.prim.ListConstruct %416, %int32_7713, %int2_7714, %int32_7715, %int8_7716, %int128_7717 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6563 = torch.aten.view %6561, %6562 : !torch.vtensor<[?,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,32,2,32,8,128],f16>
torch.bind_symbolic_shape %6563, [%293], affine_map<()[s0] -> (s0, 32, 2, 32, 8, 128)> : !torch.vtensor<[?,32,2,32,8,128],f16>
%int2097152_7718 = torch.constant.int 2097152
%6564 = torch.prim.ListConstruct %416, %int2097152_7718 : (!torch.int, !torch.int) -> !torch.list<int>
%6565 = torch.aten.view %6563, %6564 : !torch.vtensor<[?,32,2,32,8,128],f16>, !torch.list<int> -> !torch.vtensor<[?,2097152],f16>
torch.overwrite.tensor.contents %6565 overwrites %arg3 : !torch.vtensor<[?,2097152],f16>, !torch.tensor<[?,2097152],f16>
torch.bind_symbolic_shape %6565, [%293], affine_map<()[s0] -> (s0, 2097152)> : !torch.vtensor<[?,2097152],f16>
%int-2_7719 = torch.constant.int -2
%6566 = torch.aten.unsqueeze %6524, %int-2_7719 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6566, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int4_7720 = torch.constant.int 4
%int8_7721 = torch.constant.int 8
%int4_7722 = torch.constant.int 4
%int128_7723 = torch.constant.int 128
%6567 = torch.prim.ListConstruct %int4_7720, %6510, %int8_7721, %int4_7722, %int128_7723 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7724 = torch.constant.bool false
%6568 = torch.aten.expand %6566, %6567, %false_7724 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6568, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7725 = torch.constant.int 0
%6569 = torch.aten.clone %6568, %int0_7725 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6569, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7726 = torch.constant.int 4
%int32_7727 = torch.constant.int 32
%int128_7728 = torch.constant.int 128
%6570 = torch.prim.ListConstruct %int4_7726, %6510, %int32_7727, %int128_7728 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6571 = torch.aten._unsafe_view %6569, %6570 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6571, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int-2_7729 = torch.constant.int -2
%6572 = torch.aten.unsqueeze %6466, %int-2_7729 : !torch.vtensor<[4,?,8,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,1,128],f16>
torch.bind_symbolic_shape %6572, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 1, 128)> : !torch.vtensor<[4,?,8,1,128],f16>
%int1_7730 = torch.constant.int 1
%6573 = torch.aten.size.int %6460, %int1_7730 : !torch.vtensor<[4,?,1024],f16>, !torch.int -> !torch.int
%int4_7731 = torch.constant.int 4
%int8_7732 = torch.constant.int 8
%int4_7733 = torch.constant.int 4
%int128_7734 = torch.constant.int 128
%6574 = torch.prim.ListConstruct %int4_7731, %6573, %int8_7732, %int4_7733, %int128_7734 : (!torch.int, !torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%false_7735 = torch.constant.bool false
%6575 = torch.aten.expand %6572, %6574, %false_7735 : !torch.vtensor<[4,?,8,1,128],f16>, !torch.list<int>, !torch.bool -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6575, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int0_7736 = torch.constant.int 0
%6576 = torch.aten.clone %6575, %int0_7736 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.int -> !torch.vtensor<[4,?,8,4,128],f16>
torch.bind_symbolic_shape %6576, [%292], affine_map<()[s0] -> (4, s0 * 32, 8, 4, 128)> : !torch.vtensor<[4,?,8,4,128],f16>
%int4_7737 = torch.constant.int 4
%int32_7738 = torch.constant.int 32
%int128_7739 = torch.constant.int 128
%6577 = torch.prim.ListConstruct %int4_7737, %6573, %int32_7738, %int128_7739 : (!torch.int, !torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6578 = torch.aten._unsafe_view %6576, %6577 : !torch.vtensor<[4,?,8,4,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6578, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int1_7740 = torch.constant.int 1
%int2_7741 = torch.constant.int 2
%6579 = torch.aten.transpose.int %6495, %int1_7740, %int2_7741 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6579, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7742 = torch.constant.int 1
%int2_7743 = torch.constant.int 2
%6580 = torch.aten.transpose.int %6571, %int1_7742, %int2_7743 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6580, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7744 = torch.constant.int 1
%int2_7745 = torch.constant.int 2
%6581 = torch.aten.transpose.int %6578, %int1_7744, %int2_7745 : !torch.vtensor<[4,?,32,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,32,?,128],f16>
torch.bind_symbolic_shape %6581, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%float0.000000e00_7746 = torch.constant.float 0.000000e+00
%false_7747 = torch.constant.bool false
%none_7748 = torch.constant.none
%6582:2 = torch.operator "torch.aten._scaled_dot_product_flash_attention_for_cpu"(%6579, %6580, %6581, %float0.000000e00_7746, %false_7747, %320, %none_7748) : (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?,128],f16>, !torch.float, !torch.bool, !torch.vtensor<[4,1,?,?],f16>, !torch.none) -> (!torch.vtensor<[4,32,?,128],f16>, !torch.vtensor<[4,32,?],f32>)
torch.bind_symbolic_shape %6582#0, [%292], affine_map<()[s0] -> (4, 32, s0 * 32, 128)> : !torch.vtensor<[4,32,?,128],f16>
%int1_7749 = torch.constant.int 1
%int2_7750 = torch.constant.int 2
%6583 = torch.aten.transpose.int %6582#0, %int1_7749, %int2_7750 : !torch.vtensor<[4,32,?,128],f16>, !torch.int, !torch.int -> !torch.vtensor<[4,?,32,128],f16>
torch.bind_symbolic_shape %6583, [%292], affine_map<()[s0] -> (4, s0 * 32, 32, 128)> : !torch.vtensor<[4,?,32,128],f16>
%int4_7751 = torch.constant.int 4
%int4096_7752 = torch.constant.int 4096
%6584 = torch.prim.ListConstruct %int4_7751, %6481, %int4096_7752 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6585 = torch.aten.view %6583, %6584 : !torch.vtensor<[4,?,32,128],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6585, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7753 = torch.constant.int -2
%int-1_7754 = torch.constant.int -1
%6586 = torch.aten.transpose.int %284, %int-2_7753, %int-1_7754 : !torch.vtensor<[4096,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,4096],f16>
%int4_7755 = torch.constant.int 4
%6587 = torch.aten.mul.int %int4_7755, %6481 : !torch.int, !torch.int -> !torch.int
%int4096_7756 = torch.constant.int 4096
%6588 = torch.prim.ListConstruct %6587, %int4096_7756 : (!torch.int, !torch.int) -> !torch.list<int>
%6589 = torch.aten.view %6585, %6588 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6589, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6590 = torch.aten.mm %6589, %6586 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6590, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7757 = torch.constant.int 4
%int4096_7758 = torch.constant.int 4096
%6591 = torch.prim.ListConstruct %int4_7757, %6481, %int4096_7758 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6592 = torch.aten.view %6590, %6591 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6592, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7759 = torch.constant.int 1
%6593 = torch.aten.add.Tensor %6430, %6592, %int1_7759 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6593, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7760 = torch.constant.int 6
%6594 = torch.prims.convert_element_type %6593, %int6_7760 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6594, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7761 = torch.constant.int 2
%6595 = torch.aten.pow.Tensor_Scalar %6594, %int2_7761 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6595, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7762 = torch.constant.int -1
%6596 = torch.prim.ListConstruct %int-1_7762 : (!torch.int) -> !torch.list<int>
%true_7763 = torch.constant.bool true
%none_7764 = torch.constant.none
%6597 = torch.aten.mean.dim %6595, %6596, %true_7763, %none_7764 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6597, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7765 = torch.constant.float 9.9999997473787516E-6
%int1_7766 = torch.constant.int 1
%6598 = torch.aten.add.Scalar %6597, %float9.999990e-06_7765, %int1_7766 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6598, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6599 = torch.aten.rsqrt %6598 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6599, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6600 = torch.aten.mul.Tensor %6594, %6599 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6600, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6601 = torch.aten.mul.Tensor %285, %6600 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6601, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7767 = torch.constant.int 5
%6602 = torch.prims.convert_element_type %6601, %int5_7767 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6602, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7768 = torch.constant.int -2
%int-1_7769 = torch.constant.int -1
%6603 = torch.aten.transpose.int %286, %int-2_7768, %int-1_7769 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7770 = torch.constant.int 4
%6604 = torch.aten.mul.int %int4_7770, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7771 = torch.constant.int 4096
%6605 = torch.prim.ListConstruct %6604, %int4096_7771 : (!torch.int, !torch.int) -> !torch.list<int>
%6606 = torch.aten.view %6602, %6605 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6606, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6607 = torch.aten.mm %6606, %6603 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6607, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7772 = torch.constant.int 4
%int14336_7773 = torch.constant.int 14336
%6608 = torch.prim.ListConstruct %int4_7772, %294, %int14336_7773 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6609 = torch.aten.view %6607, %6608 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6609, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6610 = torch.aten.silu %6609 : !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6610, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7774 = torch.constant.int -2
%int-1_7775 = torch.constant.int -1
%6611 = torch.aten.transpose.int %287, %int-2_7774, %int-1_7775 : !torch.vtensor<[14336,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,14336],f16>
%int4_7776 = torch.constant.int 4
%6612 = torch.aten.mul.int %int4_7776, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7777 = torch.constant.int 4096
%6613 = torch.prim.ListConstruct %6612, %int4096_7777 : (!torch.int, !torch.int) -> !torch.list<int>
%6614 = torch.aten.view %6602, %6613 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6614, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6615 = torch.aten.mm %6614, %6611 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,14336],f16> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6615, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%int4_7778 = torch.constant.int 4
%int14336_7779 = torch.constant.int 14336
%6616 = torch.prim.ListConstruct %int4_7778, %294, %int14336_7779 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6617 = torch.aten.view %6615, %6616 : !torch.vtensor<[?,14336],f16>, !torch.list<int> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6617, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%6618 = torch.aten.mul.Tensor %6610, %6617 : !torch.vtensor<[4,?,14336],f16>, !torch.vtensor<[4,?,14336],f16> -> !torch.vtensor<[4,?,14336],f16>
torch.bind_symbolic_shape %6618, [%292], affine_map<()[s0] -> (4, s0 * 32, 14336)> : !torch.vtensor<[4,?,14336],f16>
%int-2_7780 = torch.constant.int -2
%int-1_7781 = torch.constant.int -1
%6619 = torch.aten.transpose.int %288, %int-2_7780, %int-1_7781 : !torch.vtensor<[4096,14336],f16>, !torch.int, !torch.int -> !torch.vtensor<[14336,4096],f16>
%int1_7782 = torch.constant.int 1
%6620 = torch.aten.size.int %6609, %int1_7782 : !torch.vtensor<[4,?,14336],f16>, !torch.int -> !torch.int
%int4_7783 = torch.constant.int 4
%6621 = torch.aten.mul.int %int4_7783, %6620 : !torch.int, !torch.int -> !torch.int
%int14336_7784 = torch.constant.int 14336
%6622 = torch.prim.ListConstruct %6621, %int14336_7784 : (!torch.int, !torch.int) -> !torch.list<int>
%6623 = torch.aten.view %6618, %6622 : !torch.vtensor<[4,?,14336],f16>, !torch.list<int> -> !torch.vtensor<[?,14336],f16>
torch.bind_symbolic_shape %6623, [%292], affine_map<()[s0] -> (s0 * 128, 14336)> : !torch.vtensor<[?,14336],f16>
%6624 = torch.aten.mm %6623, %6619 : !torch.vtensor<[?,14336],f16>, !torch.vtensor<[14336,4096],f16> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6624, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%int4_7785 = torch.constant.int 4
%int4096_7786 = torch.constant.int 4096
%6625 = torch.prim.ListConstruct %int4_7785, %6620, %int4096_7786 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6626 = torch.aten.view %6624, %6625 : !torch.vtensor<[?,4096],f16>, !torch.list<int> -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6626, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int1_7787 = torch.constant.int 1
%6627 = torch.aten.add.Tensor %6593, %6626, %int1_7787 : !torch.vtensor<[4,?,4096],f16>, !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6627, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int6_7788 = torch.constant.int 6
%6628 = torch.prims.convert_element_type %6627, %int6_7788 : !torch.vtensor<[4,?,4096],f16>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6628, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int2_7789 = torch.constant.int 2
%6629 = torch.aten.pow.Tensor_Scalar %6628, %int2_7789 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6629, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int-1_7790 = torch.constant.int -1
%6630 = torch.prim.ListConstruct %int-1_7790 : (!torch.int) -> !torch.list<int>
%true_7791 = torch.constant.bool true
%none_7792 = torch.constant.none
%6631 = torch.aten.mean.dim %6629, %6630, %true_7791, %none_7792 : !torch.vtensor<[4,?,4096],f32>, !torch.list<int>, !torch.bool, !torch.none -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6631, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%float9.999990e-06_7793 = torch.constant.float 9.9999997473787516E-6
%int1_7794 = torch.constant.int 1
%6632 = torch.aten.add.Scalar %6631, %float9.999990e-06_7793, %int1_7794 : !torch.vtensor<[4,?,1],f32>, !torch.float, !torch.int -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6632, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6633 = torch.aten.rsqrt %6632 : !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,1],f32>
torch.bind_symbolic_shape %6633, [%292], affine_map<()[s0] -> (4, s0 * 32, 1)> : !torch.vtensor<[4,?,1],f32>
%6634 = torch.aten.mul.Tensor %6628, %6633 : !torch.vtensor<[4,?,4096],f32>, !torch.vtensor<[4,?,1],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6634, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%6635 = torch.aten.mul.Tensor %289, %6634 : !torch.vtensor<[4096],f32>, !torch.vtensor<[4,?,4096],f32> -> !torch.vtensor<[4,?,4096],f32>
torch.bind_symbolic_shape %6635, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f32>
%int5_7795 = torch.constant.int 5
%6636 = torch.prims.convert_element_type %6635, %int5_7795 : !torch.vtensor<[4,?,4096],f32>, !torch.int -> !torch.vtensor<[4,?,4096],f16>
torch.bind_symbolic_shape %6636, [%292], affine_map<()[s0] -> (4, s0 * 32, 4096)> : !torch.vtensor<[4,?,4096],f16>
%int-2_7796 = torch.constant.int -2
%int-1_7797 = torch.constant.int -1
%6637 = torch.aten.transpose.int %290, %int-2_7796, %int-1_7797 : !torch.vtensor<[128256,4096],f16>, !torch.int, !torch.int -> !torch.vtensor<[4096,128256],f16>
%int4_7798 = torch.constant.int 4
%6638 = torch.aten.mul.int %int4_7798, %294 : !torch.int, !torch.int -> !torch.int
%int4096_7799 = torch.constant.int 4096
%6639 = torch.prim.ListConstruct %6638, %int4096_7799 : (!torch.int, !torch.int) -> !torch.list<int>
%6640 = torch.aten.view %6636, %6639 : !torch.vtensor<[4,?,4096],f16>, !torch.list<int> -> !torch.vtensor<[?,4096],f16>
torch.bind_symbolic_shape %6640, [%292], affine_map<()[s0] -> (s0 * 128, 4096)> : !torch.vtensor<[?,4096],f16>
%6641 = torch.aten.mm %6640, %6637 : !torch.vtensor<[?,4096],f16>, !torch.vtensor<[4096,128256],f16> -> !torch.vtensor<[?,128256],f16>
torch.bind_symbolic_shape %6641, [%292], affine_map<()[s0] -> (s0 * 128, 128256)> : !torch.vtensor<[?,128256],f16>
%int4_7800 = torch.constant.int 4
%int128256 = torch.constant.int 128256
%6642 = torch.prim.ListConstruct %int4_7800, %294, %int128256 : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
%6643 = torch.aten.view %6641, %6642 : !torch.vtensor<[?,128256],f16>, !torch.list<int> -> !torch.vtensor<[4,?,128256],f16>
torch.bind_symbolic_shape %6643, [%292], affine_map<()[s0] -> (4, s0 * 32, 128256)> : !torch.vtensor<[4,?,128256],f16>
return %6643 : !torch.vtensor<[4,?,128256],f16>
}
}