diff --git "a/locdit_p4_c4.mlmodelc/model.mil" "b/locdit_p4_c4.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/locdit_p4_c4.mlmodelc/model.mil" @@ -0,0 +1,1665 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.11.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] +{ + func main(tensor cond, tensor dt_emb, tensor mu, tensor t_emb, tensor x) { + tensor locdit_decoder_layers_1_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2101440))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2097280))))[name = string("locdit_decoder_layers_1_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2103552))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2366336))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2365760))))[name = string("locdit_decoder_layers_1_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2366656))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2629440))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2628864))))[name = string("locdit_decoder_layers_1_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2629760))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4729088))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4726976))))[name = string("locdit_decoder_layers_1_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4730176))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8932800))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8924544))))[name = string("locdit_decoder_layers_1_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8936960))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13139584))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13131328))))[name = string("locdit_decoder_layers_1_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_1_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13143744))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17340224))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17338112))))[name = string("locdit_decoder_layers_1_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17341312))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19442688))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19438528))))[name = string("locdit_decoder_layers_2_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19444800))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19707584))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19707008))))[name = string("locdit_decoder_layers_2_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19707904))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19970688))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19970112))))[name = string("locdit_decoder_layers_2_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19971008))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22070336))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22068224))))[name = string("locdit_decoder_layers_2_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22071424))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26274048))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26265792))))[name = string("locdit_decoder_layers_2_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26278208))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30480832))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30472576))))[name = string("locdit_decoder_layers_2_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_2_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30484992))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34681472))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34679360))))[name = string("locdit_decoder_layers_2_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34682560))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36783936))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36779776))))[name = string("locdit_decoder_layers_3_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36786048))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37048832))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37048256))))[name = string("locdit_decoder_layers_3_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37049152))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37311936))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37311360))))[name = string("locdit_decoder_layers_3_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37312256))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39411584))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39409472))))[name = string("locdit_decoder_layers_3_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39412672))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43615296))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43607040))))[name = string("locdit_decoder_layers_3_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43619456))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47822080))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47813824))))[name = string("locdit_decoder_layers_3_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_3_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47826240))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52022720))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52020608))))[name = string("locdit_decoder_layers_3_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52023808))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54125184))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54121024))))[name = string("locdit_decoder_layers_4_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54127296))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54390080))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54389504))))[name = string("locdit_decoder_layers_4_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54390400))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54653184))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54652608))))[name = string("locdit_decoder_layers_4_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54653504))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56752832))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56750720))))[name = string("locdit_decoder_layers_4_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(56753920))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60956544))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60948288))))[name = string("locdit_decoder_layers_4_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60960704))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65163328))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65155072))))[name = string("locdit_decoder_layers_4_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_4_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65167488))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69363968))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69361856))))[name = string("locdit_decoder_layers_4_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69365056))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71466432))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71462272))))[name = string("locdit_decoder_layers_5_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71468544))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71731328))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71730752))))[name = string("locdit_decoder_layers_5_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71731648))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71994432))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71993856))))[name = string("locdit_decoder_layers_5_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71994752))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74094080))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74091968))))[name = string("locdit_decoder_layers_5_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74095168))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78297792))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78289536))))[name = string("locdit_decoder_layers_5_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78301952))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82504576))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82496320))))[name = string("locdit_decoder_layers_5_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_5_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82508736))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86705216))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86703104))))[name = string("locdit_decoder_layers_5_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86706304))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88807680))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88803520))))[name = string("locdit_decoder_layers_6_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88809792))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89072576))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89072000))))[name = string("locdit_decoder_layers_6_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89072896))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89335680))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89335104))))[name = string("locdit_decoder_layers_6_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89336000))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91435328))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91433216))))[name = string("locdit_decoder_layers_6_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91436416))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95639040))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95630784))))[name = string("locdit_decoder_layers_6_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95643200))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99845824))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99837568))))[name = string("locdit_decoder_layers_6_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_6_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99849984))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104046464))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104044352))))[name = string("locdit_decoder_layers_6_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104047552))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106148928))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106144768))))[name = string("locdit_decoder_layers_7_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106151040))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106413824))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106413248))))[name = string("locdit_decoder_layers_7_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106414144))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106676928))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106676352))))[name = string("locdit_decoder_layers_7_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106677248))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108776576))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108774464))))[name = string("locdit_decoder_layers_7_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108777664))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112980288))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112972032))))[name = string("locdit_decoder_layers_7_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112984448))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117187072))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117178816))))[name = string("locdit_decoder_layers_7_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_7_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117191232))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121387712))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121385600))))[name = string("locdit_decoder_layers_7_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121388800))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123490176))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123486016))))[name = string("locdit_decoder_layers_8_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123492288))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123755072))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123754496))))[name = string("locdit_decoder_layers_8_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123755392))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124018176))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124017600))))[name = string("locdit_decoder_layers_8_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124018496))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126117824))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126115712))))[name = string("locdit_decoder_layers_8_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126118912))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130321536))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130313280))))[name = string("locdit_decoder_layers_8_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130325696))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134528320))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134520064))))[name = string("locdit_decoder_layers_8_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_8_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134532480))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138728960))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138726848))))[name = string("locdit_decoder_layers_8_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138730048))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140831424))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140827264))))[name = string("locdit_decoder_layers_9_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140833536))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141096320))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141095744))))[name = string("locdit_decoder_layers_9_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141096640))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141359424))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141358848))))[name = string("locdit_decoder_layers_9_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_self_attn_o_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141359744))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143459072))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143456960))))[name = string("locdit_decoder_layers_9_self_attn_o_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143460160))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147662784))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147654528))))[name = string("locdit_decoder_layers_9_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147666944))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151869568))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151861312))))[name = string("locdit_decoder_layers_9_mlp_up_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_9_mlp_down_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151873728))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156070208))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156068096))))[name = string("locdit_decoder_layers_9_mlp_down_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_10_self_attn_q_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156071296))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158172672))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158168512))))[name = string("locdit_decoder_layers_10_self_attn_q_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_10_self_attn_k_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158174784))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158437568))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158436992))))[name = string("locdit_decoder_layers_10_self_attn_k_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_10_self_attn_v_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158437888))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158700672))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158700096))))[name = string("locdit_decoder_layers_10_self_attn_v_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_10_mlp_gate_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158700992))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162903616))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162895360))))[name = string("locdit_decoder_layers_10_mlp_gate_proj_weight_cast_fp16")]; + tensor locdit_decoder_layers_10_mlp_up_proj_weight_cast_fp16 = constexpr_blockwise_shift_scale(data = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162907776))), offset = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167110400))), scale = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167102144))))[name = string("locdit_decoder_layers_10_mlp_up_proj_weight_cast_fp16")]; + string mu_pad_type_0 = const()[name = string("mu_pad_type_0"), val = string("valid")]; + tensor mu_strides_0 = const()[name = string("mu_strides_0"), val = tensor([1, 1])]; + tensor mu_pad_0 = const()[name = string("mu_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor mu_dilations_0 = const()[name = string("mu_dilations_0"), val = tensor([1, 1])]; + int32 mu_groups_0 = const()[name = string("mu_groups_0"), val = int32(1)]; + string mu_to_fp16_dtype_0 = const()[name = string("mu_to_fp16_dtype_0"), val = string("fp16")]; + tensor mu_identity_weight_to_fp16 = const()[name = string("mu_identity_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167114560)))]; + tensor mu_to_fp16 = cast(dtype = mu_to_fp16_dtype_0, x = mu)[name = string("cast_78")]; + tensor mu_cast_fp16 = conv(dilations = mu_dilations_0, groups = mu_groups_0, pad = mu_pad_0, pad_type = mu_pad_type_0, strides = mu_strides_0, weight = mu_identity_weight_to_fp16, x = mu_to_fp16)[name = string("mu_cast_fp16")]; + string x_emb_pad_type_0 = const()[name = string("x_emb_pad_type_0"), val = string("valid")]; + tensor x_emb_strides_0 = const()[name = string("x_emb_strides_0"), val = tensor([1, 1])]; + tensor x_emb_pad_0 = const()[name = string("x_emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor x_emb_dilations_0 = const()[name = string("x_emb_dilations_0"), val = tensor([1, 1])]; + int32 x_emb_groups_0 = const()[name = string("x_emb_groups_0"), val = int32(1)]; + string x_to_fp16_dtype_0 = const()[name = string("x_to_fp16_dtype_0"), val = string("fp16")]; + tensor locdit_in_proj_weight_to_fp16 = const()[name = string("locdit_in_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169211776)))]; + tensor locdit_in_proj_bias_to_fp16 = const()[name = string("locdit_in_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169342912)))]; + tensor x_to_fp16 = cast(dtype = x_to_fp16_dtype_0, x = x)[name = string("cast_77")]; + tensor x_emb_cast_fp16 = conv(bias = locdit_in_proj_bias_to_fp16, dilations = x_emb_dilations_0, groups = x_emb_groups_0, pad = x_emb_pad_0, pad_type = x_emb_pad_type_0, strides = x_emb_strides_0, weight = locdit_in_proj_weight_to_fp16, x = x_to_fp16)[name = string("x_emb_cast_fp16")]; + string cond_emb_pad_type_0 = const()[name = string("cond_emb_pad_type_0"), val = string("valid")]; + tensor cond_emb_strides_0 = const()[name = string("cond_emb_strides_0"), val = tensor([1, 1])]; + tensor cond_emb_pad_0 = const()[name = string("cond_emb_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor cond_emb_dilations_0 = const()[name = string("cond_emb_dilations_0"), val = tensor([1, 1])]; + int32 cond_emb_groups_0 = const()[name = string("cond_emb_groups_0"), val = int32(1)]; + string cond_to_fp16_dtype_0 = const()[name = string("cond_to_fp16_dtype_0"), val = string("fp16")]; + tensor locdit_cond_proj_weight_to_fp16 = const()[name = string("locdit_cond_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169345024)))]; + tensor locdit_cond_proj_bias_to_fp16 = const()[name = string("locdit_cond_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169476160)))]; + tensor cond_to_fp16 = cast(dtype = cond_to_fp16_dtype_0, x = cond)[name = string("cast_76")]; + tensor cond_emb_cast_fp16 = conv(bias = locdit_cond_proj_bias_to_fp16, dilations = cond_emb_dilations_0, groups = cond_emb_groups_0, pad = cond_emb_pad_0, pad_type = cond_emb_pad_type_0, strides = cond_emb_strides_0, weight = locdit_cond_proj_weight_to_fp16, x = cond_to_fp16)[name = string("cond_emb_cast_fp16")]; + string input_1_pad_type_0 = const()[name = string("input_1_pad_type_0"), val = string("valid")]; + tensor input_1_strides_0 = const()[name = string("input_1_strides_0"), val = tensor([1, 1])]; + tensor input_1_pad_0 = const()[name = string("input_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_1_dilations_0 = const()[name = string("input_1_dilations_0"), val = tensor([1, 1])]; + int32 input_1_groups_0 = const()[name = string("input_1_groups_0"), val = int32(1)]; + string t_emb_to_fp16_dtype_0 = const()[name = string("t_emb_to_fp16_dtype_0"), val = string("fp16")]; + tensor locdit_time_mlp_linear_1_weight_to_fp16 = const()[name = string("locdit_time_mlp_linear_1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169478272)))]; + tensor locdit_time_mlp_linear_1_bias_to_fp16 = const()[name = string("locdit_time_mlp_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171575488)))]; + tensor t_emb_to_fp16 = cast(dtype = t_emb_to_fp16_dtype_0, x = t_emb)[name = string("cast_75")]; + tensor input_1_cast_fp16 = conv(bias = locdit_time_mlp_linear_1_bias_to_fp16, dilations = input_1_dilations_0, groups = input_1_groups_0, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = input_1_strides_0, weight = locdit_time_mlp_linear_1_weight_to_fp16, x = t_emb_to_fp16)[name = string("input_1_cast_fp16")]; + tensor x_3_cast_fp16 = silu(x = input_1_cast_fp16)[name = string("x_3_cast_fp16")]; + string var_123_pad_type_0 = const()[name = string("op_123_pad_type_0"), val = string("valid")]; + tensor var_123_strides_0 = const()[name = string("op_123_strides_0"), val = tensor([1, 1])]; + tensor var_123_pad_0 = const()[name = string("op_123_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_123_dilations_0 = const()[name = string("op_123_dilations_0"), val = tensor([1, 1])]; + int32 var_123_groups_0 = const()[name = string("op_123_groups_0"), val = int32(1)]; + tensor locdit_time_mlp_linear_2_weight_to_fp16 = const()[name = string("locdit_time_mlp_linear_2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171577600)))]; + tensor locdit_time_mlp_linear_2_bias_to_fp16 = const()[name = string("locdit_time_mlp_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173674816)))]; + tensor var_123_cast_fp16 = conv(bias = locdit_time_mlp_linear_2_bias_to_fp16, dilations = var_123_dilations_0, groups = var_123_groups_0, pad = var_123_pad_0, pad_type = var_123_pad_type_0, strides = var_123_strides_0, weight = locdit_time_mlp_linear_2_weight_to_fp16, x = x_3_cast_fp16)[name = string("op_123_cast_fp16")]; + string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("valid")]; + tensor input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor([1, 1])]; + tensor input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor([1, 1])]; + int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)]; + string dt_emb_to_fp16_dtype_0 = const()[name = string("dt_emb_to_fp16_dtype_0"), val = string("fp16")]; + tensor locdit_delta_time_mlp_linear_1_weight_to_fp16 = const()[name = string("locdit_delta_time_mlp_linear_1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173676928)))]; + tensor locdit_delta_time_mlp_linear_1_bias_to_fp16 = const()[name = string("locdit_delta_time_mlp_linear_1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175774144)))]; + tensor dt_emb_to_fp16 = cast(dtype = dt_emb_to_fp16_dtype_0, x = dt_emb)[name = string("cast_74")]; + tensor input_3_cast_fp16 = conv(bias = locdit_delta_time_mlp_linear_1_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = locdit_delta_time_mlp_linear_1_weight_to_fp16, x = dt_emb_to_fp16)[name = string("input_3_cast_fp16")]; + tensor x_5_cast_fp16 = silu(x = input_3_cast_fp16)[name = string("x_5_cast_fp16")]; + string var_144_pad_type_0 = const()[name = string("op_144_pad_type_0"), val = string("valid")]; + tensor var_144_strides_0 = const()[name = string("op_144_strides_0"), val = tensor([1, 1])]; + tensor var_144_pad_0 = const()[name = string("op_144_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_144_dilations_0 = const()[name = string("op_144_dilations_0"), val = tensor([1, 1])]; + int32 var_144_groups_0 = const()[name = string("op_144_groups_0"), val = int32(1)]; + tensor locdit_delta_time_mlp_linear_2_weight_to_fp16 = const()[name = string("locdit_delta_time_mlp_linear_2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175776256)))]; + tensor locdit_delta_time_mlp_linear_2_bias_to_fp16 = const()[name = string("locdit_delta_time_mlp_linear_2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177873472)))]; + tensor var_144_cast_fp16 = conv(bias = locdit_delta_time_mlp_linear_2_bias_to_fp16, dilations = var_144_dilations_0, groups = var_144_groups_0, pad = var_144_pad_0, pad_type = var_144_pad_type_0, strides = var_144_strides_0, weight = locdit_delta_time_mlp_linear_2_weight_to_fp16, x = x_5_cast_fp16)[name = string("op_144_cast_fp16")]; + tensor t_tok_cast_fp16 = add(x = var_123_cast_fp16, y = var_144_cast_fp16)[name = string("t_tok_cast_fp16")]; + int32 var_148 = const()[name = string("op_148"), val = int32(-1)]; + bool hidden_states_1_interleave_0 = const()[name = string("hidden_states_1_interleave_0"), val = bool(false)]; + tensor hidden_states_1_cast_fp16 = concat(axis = var_148, interleave = hidden_states_1_interleave_0, values = (mu_cast_fp16, t_tok_cast_fp16, cond_emb_cast_fp16, x_emb_cast_fp16))[name = string("hidden_states_1_cast_fp16")]; + int32 var_152 = const()[name = string("op_152"), val = int32(-2)]; + int32 var_154 = const()[name = string("op_154"), val = int32(1)]; + fp16 const_0_promoted_to_fp16 = const()[name = string("const_0_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_167_cast_fp16 = mul(x = hidden_states_1_cast_fp16, y = const_0_promoted_to_fp16)[name = string("op_167_cast_fp16")]; + bool doubled_1_interleave_0 = const()[name = string("doubled_1_interleave_0"), val = bool(false)]; + tensor doubled_1_cast_fp16 = concat(axis = var_154, interleave = doubled_1_interleave_0, values = (hidden_states_1_cast_fp16, var_167_cast_fp16))[name = string("doubled_1_cast_fp16")]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + tensor out_1_gamma_0_to_fp16 = const()[name = string("out_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177875584)))]; + fp16 var_177_to_fp16 = const()[name = string("op_177_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_177_to_fp16, gamma = out_1_gamma_0_to_fp16, x = doubled_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor var_188_split_sizes_0 = const()[name = string("op_188_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_188_axis_0 = const()[name = string("op_188_axis_0"), val = int32(1)]; + tensor var_188_cast_fp16_0, tensor var_188_cast_fp16_1 = split(axis = var_188_axis_0, split_sizes = var_188_split_sizes_0, x = out_1_cast_fp16)[name = string("op_188_cast_fp16")]; + string query_states_1_pad_type_0 = const()[name = string("query_states_1_pad_type_0"), val = string("valid")]; + tensor query_states_1_strides_0 = const()[name = string("query_states_1_strides_0"), val = tensor([1, 1])]; + tensor query_states_1_pad_0 = const()[name = string("query_states_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_1_dilations_0 = const()[name = string("query_states_1_dilations_0"), val = tensor([1, 1])]; + int32 query_states_1_groups_0 = const()[name = string("query_states_1_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177879744)))]; + tensor query_states_1_cast_fp16 = conv(dilations = query_states_1_dilations_0, groups = query_states_1_groups_0, pad = query_states_1_pad_0, pad_type = query_states_1_pad_type_0, strides = query_states_1_strides_0, weight = locdit_decoder_layers_0_self_attn_q_proj_weight_to_fp16, x = var_188_cast_fp16_0)[name = string("query_states_1_cast_fp16")]; + string key_states_1_pad_type_0 = const()[name = string("key_states_1_pad_type_0"), val = string("valid")]; + tensor key_states_1_strides_0 = const()[name = string("key_states_1_strides_0"), val = tensor([1, 1])]; + tensor key_states_1_pad_0 = const()[name = string("key_states_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_1_dilations_0 = const()[name = string("key_states_1_dilations_0"), val = tensor([1, 1])]; + int32 key_states_1_groups_0 = const()[name = string("key_states_1_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182074112)))]; + tensor key_states_1_cast_fp16 = conv(dilations = key_states_1_dilations_0, groups = key_states_1_groups_0, pad = key_states_1_pad_0, pad_type = key_states_1_pad_type_0, strides = key_states_1_strides_0, weight = locdit_decoder_layers_0_self_attn_k_proj_weight_to_fp16, x = var_188_cast_fp16_0)[name = string("key_states_1_cast_fp16")]; + string value_states_1_pad_type_0 = const()[name = string("value_states_1_pad_type_0"), val = string("valid")]; + tensor value_states_1_strides_0 = const()[name = string("value_states_1_strides_0"), val = tensor([1, 1])]; + tensor value_states_1_pad_0 = const()[name = string("value_states_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_1_dilations_0 = const()[name = string("value_states_1_dilations_0"), val = tensor([1, 1])]; + int32 value_states_1_groups_0 = const()[name = string("value_states_1_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182598464)))]; + tensor value_states_1_cast_fp16 = conv(dilations = value_states_1_dilations_0, groups = value_states_1_groups_0, pad = value_states_1_pad_0, pad_type = value_states_1_pad_type_0, strides = value_states_1_strides_0, weight = locdit_decoder_layers_0_self_attn_v_proj_weight_to_fp16, x = var_188_cast_fp16_0)[name = string("value_states_1_cast_fp16")]; + tensor var_215 = const()[name = string("op_215"), val = tensor([2, 16, 128, 11])]; + tensor x_7_cast_fp16 = reshape(shape = var_215, x = query_states_1_cast_fp16)[name = string("x_7_cast_fp16")]; + tensor var_217 = const()[name = string("op_217"), val = tensor([2, 2, 128, 11])]; + tensor x_9_cast_fp16 = reshape(shape = var_217, x = key_states_1_cast_fp16)[name = string("x_9_cast_fp16")]; + tensor var_219 = const()[name = string("op_219"), val = tensor([2, 2, 128, 11])]; + tensor value_states_3_cast_fp16 = reshape(shape = var_219, x = value_states_1_cast_fp16)[name = string("value_states_3_cast_fp16")]; + tensor rope_cos_to_fp16 = const()[name = string("rope_cos_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183122816)))]; + tensor var_221_cast_fp16 = mul(x = x_7_cast_fp16, y = rope_cos_to_fp16)[name = string("op_221_cast_fp16")]; + tensor var_222_split_sizes_0 = const()[name = string("op_222_split_sizes_0"), val = tensor([64, 64])]; + int32 var_222_axis_0 = const()[name = string("op_222_axis_0"), val = int32(-2)]; + tensor var_222_cast_fp16_0, tensor var_222_cast_fp16_1 = split(axis = var_222_axis_0, split_sizes = var_222_split_sizes_0, x = x_7_cast_fp16)[name = string("op_222_cast_fp16")]; + bool var_225_interleave_0 = const()[name = string("op_225_interleave_0"), val = bool(false)]; + tensor var_225_cast_fp16 = concat(axis = var_152, interleave = var_225_interleave_0, values = (var_222_cast_fp16_1, var_222_cast_fp16_0))[name = string("op_225_cast_fp16")]; + tensor rope_sin_to_fp16 = const()[name = string("rope_sin_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183125696)))]; + tensor var_226_cast_fp16 = mul(x = var_225_cast_fp16, y = rope_sin_to_fp16)[name = string("op_226_cast_fp16")]; + tensor query_states_3_cast_fp16 = add(x = var_221_cast_fp16, y = var_226_cast_fp16)[name = string("query_states_3_cast_fp16")]; + tensor var_228_cast_fp16 = mul(x = x_9_cast_fp16, y = rope_cos_to_fp16)[name = string("op_228_cast_fp16")]; + tensor var_229_split_sizes_0 = const()[name = string("op_229_split_sizes_0"), val = tensor([64, 64])]; + int32 var_229_axis_0 = const()[name = string("op_229_axis_0"), val = int32(-2)]; + tensor var_229_cast_fp16_0, tensor var_229_cast_fp16_1 = split(axis = var_229_axis_0, split_sizes = var_229_split_sizes_0, x = x_9_cast_fp16)[name = string("op_229_cast_fp16")]; + bool var_232_interleave_0 = const()[name = string("op_232_interleave_0"), val = bool(false)]; + tensor var_232_cast_fp16 = concat(axis = var_152, interleave = var_232_interleave_0, values = (var_229_cast_fp16_1, var_229_cast_fp16_0))[name = string("op_232_cast_fp16")]; + tensor var_233_cast_fp16 = mul(x = var_232_cast_fp16, y = rope_sin_to_fp16)[name = string("op_233_cast_fp16")]; + tensor key_states_3_cast_fp16 = add(x = var_228_cast_fp16, y = var_233_cast_fp16)[name = string("key_states_3_cast_fp16")]; + tensor var_235_split_sizes_0 = const()[name = string("op_235_split_sizes_0"), val = tensor([8, 8])]; + int32 var_235_axis_0 = const()[name = string("op_235_axis_0"), val = int32(1)]; + tensor var_235_cast_fp16_0, tensor var_235_cast_fp16_1 = split(axis = var_235_axis_0, split_sizes = var_235_split_sizes_0, x = query_states_3_cast_fp16)[name = string("op_235_cast_fp16")]; + tensor var_237_split_sizes_0 = const()[name = string("op_237_split_sizes_0"), val = tensor([1, 1])]; + int32 var_237_axis_0 = const()[name = string("op_237_axis_0"), val = int32(1)]; + tensor var_237_cast_fp16_0, tensor var_237_cast_fp16_1 = split(axis = var_237_axis_0, split_sizes = var_237_split_sizes_0, x = key_states_3_cast_fp16)[name = string("op_237_cast_fp16")]; + tensor var_239_split_sizes_0 = const()[name = string("op_239_split_sizes_0"), val = tensor([1, 1])]; + int32 var_239_axis_0 = const()[name = string("op_239_axis_0"), val = int32(1)]; + tensor var_239_cast_fp16_0, tensor var_239_cast_fp16_1 = split(axis = var_239_axis_0, split_sizes = var_239_split_sizes_0, x = value_states_3_cast_fp16)[name = string("op_239_cast_fp16")]; + bool attn_weights_1_transpose_x_1 = const()[name = string("attn_weights_1_transpose_x_1"), val = bool(true)]; + bool attn_weights_1_transpose_y_1 = const()[name = string("attn_weights_1_transpose_y_1"), val = bool(false)]; + tensor attn_weights_1_cast_fp16 = matmul(transpose_x = attn_weights_1_transpose_x_1, transpose_y = attn_weights_1_transpose_y_1, x = var_237_cast_fp16_0, y = var_235_cast_fp16_0)[name = string("attn_weights_1_cast_fp16")]; + fp16 var_243_to_fp16 = const()[name = string("op_243_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_3_cast_fp16 = mul(x = attn_weights_1_cast_fp16, y = var_243_to_fp16)[name = string("attn_weights_3_cast_fp16")]; + tensor attn_weights_5_cast_fp16 = softmax(axis = var_152, x = attn_weights_3_cast_fp16)[name = string("attn_weights_5_cast_fp16")]; + bool var_246_transpose_x_0 = const()[name = string("op_246_transpose_x_0"), val = bool(false)]; + bool var_246_transpose_y_0 = const()[name = string("op_246_transpose_y_0"), val = bool(false)]; + tensor var_246_cast_fp16 = matmul(transpose_x = var_246_transpose_x_0, transpose_y = var_246_transpose_y_0, x = var_239_cast_fp16_0, y = attn_weights_5_cast_fp16)[name = string("op_246_cast_fp16")]; + bool attn_weights_7_transpose_x_1 = const()[name = string("attn_weights_7_transpose_x_1"), val = bool(true)]; + bool attn_weights_7_transpose_y_1 = const()[name = string("attn_weights_7_transpose_y_1"), val = bool(false)]; + tensor attn_weights_7_cast_fp16 = matmul(transpose_x = attn_weights_7_transpose_x_1, transpose_y = attn_weights_7_transpose_y_1, x = var_237_cast_fp16_1, y = var_235_cast_fp16_1)[name = string("attn_weights_7_cast_fp16")]; + fp16 var_249_to_fp16 = const()[name = string("op_249_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_9_cast_fp16 = mul(x = attn_weights_7_cast_fp16, y = var_249_to_fp16)[name = string("attn_weights_9_cast_fp16")]; + tensor attn_weights_11_cast_fp16 = softmax(axis = var_152, x = attn_weights_9_cast_fp16)[name = string("attn_weights_11_cast_fp16")]; + bool attn_out_1_transpose_x_0 = const()[name = string("attn_out_1_transpose_x_0"), val = bool(false)]; + bool attn_out_1_transpose_y_0 = const()[name = string("attn_out_1_transpose_y_0"), val = bool(false)]; + tensor attn_out_1_cast_fp16 = matmul(transpose_x = attn_out_1_transpose_x_0, transpose_y = attn_out_1_transpose_y_0, x = var_239_cast_fp16_1, y = attn_weights_11_cast_fp16)[name = string("attn_out_1_cast_fp16")]; + bool attn_output_1_interleave_0 = const()[name = string("attn_output_1_interleave_0"), val = bool(false)]; + tensor attn_output_1_cast_fp16 = concat(axis = var_154, interleave = attn_output_1_interleave_0, values = (var_246_cast_fp16, attn_out_1_cast_fp16))[name = string("attn_output_1_cast_fp16")]; + tensor var_255 = const()[name = string("op_255"), val = tensor([2, 2048, 1, 11])]; + tensor x_11_cast_fp16 = reshape(shape = var_255, x = attn_output_1_cast_fp16)[name = string("x_11_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183128576)))]; + tensor hidden_states_5_cast_fp16 = conv(dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = locdit_decoder_layers_0_self_attn_o_proj_weight_to_fp16, x = x_11_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor hidden_states_7_cast_fp16 = add(x = hidden_states_1_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + fp16 const_4_promoted_to_fp16 = const()[name = string("const_4_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_265_cast_fp16 = mul(x = hidden_states_7_cast_fp16, y = const_4_promoted_to_fp16)[name = string("op_265_cast_fp16")]; + bool doubled_5_interleave_0 = const()[name = string("doubled_5_interleave_0"), val = bool(false)]; + tensor doubled_5_cast_fp16 = concat(axis = var_154, interleave = doubled_5_interleave_0, values = (hidden_states_7_cast_fp16, var_265_cast_fp16))[name = string("doubled_5_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + tensor out_3_gamma_0_to_fp16 = const()[name = string("out_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187322944)))]; + fp16 var_275_to_fp16 = const()[name = string("op_275_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_275_to_fp16, gamma = out_3_gamma_0_to_fp16, x = doubled_5_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor var_286_split_sizes_0 = const()[name = string("op_286_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_286_axis_0 = const()[name = string("op_286_axis_0"), val = int32(1)]; + tensor var_286_cast_fp16_0, tensor var_286_cast_fp16_1 = split(axis = var_286_axis_0, split_sizes = var_286_split_sizes_0, x = out_3_cast_fp16)[name = string("op_286_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_mlp_gate_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187327104)))]; + tensor input_5_cast_fp16 = conv(dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = locdit_decoder_layers_0_mlp_gate_proj_weight_to_fp16, x = var_286_cast_fp16_0)[name = string("input_5_cast_fp16")]; + tensor var_298_cast_fp16 = silu(x = input_5_cast_fp16)[name = string("op_298_cast_fp16")]; + string var_304_pad_type_0 = const()[name = string("op_304_pad_type_0"), val = string("valid")]; + tensor var_304_strides_0 = const()[name = string("op_304_strides_0"), val = tensor([1, 1])]; + tensor var_304_pad_0 = const()[name = string("op_304_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_304_dilations_0 = const()[name = string("op_304_dilations_0"), val = tensor([1, 1])]; + int32 var_304_groups_0 = const()[name = string("op_304_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_mlp_up_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195715776)))]; + tensor var_304_cast_fp16 = conv(dilations = var_304_dilations_0, groups = var_304_groups_0, pad = var_304_pad_0, pad_type = var_304_pad_type_0, strides = var_304_strides_0, weight = locdit_decoder_layers_0_mlp_up_proj_weight_to_fp16, x = var_286_cast_fp16_0)[name = string("op_304_cast_fp16")]; + tensor x_15_cast_fp16 = mul(x = var_298_cast_fp16, y = var_304_cast_fp16)[name = string("x_15_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_0_mlp_down_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_0_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204104448)))]; + tensor hidden_states_9_cast_fp16 = conv(dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = locdit_decoder_layers_0_mlp_down_proj_weight_to_fp16, x = x_15_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor hidden_states_11_cast_fp16 = add(x = hidden_states_7_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + int32 var_315 = const()[name = string("op_315"), val = int32(-2)]; + int32 var_317 = const()[name = string("op_317"), val = int32(1)]; + fp16 const_6_promoted_to_fp16 = const()[name = string("const_6_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_330_cast_fp16 = mul(x = hidden_states_11_cast_fp16, y = const_6_promoted_to_fp16)[name = string("op_330_cast_fp16")]; + bool doubled_9_interleave_0 = const()[name = string("doubled_9_interleave_0"), val = bool(false)]; + tensor doubled_9_cast_fp16 = concat(axis = var_317, interleave = doubled_9_interleave_0, values = (hidden_states_11_cast_fp16, var_330_cast_fp16))[name = string("doubled_9_cast_fp16")]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + tensor out_5_gamma_0_to_fp16 = const()[name = string("out_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212493120)))]; + fp16 var_340_to_fp16 = const()[name = string("op_340_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_340_to_fp16, gamma = out_5_gamma_0_to_fp16, x = doubled_9_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor var_351_split_sizes_0 = const()[name = string("op_351_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_351_axis_0 = const()[name = string("op_351_axis_0"), val = int32(1)]; + tensor var_351_cast_fp16_0, tensor var_351_cast_fp16_1 = split(axis = var_351_axis_0, split_sizes = var_351_split_sizes_0, x = out_5_cast_fp16)[name = string("op_351_cast_fp16")]; + string query_states_5_pad_type_0 = const()[name = string("query_states_5_pad_type_0"), val = string("valid")]; + tensor query_states_5_strides_0 = const()[name = string("query_states_5_strides_0"), val = tensor([1, 1])]; + tensor query_states_5_pad_0 = const()[name = string("query_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_5_dilations_0 = const()[name = string("query_states_5_dilations_0"), val = tensor([1, 1])]; + int32 query_states_5_groups_0 = const()[name = string("query_states_5_groups_0"), val = int32(1)]; + tensor query_states_5_cast_fp16 = conv(dilations = query_states_5_dilations_0, groups = query_states_5_groups_0, pad = query_states_5_pad_0, pad_type = query_states_5_pad_type_0, strides = query_states_5_strides_0, weight = locdit_decoder_layers_1_self_attn_q_proj_weight_cast_fp16, x = var_351_cast_fp16_0)[name = string("query_states_5_cast_fp16")]; + string key_states_5_pad_type_0 = const()[name = string("key_states_5_pad_type_0"), val = string("valid")]; + tensor key_states_5_strides_0 = const()[name = string("key_states_5_strides_0"), val = tensor([1, 1])]; + tensor key_states_5_pad_0 = const()[name = string("key_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_5_dilations_0 = const()[name = string("key_states_5_dilations_0"), val = tensor([1, 1])]; + int32 key_states_5_groups_0 = const()[name = string("key_states_5_groups_0"), val = int32(1)]; + tensor key_states_5_cast_fp16 = conv(dilations = key_states_5_dilations_0, groups = key_states_5_groups_0, pad = key_states_5_pad_0, pad_type = key_states_5_pad_type_0, strides = key_states_5_strides_0, weight = locdit_decoder_layers_1_self_attn_k_proj_weight_cast_fp16, x = var_351_cast_fp16_0)[name = string("key_states_5_cast_fp16")]; + string value_states_5_pad_type_0 = const()[name = string("value_states_5_pad_type_0"), val = string("valid")]; + tensor value_states_5_strides_0 = const()[name = string("value_states_5_strides_0"), val = tensor([1, 1])]; + tensor value_states_5_pad_0 = const()[name = string("value_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_5_dilations_0 = const()[name = string("value_states_5_dilations_0"), val = tensor([1, 1])]; + int32 value_states_5_groups_0 = const()[name = string("value_states_5_groups_0"), val = int32(1)]; + tensor value_states_5_cast_fp16 = conv(dilations = value_states_5_dilations_0, groups = value_states_5_groups_0, pad = value_states_5_pad_0, pad_type = value_states_5_pad_type_0, strides = value_states_5_strides_0, weight = locdit_decoder_layers_1_self_attn_v_proj_weight_cast_fp16, x = var_351_cast_fp16_0)[name = string("value_states_5_cast_fp16")]; + tensor var_378 = const()[name = string("op_378"), val = tensor([2, 16, 128, 11])]; + tensor x_17_cast_fp16 = reshape(shape = var_378, x = query_states_5_cast_fp16)[name = string("x_17_cast_fp16")]; + tensor var_380 = const()[name = string("op_380"), val = tensor([2, 2, 128, 11])]; + tensor x_19_cast_fp16 = reshape(shape = var_380, x = key_states_5_cast_fp16)[name = string("x_19_cast_fp16")]; + tensor var_382 = const()[name = string("op_382"), val = tensor([2, 2, 128, 11])]; + tensor value_states_7_cast_fp16 = reshape(shape = var_382, x = value_states_5_cast_fp16)[name = string("value_states_7_cast_fp16")]; + tensor var_384_cast_fp16 = mul(x = x_17_cast_fp16, y = rope_cos_to_fp16)[name = string("op_384_cast_fp16")]; + tensor var_385_split_sizes_0 = const()[name = string("op_385_split_sizes_0"), val = tensor([64, 64])]; + int32 var_385_axis_0 = const()[name = string("op_385_axis_0"), val = int32(-2)]; + tensor var_385_cast_fp16_0, tensor var_385_cast_fp16_1 = split(axis = var_385_axis_0, split_sizes = var_385_split_sizes_0, x = x_17_cast_fp16)[name = string("op_385_cast_fp16")]; + bool var_388_interleave_0 = const()[name = string("op_388_interleave_0"), val = bool(false)]; + tensor var_388_cast_fp16 = concat(axis = var_315, interleave = var_388_interleave_0, values = (var_385_cast_fp16_1, var_385_cast_fp16_0))[name = string("op_388_cast_fp16")]; + tensor var_389_cast_fp16 = mul(x = var_388_cast_fp16, y = rope_sin_to_fp16)[name = string("op_389_cast_fp16")]; + tensor query_states_7_cast_fp16 = add(x = var_384_cast_fp16, y = var_389_cast_fp16)[name = string("query_states_7_cast_fp16")]; + tensor var_391_cast_fp16 = mul(x = x_19_cast_fp16, y = rope_cos_to_fp16)[name = string("op_391_cast_fp16")]; + tensor var_392_split_sizes_0 = const()[name = string("op_392_split_sizes_0"), val = tensor([64, 64])]; + int32 var_392_axis_0 = const()[name = string("op_392_axis_0"), val = int32(-2)]; + tensor var_392_cast_fp16_0, tensor var_392_cast_fp16_1 = split(axis = var_392_axis_0, split_sizes = var_392_split_sizes_0, x = x_19_cast_fp16)[name = string("op_392_cast_fp16")]; + bool var_395_interleave_0 = const()[name = string("op_395_interleave_0"), val = bool(false)]; + tensor var_395_cast_fp16 = concat(axis = var_315, interleave = var_395_interleave_0, values = (var_392_cast_fp16_1, var_392_cast_fp16_0))[name = string("op_395_cast_fp16")]; + tensor var_396_cast_fp16 = mul(x = var_395_cast_fp16, y = rope_sin_to_fp16)[name = string("op_396_cast_fp16")]; + tensor key_states_7_cast_fp16 = add(x = var_391_cast_fp16, y = var_396_cast_fp16)[name = string("key_states_7_cast_fp16")]; + tensor var_398_split_sizes_0 = const()[name = string("op_398_split_sizes_0"), val = tensor([8, 8])]; + int32 var_398_axis_0 = const()[name = string("op_398_axis_0"), val = int32(1)]; + tensor var_398_cast_fp16_0, tensor var_398_cast_fp16_1 = split(axis = var_398_axis_0, split_sizes = var_398_split_sizes_0, x = query_states_7_cast_fp16)[name = string("op_398_cast_fp16")]; + tensor var_400_split_sizes_0 = const()[name = string("op_400_split_sizes_0"), val = tensor([1, 1])]; + int32 var_400_axis_0 = const()[name = string("op_400_axis_0"), val = int32(1)]; + tensor var_400_cast_fp16_0, tensor var_400_cast_fp16_1 = split(axis = var_400_axis_0, split_sizes = var_400_split_sizes_0, x = key_states_7_cast_fp16)[name = string("op_400_cast_fp16")]; + tensor var_402_split_sizes_0 = const()[name = string("op_402_split_sizes_0"), val = tensor([1, 1])]; + int32 var_402_axis_0 = const()[name = string("op_402_axis_0"), val = int32(1)]; + tensor var_402_cast_fp16_0, tensor var_402_cast_fp16_1 = split(axis = var_402_axis_0, split_sizes = var_402_split_sizes_0, x = value_states_7_cast_fp16)[name = string("op_402_cast_fp16")]; + bool attn_weights_13_transpose_x_1 = const()[name = string("attn_weights_13_transpose_x_1"), val = bool(true)]; + bool attn_weights_13_transpose_y_1 = const()[name = string("attn_weights_13_transpose_y_1"), val = bool(false)]; + tensor attn_weights_13_cast_fp16 = matmul(transpose_x = attn_weights_13_transpose_x_1, transpose_y = attn_weights_13_transpose_y_1, x = var_400_cast_fp16_0, y = var_398_cast_fp16_0)[name = string("attn_weights_13_cast_fp16")]; + fp16 var_406_to_fp16 = const()[name = string("op_406_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_15_cast_fp16 = mul(x = attn_weights_13_cast_fp16, y = var_406_to_fp16)[name = string("attn_weights_15_cast_fp16")]; + tensor attn_weights_17_cast_fp16 = softmax(axis = var_315, x = attn_weights_15_cast_fp16)[name = string("attn_weights_17_cast_fp16")]; + bool var_409_transpose_x_0 = const()[name = string("op_409_transpose_x_0"), val = bool(false)]; + bool var_409_transpose_y_0 = const()[name = string("op_409_transpose_y_0"), val = bool(false)]; + tensor var_409_cast_fp16 = matmul(transpose_x = var_409_transpose_x_0, transpose_y = var_409_transpose_y_0, x = var_402_cast_fp16_0, y = attn_weights_17_cast_fp16)[name = string("op_409_cast_fp16")]; + bool attn_weights_19_transpose_x_1 = const()[name = string("attn_weights_19_transpose_x_1"), val = bool(true)]; + bool attn_weights_19_transpose_y_1 = const()[name = string("attn_weights_19_transpose_y_1"), val = bool(false)]; + tensor attn_weights_19_cast_fp16 = matmul(transpose_x = attn_weights_19_transpose_x_1, transpose_y = attn_weights_19_transpose_y_1, x = var_400_cast_fp16_1, y = var_398_cast_fp16_1)[name = string("attn_weights_19_cast_fp16")]; + fp16 var_412_to_fp16 = const()[name = string("op_412_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_21_cast_fp16 = mul(x = attn_weights_19_cast_fp16, y = var_412_to_fp16)[name = string("attn_weights_21_cast_fp16")]; + tensor attn_weights_23_cast_fp16 = softmax(axis = var_315, x = attn_weights_21_cast_fp16)[name = string("attn_weights_23_cast_fp16")]; + bool attn_out_3_transpose_x_0 = const()[name = string("attn_out_3_transpose_x_0"), val = bool(false)]; + bool attn_out_3_transpose_y_0 = const()[name = string("attn_out_3_transpose_y_0"), val = bool(false)]; + tensor attn_out_3_cast_fp16 = matmul(transpose_x = attn_out_3_transpose_x_0, transpose_y = attn_out_3_transpose_y_0, x = var_402_cast_fp16_1, y = attn_weights_23_cast_fp16)[name = string("attn_out_3_cast_fp16")]; + bool attn_output_3_interleave_0 = const()[name = string("attn_output_3_interleave_0"), val = bool(false)]; + tensor attn_output_3_cast_fp16 = concat(axis = var_317, interleave = attn_output_3_interleave_0, values = (var_409_cast_fp16, attn_out_3_cast_fp16))[name = string("attn_output_3_cast_fp16")]; + tensor var_418 = const()[name = string("op_418"), val = tensor([2, 2048, 1, 11])]; + tensor x_21_cast_fp16 = reshape(shape = var_418, x = attn_output_3_cast_fp16)[name = string("x_21_cast_fp16")]; + string hidden_states_15_pad_type_0 = const()[name = string("hidden_states_15_pad_type_0"), val = string("valid")]; + tensor hidden_states_15_strides_0 = const()[name = string("hidden_states_15_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_15_pad_0 = const()[name = string("hidden_states_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_15_dilations_0 = const()[name = string("hidden_states_15_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_15_groups_0 = const()[name = string("hidden_states_15_groups_0"), val = int32(1)]; + tensor hidden_states_15_cast_fp16 = conv(dilations = hidden_states_15_dilations_0, groups = hidden_states_15_groups_0, pad = hidden_states_15_pad_0, pad_type = hidden_states_15_pad_type_0, strides = hidden_states_15_strides_0, weight = locdit_decoder_layers_1_self_attn_o_proj_weight_cast_fp16, x = x_21_cast_fp16)[name = string("hidden_states_15_cast_fp16")]; + tensor hidden_states_17_cast_fp16 = add(x = hidden_states_11_cast_fp16, y = hidden_states_15_cast_fp16)[name = string("hidden_states_17_cast_fp16")]; + fp16 const_10_promoted_to_fp16 = const()[name = string("const_10_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_428_cast_fp16 = mul(x = hidden_states_17_cast_fp16, y = const_10_promoted_to_fp16)[name = string("op_428_cast_fp16")]; + bool doubled_13_interleave_0 = const()[name = string("doubled_13_interleave_0"), val = bool(false)]; + tensor doubled_13_cast_fp16 = concat(axis = var_317, interleave = doubled_13_interleave_0, values = (hidden_states_17_cast_fp16, var_428_cast_fp16))[name = string("doubled_13_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + tensor out_7_gamma_0_to_fp16 = const()[name = string("out_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212497280)))]; + fp16 var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_438_to_fp16, gamma = out_7_gamma_0_to_fp16, x = doubled_13_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor var_449_split_sizes_0 = const()[name = string("op_449_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_449_axis_0 = const()[name = string("op_449_axis_0"), val = int32(1)]; + tensor var_449_cast_fp16_0, tensor var_449_cast_fp16_1 = split(axis = var_449_axis_0, split_sizes = var_449_split_sizes_0, x = out_7_cast_fp16)[name = string("op_449_cast_fp16")]; + string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("valid")]; + tensor input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor([1, 1])]; + int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(1)]; + tensor input_7_cast_fp16 = conv(dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = locdit_decoder_layers_1_mlp_gate_proj_weight_cast_fp16, x = var_449_cast_fp16_0)[name = string("input_7_cast_fp16")]; + tensor var_461_cast_fp16 = silu(x = input_7_cast_fp16)[name = string("op_461_cast_fp16")]; + string var_467_pad_type_0 = const()[name = string("op_467_pad_type_0"), val = string("valid")]; + tensor var_467_strides_0 = const()[name = string("op_467_strides_0"), val = tensor([1, 1])]; + tensor var_467_pad_0 = const()[name = string("op_467_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_467_dilations_0 = const()[name = string("op_467_dilations_0"), val = tensor([1, 1])]; + int32 var_467_groups_0 = const()[name = string("op_467_groups_0"), val = int32(1)]; + tensor var_467_cast_fp16 = conv(dilations = var_467_dilations_0, groups = var_467_groups_0, pad = var_467_pad_0, pad_type = var_467_pad_type_0, strides = var_467_strides_0, weight = locdit_decoder_layers_1_mlp_up_proj_weight_cast_fp16, x = var_449_cast_fp16_0)[name = string("op_467_cast_fp16")]; + tensor x_25_cast_fp16 = mul(x = var_461_cast_fp16, y = var_467_cast_fp16)[name = string("x_25_cast_fp16")]; + string hidden_states_19_pad_type_0 = const()[name = string("hidden_states_19_pad_type_0"), val = string("valid")]; + tensor hidden_states_19_strides_0 = const()[name = string("hidden_states_19_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_19_pad_0 = const()[name = string("hidden_states_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_19_dilations_0 = const()[name = string("hidden_states_19_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_19_groups_0 = const()[name = string("hidden_states_19_groups_0"), val = int32(1)]; + tensor hidden_states_19_cast_fp16 = conv(dilations = hidden_states_19_dilations_0, groups = hidden_states_19_groups_0, pad = hidden_states_19_pad_0, pad_type = hidden_states_19_pad_type_0, strides = hidden_states_19_strides_0, weight = locdit_decoder_layers_1_mlp_down_proj_weight_cast_fp16, x = x_25_cast_fp16)[name = string("hidden_states_19_cast_fp16")]; + tensor hidden_states_21_cast_fp16 = add(x = hidden_states_17_cast_fp16, y = hidden_states_19_cast_fp16)[name = string("hidden_states_21_cast_fp16")]; + int32 var_478 = const()[name = string("op_478"), val = int32(-2)]; + int32 var_480 = const()[name = string("op_480"), val = int32(1)]; + fp16 const_12_promoted_to_fp16 = const()[name = string("const_12_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_493_cast_fp16 = mul(x = hidden_states_21_cast_fp16, y = const_12_promoted_to_fp16)[name = string("op_493_cast_fp16")]; + bool doubled_17_interleave_0 = const()[name = string("doubled_17_interleave_0"), val = bool(false)]; + tensor doubled_17_cast_fp16 = concat(axis = var_480, interleave = doubled_17_interleave_0, values = (hidden_states_21_cast_fp16, var_493_cast_fp16))[name = string("doubled_17_cast_fp16")]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + tensor out_9_gamma_0_to_fp16 = const()[name = string("out_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212501440)))]; + fp16 var_503_to_fp16 = const()[name = string("op_503_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_503_to_fp16, gamma = out_9_gamma_0_to_fp16, x = doubled_17_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor var_514_split_sizes_0 = const()[name = string("op_514_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_514_axis_0 = const()[name = string("op_514_axis_0"), val = int32(1)]; + tensor var_514_cast_fp16_0, tensor var_514_cast_fp16_1 = split(axis = var_514_axis_0, split_sizes = var_514_split_sizes_0, x = out_9_cast_fp16)[name = string("op_514_cast_fp16")]; + string query_states_9_pad_type_0 = const()[name = string("query_states_9_pad_type_0"), val = string("valid")]; + tensor query_states_9_strides_0 = const()[name = string("query_states_9_strides_0"), val = tensor([1, 1])]; + tensor query_states_9_pad_0 = const()[name = string("query_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_9_dilations_0 = const()[name = string("query_states_9_dilations_0"), val = tensor([1, 1])]; + int32 query_states_9_groups_0 = const()[name = string("query_states_9_groups_0"), val = int32(1)]; + tensor query_states_9_cast_fp16 = conv(dilations = query_states_9_dilations_0, groups = query_states_9_groups_0, pad = query_states_9_pad_0, pad_type = query_states_9_pad_type_0, strides = query_states_9_strides_0, weight = locdit_decoder_layers_2_self_attn_q_proj_weight_cast_fp16, x = var_514_cast_fp16_0)[name = string("query_states_9_cast_fp16")]; + string key_states_9_pad_type_0 = const()[name = string("key_states_9_pad_type_0"), val = string("valid")]; + tensor key_states_9_strides_0 = const()[name = string("key_states_9_strides_0"), val = tensor([1, 1])]; + tensor key_states_9_pad_0 = const()[name = string("key_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_9_dilations_0 = const()[name = string("key_states_9_dilations_0"), val = tensor([1, 1])]; + int32 key_states_9_groups_0 = const()[name = string("key_states_9_groups_0"), val = int32(1)]; + tensor key_states_9_cast_fp16 = conv(dilations = key_states_9_dilations_0, groups = key_states_9_groups_0, pad = key_states_9_pad_0, pad_type = key_states_9_pad_type_0, strides = key_states_9_strides_0, weight = locdit_decoder_layers_2_self_attn_k_proj_weight_cast_fp16, x = var_514_cast_fp16_0)[name = string("key_states_9_cast_fp16")]; + string value_states_9_pad_type_0 = const()[name = string("value_states_9_pad_type_0"), val = string("valid")]; + tensor value_states_9_strides_0 = const()[name = string("value_states_9_strides_0"), val = tensor([1, 1])]; + tensor value_states_9_pad_0 = const()[name = string("value_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_9_dilations_0 = const()[name = string("value_states_9_dilations_0"), val = tensor([1, 1])]; + int32 value_states_9_groups_0 = const()[name = string("value_states_9_groups_0"), val = int32(1)]; + tensor value_states_9_cast_fp16 = conv(dilations = value_states_9_dilations_0, groups = value_states_9_groups_0, pad = value_states_9_pad_0, pad_type = value_states_9_pad_type_0, strides = value_states_9_strides_0, weight = locdit_decoder_layers_2_self_attn_v_proj_weight_cast_fp16, x = var_514_cast_fp16_0)[name = string("value_states_9_cast_fp16")]; + tensor var_541 = const()[name = string("op_541"), val = tensor([2, 16, 128, 11])]; + tensor x_27_cast_fp16 = reshape(shape = var_541, x = query_states_9_cast_fp16)[name = string("x_27_cast_fp16")]; + tensor var_543 = const()[name = string("op_543"), val = tensor([2, 2, 128, 11])]; + tensor x_29_cast_fp16 = reshape(shape = var_543, x = key_states_9_cast_fp16)[name = string("x_29_cast_fp16")]; + tensor var_545 = const()[name = string("op_545"), val = tensor([2, 2, 128, 11])]; + tensor value_states_11_cast_fp16 = reshape(shape = var_545, x = value_states_9_cast_fp16)[name = string("value_states_11_cast_fp16")]; + tensor var_547_cast_fp16 = mul(x = x_27_cast_fp16, y = rope_cos_to_fp16)[name = string("op_547_cast_fp16")]; + tensor var_548_split_sizes_0 = const()[name = string("op_548_split_sizes_0"), val = tensor([64, 64])]; + int32 var_548_axis_0 = const()[name = string("op_548_axis_0"), val = int32(-2)]; + tensor var_548_cast_fp16_0, tensor var_548_cast_fp16_1 = split(axis = var_548_axis_0, split_sizes = var_548_split_sizes_0, x = x_27_cast_fp16)[name = string("op_548_cast_fp16")]; + bool var_551_interleave_0 = const()[name = string("op_551_interleave_0"), val = bool(false)]; + tensor var_551_cast_fp16 = concat(axis = var_478, interleave = var_551_interleave_0, values = (var_548_cast_fp16_1, var_548_cast_fp16_0))[name = string("op_551_cast_fp16")]; + tensor var_552_cast_fp16 = mul(x = var_551_cast_fp16, y = rope_sin_to_fp16)[name = string("op_552_cast_fp16")]; + tensor query_states_11_cast_fp16 = add(x = var_547_cast_fp16, y = var_552_cast_fp16)[name = string("query_states_11_cast_fp16")]; + tensor var_554_cast_fp16 = mul(x = x_29_cast_fp16, y = rope_cos_to_fp16)[name = string("op_554_cast_fp16")]; + tensor var_555_split_sizes_0 = const()[name = string("op_555_split_sizes_0"), val = tensor([64, 64])]; + int32 var_555_axis_0 = const()[name = string("op_555_axis_0"), val = int32(-2)]; + tensor var_555_cast_fp16_0, tensor var_555_cast_fp16_1 = split(axis = var_555_axis_0, split_sizes = var_555_split_sizes_0, x = x_29_cast_fp16)[name = string("op_555_cast_fp16")]; + bool var_558_interleave_0 = const()[name = string("op_558_interleave_0"), val = bool(false)]; + tensor var_558_cast_fp16 = concat(axis = var_478, interleave = var_558_interleave_0, values = (var_555_cast_fp16_1, var_555_cast_fp16_0))[name = string("op_558_cast_fp16")]; + tensor var_559_cast_fp16 = mul(x = var_558_cast_fp16, y = rope_sin_to_fp16)[name = string("op_559_cast_fp16")]; + tensor key_states_11_cast_fp16 = add(x = var_554_cast_fp16, y = var_559_cast_fp16)[name = string("key_states_11_cast_fp16")]; + tensor var_561_split_sizes_0 = const()[name = string("op_561_split_sizes_0"), val = tensor([8, 8])]; + int32 var_561_axis_0 = const()[name = string("op_561_axis_0"), val = int32(1)]; + tensor var_561_cast_fp16_0, tensor var_561_cast_fp16_1 = split(axis = var_561_axis_0, split_sizes = var_561_split_sizes_0, x = query_states_11_cast_fp16)[name = string("op_561_cast_fp16")]; + tensor var_563_split_sizes_0 = const()[name = string("op_563_split_sizes_0"), val = tensor([1, 1])]; + int32 var_563_axis_0 = const()[name = string("op_563_axis_0"), val = int32(1)]; + tensor var_563_cast_fp16_0, tensor var_563_cast_fp16_1 = split(axis = var_563_axis_0, split_sizes = var_563_split_sizes_0, x = key_states_11_cast_fp16)[name = string("op_563_cast_fp16")]; + tensor var_565_split_sizes_0 = const()[name = string("op_565_split_sizes_0"), val = tensor([1, 1])]; + int32 var_565_axis_0 = const()[name = string("op_565_axis_0"), val = int32(1)]; + tensor var_565_cast_fp16_0, tensor var_565_cast_fp16_1 = split(axis = var_565_axis_0, split_sizes = var_565_split_sizes_0, x = value_states_11_cast_fp16)[name = string("op_565_cast_fp16")]; + bool attn_weights_25_transpose_x_1 = const()[name = string("attn_weights_25_transpose_x_1"), val = bool(true)]; + bool attn_weights_25_transpose_y_1 = const()[name = string("attn_weights_25_transpose_y_1"), val = bool(false)]; + tensor attn_weights_25_cast_fp16 = matmul(transpose_x = attn_weights_25_transpose_x_1, transpose_y = attn_weights_25_transpose_y_1, x = var_563_cast_fp16_0, y = var_561_cast_fp16_0)[name = string("attn_weights_25_cast_fp16")]; + fp16 var_569_to_fp16 = const()[name = string("op_569_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_27_cast_fp16 = mul(x = attn_weights_25_cast_fp16, y = var_569_to_fp16)[name = string("attn_weights_27_cast_fp16")]; + tensor attn_weights_29_cast_fp16 = softmax(axis = var_478, x = attn_weights_27_cast_fp16)[name = string("attn_weights_29_cast_fp16")]; + bool var_572_transpose_x_0 = const()[name = string("op_572_transpose_x_0"), val = bool(false)]; + bool var_572_transpose_y_0 = const()[name = string("op_572_transpose_y_0"), val = bool(false)]; + tensor var_572_cast_fp16 = matmul(transpose_x = var_572_transpose_x_0, transpose_y = var_572_transpose_y_0, x = var_565_cast_fp16_0, y = attn_weights_29_cast_fp16)[name = string("op_572_cast_fp16")]; + bool attn_weights_31_transpose_x_1 = const()[name = string("attn_weights_31_transpose_x_1"), val = bool(true)]; + bool attn_weights_31_transpose_y_1 = const()[name = string("attn_weights_31_transpose_y_1"), val = bool(false)]; + tensor attn_weights_31_cast_fp16 = matmul(transpose_x = attn_weights_31_transpose_x_1, transpose_y = attn_weights_31_transpose_y_1, x = var_563_cast_fp16_1, y = var_561_cast_fp16_1)[name = string("attn_weights_31_cast_fp16")]; + fp16 var_575_to_fp16 = const()[name = string("op_575_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_33_cast_fp16 = mul(x = attn_weights_31_cast_fp16, y = var_575_to_fp16)[name = string("attn_weights_33_cast_fp16")]; + tensor attn_weights_35_cast_fp16 = softmax(axis = var_478, x = attn_weights_33_cast_fp16)[name = string("attn_weights_35_cast_fp16")]; + bool attn_out_5_transpose_x_0 = const()[name = string("attn_out_5_transpose_x_0"), val = bool(false)]; + bool attn_out_5_transpose_y_0 = const()[name = string("attn_out_5_transpose_y_0"), val = bool(false)]; + tensor attn_out_5_cast_fp16 = matmul(transpose_x = attn_out_5_transpose_x_0, transpose_y = attn_out_5_transpose_y_0, x = var_565_cast_fp16_1, y = attn_weights_35_cast_fp16)[name = string("attn_out_5_cast_fp16")]; + bool attn_output_5_interleave_0 = const()[name = string("attn_output_5_interleave_0"), val = bool(false)]; + tensor attn_output_5_cast_fp16 = concat(axis = var_480, interleave = attn_output_5_interleave_0, values = (var_572_cast_fp16, attn_out_5_cast_fp16))[name = string("attn_output_5_cast_fp16")]; + tensor var_581 = const()[name = string("op_581"), val = tensor([2, 2048, 1, 11])]; + tensor x_31_cast_fp16 = reshape(shape = var_581, x = attn_output_5_cast_fp16)[name = string("x_31_cast_fp16")]; + string hidden_states_25_pad_type_0 = const()[name = string("hidden_states_25_pad_type_0"), val = string("valid")]; + tensor hidden_states_25_strides_0 = const()[name = string("hidden_states_25_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_25_pad_0 = const()[name = string("hidden_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_25_dilations_0 = const()[name = string("hidden_states_25_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_25_groups_0 = const()[name = string("hidden_states_25_groups_0"), val = int32(1)]; + tensor hidden_states_25_cast_fp16 = conv(dilations = hidden_states_25_dilations_0, groups = hidden_states_25_groups_0, pad = hidden_states_25_pad_0, pad_type = hidden_states_25_pad_type_0, strides = hidden_states_25_strides_0, weight = locdit_decoder_layers_2_self_attn_o_proj_weight_cast_fp16, x = x_31_cast_fp16)[name = string("hidden_states_25_cast_fp16")]; + tensor hidden_states_27_cast_fp16 = add(x = hidden_states_21_cast_fp16, y = hidden_states_25_cast_fp16)[name = string("hidden_states_27_cast_fp16")]; + fp16 const_16_promoted_to_fp16 = const()[name = string("const_16_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_591_cast_fp16 = mul(x = hidden_states_27_cast_fp16, y = const_16_promoted_to_fp16)[name = string("op_591_cast_fp16")]; + bool doubled_21_interleave_0 = const()[name = string("doubled_21_interleave_0"), val = bool(false)]; + tensor doubled_21_cast_fp16 = concat(axis = var_480, interleave = doubled_21_interleave_0, values = (hidden_states_27_cast_fp16, var_591_cast_fp16))[name = string("doubled_21_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + tensor out_11_gamma_0_to_fp16 = const()[name = string("out_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212505600)))]; + fp16 var_601_to_fp16 = const()[name = string("op_601_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_601_to_fp16, gamma = out_11_gamma_0_to_fp16, x = doubled_21_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor var_612_split_sizes_0 = const()[name = string("op_612_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_612_axis_0 = const()[name = string("op_612_axis_0"), val = int32(1)]; + tensor var_612_cast_fp16_0, tensor var_612_cast_fp16_1 = split(axis = var_612_axis_0, split_sizes = var_612_split_sizes_0, x = out_11_cast_fp16)[name = string("op_612_cast_fp16")]; + string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")]; + tensor input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor([1, 1])]; + tensor input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor([1, 1])]; + int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)]; + tensor input_9_cast_fp16 = conv(dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = locdit_decoder_layers_2_mlp_gate_proj_weight_cast_fp16, x = var_612_cast_fp16_0)[name = string("input_9_cast_fp16")]; + tensor var_624_cast_fp16 = silu(x = input_9_cast_fp16)[name = string("op_624_cast_fp16")]; + string var_630_pad_type_0 = const()[name = string("op_630_pad_type_0"), val = string("valid")]; + tensor var_630_strides_0 = const()[name = string("op_630_strides_0"), val = tensor([1, 1])]; + tensor var_630_pad_0 = const()[name = string("op_630_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_630_dilations_0 = const()[name = string("op_630_dilations_0"), val = tensor([1, 1])]; + int32 var_630_groups_0 = const()[name = string("op_630_groups_0"), val = int32(1)]; + tensor var_630_cast_fp16 = conv(dilations = var_630_dilations_0, groups = var_630_groups_0, pad = var_630_pad_0, pad_type = var_630_pad_type_0, strides = var_630_strides_0, weight = locdit_decoder_layers_2_mlp_up_proj_weight_cast_fp16, x = var_612_cast_fp16_0)[name = string("op_630_cast_fp16")]; + tensor x_35_cast_fp16 = mul(x = var_624_cast_fp16, y = var_630_cast_fp16)[name = string("x_35_cast_fp16")]; + string hidden_states_29_pad_type_0 = const()[name = string("hidden_states_29_pad_type_0"), val = string("valid")]; + tensor hidden_states_29_strides_0 = const()[name = string("hidden_states_29_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_29_pad_0 = const()[name = string("hidden_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_29_dilations_0 = const()[name = string("hidden_states_29_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_29_groups_0 = const()[name = string("hidden_states_29_groups_0"), val = int32(1)]; + tensor hidden_states_29_cast_fp16 = conv(dilations = hidden_states_29_dilations_0, groups = hidden_states_29_groups_0, pad = hidden_states_29_pad_0, pad_type = hidden_states_29_pad_type_0, strides = hidden_states_29_strides_0, weight = locdit_decoder_layers_2_mlp_down_proj_weight_cast_fp16, x = x_35_cast_fp16)[name = string("hidden_states_29_cast_fp16")]; + tensor hidden_states_31_cast_fp16 = add(x = hidden_states_27_cast_fp16, y = hidden_states_29_cast_fp16)[name = string("hidden_states_31_cast_fp16")]; + int32 var_641 = const()[name = string("op_641"), val = int32(-2)]; + int32 var_643 = const()[name = string("op_643"), val = int32(1)]; + fp16 const_18_promoted_to_fp16 = const()[name = string("const_18_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_656_cast_fp16 = mul(x = hidden_states_31_cast_fp16, y = const_18_promoted_to_fp16)[name = string("op_656_cast_fp16")]; + bool doubled_25_interleave_0 = const()[name = string("doubled_25_interleave_0"), val = bool(false)]; + tensor doubled_25_cast_fp16 = concat(axis = var_643, interleave = doubled_25_interleave_0, values = (hidden_states_31_cast_fp16, var_656_cast_fp16))[name = string("doubled_25_cast_fp16")]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + tensor out_13_gamma_0_to_fp16 = const()[name = string("out_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212509760)))]; + fp16 var_666_to_fp16 = const()[name = string("op_666_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_666_to_fp16, gamma = out_13_gamma_0_to_fp16, x = doubled_25_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor var_677_split_sizes_0 = const()[name = string("op_677_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_677_axis_0 = const()[name = string("op_677_axis_0"), val = int32(1)]; + tensor var_677_cast_fp16_0, tensor var_677_cast_fp16_1 = split(axis = var_677_axis_0, split_sizes = var_677_split_sizes_0, x = out_13_cast_fp16)[name = string("op_677_cast_fp16")]; + string query_states_13_pad_type_0 = const()[name = string("query_states_13_pad_type_0"), val = string("valid")]; + tensor query_states_13_strides_0 = const()[name = string("query_states_13_strides_0"), val = tensor([1, 1])]; + tensor query_states_13_pad_0 = const()[name = string("query_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_13_dilations_0 = const()[name = string("query_states_13_dilations_0"), val = tensor([1, 1])]; + int32 query_states_13_groups_0 = const()[name = string("query_states_13_groups_0"), val = int32(1)]; + tensor query_states_13_cast_fp16 = conv(dilations = query_states_13_dilations_0, groups = query_states_13_groups_0, pad = query_states_13_pad_0, pad_type = query_states_13_pad_type_0, strides = query_states_13_strides_0, weight = locdit_decoder_layers_3_self_attn_q_proj_weight_cast_fp16, x = var_677_cast_fp16_0)[name = string("query_states_13_cast_fp16")]; + string key_states_13_pad_type_0 = const()[name = string("key_states_13_pad_type_0"), val = string("valid")]; + tensor key_states_13_strides_0 = const()[name = string("key_states_13_strides_0"), val = tensor([1, 1])]; + tensor key_states_13_pad_0 = const()[name = string("key_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_13_dilations_0 = const()[name = string("key_states_13_dilations_0"), val = tensor([1, 1])]; + int32 key_states_13_groups_0 = const()[name = string("key_states_13_groups_0"), val = int32(1)]; + tensor key_states_13_cast_fp16 = conv(dilations = key_states_13_dilations_0, groups = key_states_13_groups_0, pad = key_states_13_pad_0, pad_type = key_states_13_pad_type_0, strides = key_states_13_strides_0, weight = locdit_decoder_layers_3_self_attn_k_proj_weight_cast_fp16, x = var_677_cast_fp16_0)[name = string("key_states_13_cast_fp16")]; + string value_states_13_pad_type_0 = const()[name = string("value_states_13_pad_type_0"), val = string("valid")]; + tensor value_states_13_strides_0 = const()[name = string("value_states_13_strides_0"), val = tensor([1, 1])]; + tensor value_states_13_pad_0 = const()[name = string("value_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_13_dilations_0 = const()[name = string("value_states_13_dilations_0"), val = tensor([1, 1])]; + int32 value_states_13_groups_0 = const()[name = string("value_states_13_groups_0"), val = int32(1)]; + tensor value_states_13_cast_fp16 = conv(dilations = value_states_13_dilations_0, groups = value_states_13_groups_0, pad = value_states_13_pad_0, pad_type = value_states_13_pad_type_0, strides = value_states_13_strides_0, weight = locdit_decoder_layers_3_self_attn_v_proj_weight_cast_fp16, x = var_677_cast_fp16_0)[name = string("value_states_13_cast_fp16")]; + tensor var_704 = const()[name = string("op_704"), val = tensor([2, 16, 128, 11])]; + tensor x_37_cast_fp16 = reshape(shape = var_704, x = query_states_13_cast_fp16)[name = string("x_37_cast_fp16")]; + tensor var_706 = const()[name = string("op_706"), val = tensor([2, 2, 128, 11])]; + tensor x_39_cast_fp16 = reshape(shape = var_706, x = key_states_13_cast_fp16)[name = string("x_39_cast_fp16")]; + tensor var_708 = const()[name = string("op_708"), val = tensor([2, 2, 128, 11])]; + tensor value_states_15_cast_fp16 = reshape(shape = var_708, x = value_states_13_cast_fp16)[name = string("value_states_15_cast_fp16")]; + tensor var_710_cast_fp16 = mul(x = x_37_cast_fp16, y = rope_cos_to_fp16)[name = string("op_710_cast_fp16")]; + tensor var_711_split_sizes_0 = const()[name = string("op_711_split_sizes_0"), val = tensor([64, 64])]; + int32 var_711_axis_0 = const()[name = string("op_711_axis_0"), val = int32(-2)]; + tensor var_711_cast_fp16_0, tensor var_711_cast_fp16_1 = split(axis = var_711_axis_0, split_sizes = var_711_split_sizes_0, x = x_37_cast_fp16)[name = string("op_711_cast_fp16")]; + bool var_714_interleave_0 = const()[name = string("op_714_interleave_0"), val = bool(false)]; + tensor var_714_cast_fp16 = concat(axis = var_641, interleave = var_714_interleave_0, values = (var_711_cast_fp16_1, var_711_cast_fp16_0))[name = string("op_714_cast_fp16")]; + tensor var_715_cast_fp16 = mul(x = var_714_cast_fp16, y = rope_sin_to_fp16)[name = string("op_715_cast_fp16")]; + tensor query_states_15_cast_fp16 = add(x = var_710_cast_fp16, y = var_715_cast_fp16)[name = string("query_states_15_cast_fp16")]; + tensor var_717_cast_fp16 = mul(x = x_39_cast_fp16, y = rope_cos_to_fp16)[name = string("op_717_cast_fp16")]; + tensor var_718_split_sizes_0 = const()[name = string("op_718_split_sizes_0"), val = tensor([64, 64])]; + int32 var_718_axis_0 = const()[name = string("op_718_axis_0"), val = int32(-2)]; + tensor var_718_cast_fp16_0, tensor var_718_cast_fp16_1 = split(axis = var_718_axis_0, split_sizes = var_718_split_sizes_0, x = x_39_cast_fp16)[name = string("op_718_cast_fp16")]; + bool var_721_interleave_0 = const()[name = string("op_721_interleave_0"), val = bool(false)]; + tensor var_721_cast_fp16 = concat(axis = var_641, interleave = var_721_interleave_0, values = (var_718_cast_fp16_1, var_718_cast_fp16_0))[name = string("op_721_cast_fp16")]; + tensor var_722_cast_fp16 = mul(x = var_721_cast_fp16, y = rope_sin_to_fp16)[name = string("op_722_cast_fp16")]; + tensor key_states_15_cast_fp16 = add(x = var_717_cast_fp16, y = var_722_cast_fp16)[name = string("key_states_15_cast_fp16")]; + tensor var_724_split_sizes_0 = const()[name = string("op_724_split_sizes_0"), val = tensor([8, 8])]; + int32 var_724_axis_0 = const()[name = string("op_724_axis_0"), val = int32(1)]; + tensor var_724_cast_fp16_0, tensor var_724_cast_fp16_1 = split(axis = var_724_axis_0, split_sizes = var_724_split_sizes_0, x = query_states_15_cast_fp16)[name = string("op_724_cast_fp16")]; + tensor var_726_split_sizes_0 = const()[name = string("op_726_split_sizes_0"), val = tensor([1, 1])]; + int32 var_726_axis_0 = const()[name = string("op_726_axis_0"), val = int32(1)]; + tensor var_726_cast_fp16_0, tensor var_726_cast_fp16_1 = split(axis = var_726_axis_0, split_sizes = var_726_split_sizes_0, x = key_states_15_cast_fp16)[name = string("op_726_cast_fp16")]; + tensor var_728_split_sizes_0 = const()[name = string("op_728_split_sizes_0"), val = tensor([1, 1])]; + int32 var_728_axis_0 = const()[name = string("op_728_axis_0"), val = int32(1)]; + tensor var_728_cast_fp16_0, tensor var_728_cast_fp16_1 = split(axis = var_728_axis_0, split_sizes = var_728_split_sizes_0, x = value_states_15_cast_fp16)[name = string("op_728_cast_fp16")]; + bool attn_weights_37_transpose_x_1 = const()[name = string("attn_weights_37_transpose_x_1"), val = bool(true)]; + bool attn_weights_37_transpose_y_1 = const()[name = string("attn_weights_37_transpose_y_1"), val = bool(false)]; + tensor attn_weights_37_cast_fp16 = matmul(transpose_x = attn_weights_37_transpose_x_1, transpose_y = attn_weights_37_transpose_y_1, x = var_726_cast_fp16_0, y = var_724_cast_fp16_0)[name = string("attn_weights_37_cast_fp16")]; + fp16 var_732_to_fp16 = const()[name = string("op_732_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_39_cast_fp16 = mul(x = attn_weights_37_cast_fp16, y = var_732_to_fp16)[name = string("attn_weights_39_cast_fp16")]; + tensor attn_weights_41_cast_fp16 = softmax(axis = var_641, x = attn_weights_39_cast_fp16)[name = string("attn_weights_41_cast_fp16")]; + bool var_735_transpose_x_0 = const()[name = string("op_735_transpose_x_0"), val = bool(false)]; + bool var_735_transpose_y_0 = const()[name = string("op_735_transpose_y_0"), val = bool(false)]; + tensor var_735_cast_fp16 = matmul(transpose_x = var_735_transpose_x_0, transpose_y = var_735_transpose_y_0, x = var_728_cast_fp16_0, y = attn_weights_41_cast_fp16)[name = string("op_735_cast_fp16")]; + bool attn_weights_43_transpose_x_1 = const()[name = string("attn_weights_43_transpose_x_1"), val = bool(true)]; + bool attn_weights_43_transpose_y_1 = const()[name = string("attn_weights_43_transpose_y_1"), val = bool(false)]; + tensor attn_weights_43_cast_fp16 = matmul(transpose_x = attn_weights_43_transpose_x_1, transpose_y = attn_weights_43_transpose_y_1, x = var_726_cast_fp16_1, y = var_724_cast_fp16_1)[name = string("attn_weights_43_cast_fp16")]; + fp16 var_738_to_fp16 = const()[name = string("op_738_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_45_cast_fp16 = mul(x = attn_weights_43_cast_fp16, y = var_738_to_fp16)[name = string("attn_weights_45_cast_fp16")]; + tensor attn_weights_47_cast_fp16 = softmax(axis = var_641, x = attn_weights_45_cast_fp16)[name = string("attn_weights_47_cast_fp16")]; + bool attn_out_7_transpose_x_0 = const()[name = string("attn_out_7_transpose_x_0"), val = bool(false)]; + bool attn_out_7_transpose_y_0 = const()[name = string("attn_out_7_transpose_y_0"), val = bool(false)]; + tensor attn_out_7_cast_fp16 = matmul(transpose_x = attn_out_7_transpose_x_0, transpose_y = attn_out_7_transpose_y_0, x = var_728_cast_fp16_1, y = attn_weights_47_cast_fp16)[name = string("attn_out_7_cast_fp16")]; + bool attn_output_7_interleave_0 = const()[name = string("attn_output_7_interleave_0"), val = bool(false)]; + tensor attn_output_7_cast_fp16 = concat(axis = var_643, interleave = attn_output_7_interleave_0, values = (var_735_cast_fp16, attn_out_7_cast_fp16))[name = string("attn_output_7_cast_fp16")]; + tensor var_744 = const()[name = string("op_744"), val = tensor([2, 2048, 1, 11])]; + tensor x_41_cast_fp16 = reshape(shape = var_744, x = attn_output_7_cast_fp16)[name = string("x_41_cast_fp16")]; + string hidden_states_35_pad_type_0 = const()[name = string("hidden_states_35_pad_type_0"), val = string("valid")]; + tensor hidden_states_35_strides_0 = const()[name = string("hidden_states_35_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_35_pad_0 = const()[name = string("hidden_states_35_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_35_dilations_0 = const()[name = string("hidden_states_35_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_35_groups_0 = const()[name = string("hidden_states_35_groups_0"), val = int32(1)]; + tensor hidden_states_35_cast_fp16 = conv(dilations = hidden_states_35_dilations_0, groups = hidden_states_35_groups_0, pad = hidden_states_35_pad_0, pad_type = hidden_states_35_pad_type_0, strides = hidden_states_35_strides_0, weight = locdit_decoder_layers_3_self_attn_o_proj_weight_cast_fp16, x = x_41_cast_fp16)[name = string("hidden_states_35_cast_fp16")]; + tensor hidden_states_37_cast_fp16 = add(x = hidden_states_31_cast_fp16, y = hidden_states_35_cast_fp16)[name = string("hidden_states_37_cast_fp16")]; + fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_754_cast_fp16 = mul(x = hidden_states_37_cast_fp16, y = const_22_promoted_to_fp16)[name = string("op_754_cast_fp16")]; + bool doubled_29_interleave_0 = const()[name = string("doubled_29_interleave_0"), val = bool(false)]; + tensor doubled_29_cast_fp16 = concat(axis = var_643, interleave = doubled_29_interleave_0, values = (hidden_states_37_cast_fp16, var_754_cast_fp16))[name = string("doubled_29_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + tensor out_15_gamma_0_to_fp16 = const()[name = string("out_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212513920)))]; + fp16 var_764_to_fp16 = const()[name = string("op_764_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_764_to_fp16, gamma = out_15_gamma_0_to_fp16, x = doubled_29_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor var_775_split_sizes_0 = const()[name = string("op_775_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_775_axis_0 = const()[name = string("op_775_axis_0"), val = int32(1)]; + tensor var_775_cast_fp16_0, tensor var_775_cast_fp16_1 = split(axis = var_775_axis_0, split_sizes = var_775_split_sizes_0, x = out_15_cast_fp16)[name = string("op_775_cast_fp16")]; + string input_11_pad_type_0 = const()[name = string("input_11_pad_type_0"), val = string("valid")]; + tensor input_11_strides_0 = const()[name = string("input_11_strides_0"), val = tensor([1, 1])]; + tensor input_11_pad_0 = const()[name = string("input_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_11_dilations_0 = const()[name = string("input_11_dilations_0"), val = tensor([1, 1])]; + int32 input_11_groups_0 = const()[name = string("input_11_groups_0"), val = int32(1)]; + tensor input_11_cast_fp16 = conv(dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = locdit_decoder_layers_3_mlp_gate_proj_weight_cast_fp16, x = var_775_cast_fp16_0)[name = string("input_11_cast_fp16")]; + tensor var_787_cast_fp16 = silu(x = input_11_cast_fp16)[name = string("op_787_cast_fp16")]; + string var_793_pad_type_0 = const()[name = string("op_793_pad_type_0"), val = string("valid")]; + tensor var_793_strides_0 = const()[name = string("op_793_strides_0"), val = tensor([1, 1])]; + tensor var_793_pad_0 = const()[name = string("op_793_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_793_dilations_0 = const()[name = string("op_793_dilations_0"), val = tensor([1, 1])]; + int32 var_793_groups_0 = const()[name = string("op_793_groups_0"), val = int32(1)]; + tensor var_793_cast_fp16 = conv(dilations = var_793_dilations_0, groups = var_793_groups_0, pad = var_793_pad_0, pad_type = var_793_pad_type_0, strides = var_793_strides_0, weight = locdit_decoder_layers_3_mlp_up_proj_weight_cast_fp16, x = var_775_cast_fp16_0)[name = string("op_793_cast_fp16")]; + tensor x_45_cast_fp16 = mul(x = var_787_cast_fp16, y = var_793_cast_fp16)[name = string("x_45_cast_fp16")]; + string hidden_states_39_pad_type_0 = const()[name = string("hidden_states_39_pad_type_0"), val = string("valid")]; + tensor hidden_states_39_strides_0 = const()[name = string("hidden_states_39_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_39_pad_0 = const()[name = string("hidden_states_39_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_39_dilations_0 = const()[name = string("hidden_states_39_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_39_groups_0 = const()[name = string("hidden_states_39_groups_0"), val = int32(1)]; + tensor hidden_states_39_cast_fp16 = conv(dilations = hidden_states_39_dilations_0, groups = hidden_states_39_groups_0, pad = hidden_states_39_pad_0, pad_type = hidden_states_39_pad_type_0, strides = hidden_states_39_strides_0, weight = locdit_decoder_layers_3_mlp_down_proj_weight_cast_fp16, x = x_45_cast_fp16)[name = string("hidden_states_39_cast_fp16")]; + tensor hidden_states_41_cast_fp16 = add(x = hidden_states_37_cast_fp16, y = hidden_states_39_cast_fp16)[name = string("hidden_states_41_cast_fp16")]; + int32 var_804 = const()[name = string("op_804"), val = int32(-2)]; + int32 var_806 = const()[name = string("op_806"), val = int32(1)]; + fp16 const_24_promoted_to_fp16 = const()[name = string("const_24_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_819_cast_fp16 = mul(x = hidden_states_41_cast_fp16, y = const_24_promoted_to_fp16)[name = string("op_819_cast_fp16")]; + bool doubled_33_interleave_0 = const()[name = string("doubled_33_interleave_0"), val = bool(false)]; + tensor doubled_33_cast_fp16 = concat(axis = var_806, interleave = doubled_33_interleave_0, values = (hidden_states_41_cast_fp16, var_819_cast_fp16))[name = string("doubled_33_cast_fp16")]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + tensor out_17_gamma_0_to_fp16 = const()[name = string("out_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212518080)))]; + fp16 var_829_to_fp16 = const()[name = string("op_829_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_829_to_fp16, gamma = out_17_gamma_0_to_fp16, x = doubled_33_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor var_840_split_sizes_0 = const()[name = string("op_840_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_840_axis_0 = const()[name = string("op_840_axis_0"), val = int32(1)]; + tensor var_840_cast_fp16_0, tensor var_840_cast_fp16_1 = split(axis = var_840_axis_0, split_sizes = var_840_split_sizes_0, x = out_17_cast_fp16)[name = string("op_840_cast_fp16")]; + string query_states_17_pad_type_0 = const()[name = string("query_states_17_pad_type_0"), val = string("valid")]; + tensor query_states_17_strides_0 = const()[name = string("query_states_17_strides_0"), val = tensor([1, 1])]; + tensor query_states_17_pad_0 = const()[name = string("query_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_17_dilations_0 = const()[name = string("query_states_17_dilations_0"), val = tensor([1, 1])]; + int32 query_states_17_groups_0 = const()[name = string("query_states_17_groups_0"), val = int32(1)]; + tensor query_states_17_cast_fp16 = conv(dilations = query_states_17_dilations_0, groups = query_states_17_groups_0, pad = query_states_17_pad_0, pad_type = query_states_17_pad_type_0, strides = query_states_17_strides_0, weight = locdit_decoder_layers_4_self_attn_q_proj_weight_cast_fp16, x = var_840_cast_fp16_0)[name = string("query_states_17_cast_fp16")]; + string key_states_17_pad_type_0 = const()[name = string("key_states_17_pad_type_0"), val = string("valid")]; + tensor key_states_17_strides_0 = const()[name = string("key_states_17_strides_0"), val = tensor([1, 1])]; + tensor key_states_17_pad_0 = const()[name = string("key_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_17_dilations_0 = const()[name = string("key_states_17_dilations_0"), val = tensor([1, 1])]; + int32 key_states_17_groups_0 = const()[name = string("key_states_17_groups_0"), val = int32(1)]; + tensor key_states_17_cast_fp16 = conv(dilations = key_states_17_dilations_0, groups = key_states_17_groups_0, pad = key_states_17_pad_0, pad_type = key_states_17_pad_type_0, strides = key_states_17_strides_0, weight = locdit_decoder_layers_4_self_attn_k_proj_weight_cast_fp16, x = var_840_cast_fp16_0)[name = string("key_states_17_cast_fp16")]; + string value_states_17_pad_type_0 = const()[name = string("value_states_17_pad_type_0"), val = string("valid")]; + tensor value_states_17_strides_0 = const()[name = string("value_states_17_strides_0"), val = tensor([1, 1])]; + tensor value_states_17_pad_0 = const()[name = string("value_states_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_17_dilations_0 = const()[name = string("value_states_17_dilations_0"), val = tensor([1, 1])]; + int32 value_states_17_groups_0 = const()[name = string("value_states_17_groups_0"), val = int32(1)]; + tensor value_states_17_cast_fp16 = conv(dilations = value_states_17_dilations_0, groups = value_states_17_groups_0, pad = value_states_17_pad_0, pad_type = value_states_17_pad_type_0, strides = value_states_17_strides_0, weight = locdit_decoder_layers_4_self_attn_v_proj_weight_cast_fp16, x = var_840_cast_fp16_0)[name = string("value_states_17_cast_fp16")]; + tensor var_867 = const()[name = string("op_867"), val = tensor([2, 16, 128, 11])]; + tensor x_47_cast_fp16 = reshape(shape = var_867, x = query_states_17_cast_fp16)[name = string("x_47_cast_fp16")]; + tensor var_869 = const()[name = string("op_869"), val = tensor([2, 2, 128, 11])]; + tensor x_49_cast_fp16 = reshape(shape = var_869, x = key_states_17_cast_fp16)[name = string("x_49_cast_fp16")]; + tensor var_871 = const()[name = string("op_871"), val = tensor([2, 2, 128, 11])]; + tensor value_states_19_cast_fp16 = reshape(shape = var_871, x = value_states_17_cast_fp16)[name = string("value_states_19_cast_fp16")]; + tensor var_873_cast_fp16 = mul(x = x_47_cast_fp16, y = rope_cos_to_fp16)[name = string("op_873_cast_fp16")]; + tensor var_874_split_sizes_0 = const()[name = string("op_874_split_sizes_0"), val = tensor([64, 64])]; + int32 var_874_axis_0 = const()[name = string("op_874_axis_0"), val = int32(-2)]; + tensor var_874_cast_fp16_0, tensor var_874_cast_fp16_1 = split(axis = var_874_axis_0, split_sizes = var_874_split_sizes_0, x = x_47_cast_fp16)[name = string("op_874_cast_fp16")]; + bool var_877_interleave_0 = const()[name = string("op_877_interleave_0"), val = bool(false)]; + tensor var_877_cast_fp16 = concat(axis = var_804, interleave = var_877_interleave_0, values = (var_874_cast_fp16_1, var_874_cast_fp16_0))[name = string("op_877_cast_fp16")]; + tensor var_878_cast_fp16 = mul(x = var_877_cast_fp16, y = rope_sin_to_fp16)[name = string("op_878_cast_fp16")]; + tensor query_states_19_cast_fp16 = add(x = var_873_cast_fp16, y = var_878_cast_fp16)[name = string("query_states_19_cast_fp16")]; + tensor var_880_cast_fp16 = mul(x = x_49_cast_fp16, y = rope_cos_to_fp16)[name = string("op_880_cast_fp16")]; + tensor var_881_split_sizes_0 = const()[name = string("op_881_split_sizes_0"), val = tensor([64, 64])]; + int32 var_881_axis_0 = const()[name = string("op_881_axis_0"), val = int32(-2)]; + tensor var_881_cast_fp16_0, tensor var_881_cast_fp16_1 = split(axis = var_881_axis_0, split_sizes = var_881_split_sizes_0, x = x_49_cast_fp16)[name = string("op_881_cast_fp16")]; + bool var_884_interleave_0 = const()[name = string("op_884_interleave_0"), val = bool(false)]; + tensor var_884_cast_fp16 = concat(axis = var_804, interleave = var_884_interleave_0, values = (var_881_cast_fp16_1, var_881_cast_fp16_0))[name = string("op_884_cast_fp16")]; + tensor var_885_cast_fp16 = mul(x = var_884_cast_fp16, y = rope_sin_to_fp16)[name = string("op_885_cast_fp16")]; + tensor key_states_19_cast_fp16 = add(x = var_880_cast_fp16, y = var_885_cast_fp16)[name = string("key_states_19_cast_fp16")]; + tensor var_887_split_sizes_0 = const()[name = string("op_887_split_sizes_0"), val = tensor([8, 8])]; + int32 var_887_axis_0 = const()[name = string("op_887_axis_0"), val = int32(1)]; + tensor var_887_cast_fp16_0, tensor var_887_cast_fp16_1 = split(axis = var_887_axis_0, split_sizes = var_887_split_sizes_0, x = query_states_19_cast_fp16)[name = string("op_887_cast_fp16")]; + tensor var_889_split_sizes_0 = const()[name = string("op_889_split_sizes_0"), val = tensor([1, 1])]; + int32 var_889_axis_0 = const()[name = string("op_889_axis_0"), val = int32(1)]; + tensor var_889_cast_fp16_0, tensor var_889_cast_fp16_1 = split(axis = var_889_axis_0, split_sizes = var_889_split_sizes_0, x = key_states_19_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_891_split_sizes_0 = const()[name = string("op_891_split_sizes_0"), val = tensor([1, 1])]; + int32 var_891_axis_0 = const()[name = string("op_891_axis_0"), val = int32(1)]; + tensor var_891_cast_fp16_0, tensor var_891_cast_fp16_1 = split(axis = var_891_axis_0, split_sizes = var_891_split_sizes_0, x = value_states_19_cast_fp16)[name = string("op_891_cast_fp16")]; + bool attn_weights_49_transpose_x_1 = const()[name = string("attn_weights_49_transpose_x_1"), val = bool(true)]; + bool attn_weights_49_transpose_y_1 = const()[name = string("attn_weights_49_transpose_y_1"), val = bool(false)]; + tensor attn_weights_49_cast_fp16 = matmul(transpose_x = attn_weights_49_transpose_x_1, transpose_y = attn_weights_49_transpose_y_1, x = var_889_cast_fp16_0, y = var_887_cast_fp16_0)[name = string("attn_weights_49_cast_fp16")]; + fp16 var_895_to_fp16 = const()[name = string("op_895_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_51_cast_fp16 = mul(x = attn_weights_49_cast_fp16, y = var_895_to_fp16)[name = string("attn_weights_51_cast_fp16")]; + tensor attn_weights_53_cast_fp16 = softmax(axis = var_804, x = attn_weights_51_cast_fp16)[name = string("attn_weights_53_cast_fp16")]; + bool var_898_transpose_x_0 = const()[name = string("op_898_transpose_x_0"), val = bool(false)]; + bool var_898_transpose_y_0 = const()[name = string("op_898_transpose_y_0"), val = bool(false)]; + tensor var_898_cast_fp16 = matmul(transpose_x = var_898_transpose_x_0, transpose_y = var_898_transpose_y_0, x = var_891_cast_fp16_0, y = attn_weights_53_cast_fp16)[name = string("op_898_cast_fp16")]; + bool attn_weights_55_transpose_x_1 = const()[name = string("attn_weights_55_transpose_x_1"), val = bool(true)]; + bool attn_weights_55_transpose_y_1 = const()[name = string("attn_weights_55_transpose_y_1"), val = bool(false)]; + tensor attn_weights_55_cast_fp16 = matmul(transpose_x = attn_weights_55_transpose_x_1, transpose_y = attn_weights_55_transpose_y_1, x = var_889_cast_fp16_1, y = var_887_cast_fp16_1)[name = string("attn_weights_55_cast_fp16")]; + fp16 var_901_to_fp16 = const()[name = string("op_901_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_57_cast_fp16 = mul(x = attn_weights_55_cast_fp16, y = var_901_to_fp16)[name = string("attn_weights_57_cast_fp16")]; + tensor attn_weights_59_cast_fp16 = softmax(axis = var_804, x = attn_weights_57_cast_fp16)[name = string("attn_weights_59_cast_fp16")]; + bool attn_out_9_transpose_x_0 = const()[name = string("attn_out_9_transpose_x_0"), val = bool(false)]; + bool attn_out_9_transpose_y_0 = const()[name = string("attn_out_9_transpose_y_0"), val = bool(false)]; + tensor attn_out_9_cast_fp16 = matmul(transpose_x = attn_out_9_transpose_x_0, transpose_y = attn_out_9_transpose_y_0, x = var_891_cast_fp16_1, y = attn_weights_59_cast_fp16)[name = string("attn_out_9_cast_fp16")]; + bool attn_output_9_interleave_0 = const()[name = string("attn_output_9_interleave_0"), val = bool(false)]; + tensor attn_output_9_cast_fp16 = concat(axis = var_806, interleave = attn_output_9_interleave_0, values = (var_898_cast_fp16, attn_out_9_cast_fp16))[name = string("attn_output_9_cast_fp16")]; + tensor var_907 = const()[name = string("op_907"), val = tensor([2, 2048, 1, 11])]; + tensor x_51_cast_fp16 = reshape(shape = var_907, x = attn_output_9_cast_fp16)[name = string("x_51_cast_fp16")]; + string hidden_states_45_pad_type_0 = const()[name = string("hidden_states_45_pad_type_0"), val = string("valid")]; + tensor hidden_states_45_strides_0 = const()[name = string("hidden_states_45_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_45_pad_0 = const()[name = string("hidden_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_45_dilations_0 = const()[name = string("hidden_states_45_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_45_groups_0 = const()[name = string("hidden_states_45_groups_0"), val = int32(1)]; + tensor hidden_states_45_cast_fp16 = conv(dilations = hidden_states_45_dilations_0, groups = hidden_states_45_groups_0, pad = hidden_states_45_pad_0, pad_type = hidden_states_45_pad_type_0, strides = hidden_states_45_strides_0, weight = locdit_decoder_layers_4_self_attn_o_proj_weight_cast_fp16, x = x_51_cast_fp16)[name = string("hidden_states_45_cast_fp16")]; + tensor hidden_states_47_cast_fp16 = add(x = hidden_states_41_cast_fp16, y = hidden_states_45_cast_fp16)[name = string("hidden_states_47_cast_fp16")]; + fp16 const_28_promoted_to_fp16 = const()[name = string("const_28_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_917_cast_fp16 = mul(x = hidden_states_47_cast_fp16, y = const_28_promoted_to_fp16)[name = string("op_917_cast_fp16")]; + bool doubled_37_interleave_0 = const()[name = string("doubled_37_interleave_0"), val = bool(false)]; + tensor doubled_37_cast_fp16 = concat(axis = var_806, interleave = doubled_37_interleave_0, values = (hidden_states_47_cast_fp16, var_917_cast_fp16))[name = string("doubled_37_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + tensor out_19_gamma_0_to_fp16 = const()[name = string("out_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212522240)))]; + fp16 var_927_to_fp16 = const()[name = string("op_927_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_927_to_fp16, gamma = out_19_gamma_0_to_fp16, x = doubled_37_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor var_938_split_sizes_0 = const()[name = string("op_938_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_938_axis_0 = const()[name = string("op_938_axis_0"), val = int32(1)]; + tensor var_938_cast_fp16_0, tensor var_938_cast_fp16_1 = split(axis = var_938_axis_0, split_sizes = var_938_split_sizes_0, x = out_19_cast_fp16)[name = string("op_938_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor input_13_cast_fp16 = conv(dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = locdit_decoder_layers_4_mlp_gate_proj_weight_cast_fp16, x = var_938_cast_fp16_0)[name = string("input_13_cast_fp16")]; + tensor var_950_cast_fp16 = silu(x = input_13_cast_fp16)[name = string("op_950_cast_fp16")]; + string var_956_pad_type_0 = const()[name = string("op_956_pad_type_0"), val = string("valid")]; + tensor var_956_strides_0 = const()[name = string("op_956_strides_0"), val = tensor([1, 1])]; + tensor var_956_pad_0 = const()[name = string("op_956_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_956_dilations_0 = const()[name = string("op_956_dilations_0"), val = tensor([1, 1])]; + int32 var_956_groups_0 = const()[name = string("op_956_groups_0"), val = int32(1)]; + tensor var_956_cast_fp16 = conv(dilations = var_956_dilations_0, groups = var_956_groups_0, pad = var_956_pad_0, pad_type = var_956_pad_type_0, strides = var_956_strides_0, weight = locdit_decoder_layers_4_mlp_up_proj_weight_cast_fp16, x = var_938_cast_fp16_0)[name = string("op_956_cast_fp16")]; + tensor x_55_cast_fp16 = mul(x = var_950_cast_fp16, y = var_956_cast_fp16)[name = string("x_55_cast_fp16")]; + string hidden_states_49_pad_type_0 = const()[name = string("hidden_states_49_pad_type_0"), val = string("valid")]; + tensor hidden_states_49_strides_0 = const()[name = string("hidden_states_49_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_49_pad_0 = const()[name = string("hidden_states_49_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_49_dilations_0 = const()[name = string("hidden_states_49_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_49_groups_0 = const()[name = string("hidden_states_49_groups_0"), val = int32(1)]; + tensor hidden_states_49_cast_fp16 = conv(dilations = hidden_states_49_dilations_0, groups = hidden_states_49_groups_0, pad = hidden_states_49_pad_0, pad_type = hidden_states_49_pad_type_0, strides = hidden_states_49_strides_0, weight = locdit_decoder_layers_4_mlp_down_proj_weight_cast_fp16, x = x_55_cast_fp16)[name = string("hidden_states_49_cast_fp16")]; + tensor hidden_states_51_cast_fp16 = add(x = hidden_states_47_cast_fp16, y = hidden_states_49_cast_fp16)[name = string("hidden_states_51_cast_fp16")]; + int32 var_967 = const()[name = string("op_967"), val = int32(-2)]; + int32 var_969 = const()[name = string("op_969"), val = int32(1)]; + fp16 const_30_promoted_to_fp16 = const()[name = string("const_30_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_982_cast_fp16 = mul(x = hidden_states_51_cast_fp16, y = const_30_promoted_to_fp16)[name = string("op_982_cast_fp16")]; + bool doubled_41_interleave_0 = const()[name = string("doubled_41_interleave_0"), val = bool(false)]; + tensor doubled_41_cast_fp16 = concat(axis = var_969, interleave = doubled_41_interleave_0, values = (hidden_states_51_cast_fp16, var_982_cast_fp16))[name = string("doubled_41_cast_fp16")]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + tensor out_21_gamma_0_to_fp16 = const()[name = string("out_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212526400)))]; + fp16 var_992_to_fp16 = const()[name = string("op_992_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_992_to_fp16, gamma = out_21_gamma_0_to_fp16, x = doubled_41_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor var_1003_split_sizes_0 = const()[name = string("op_1003_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1003_axis_0 = const()[name = string("op_1003_axis_0"), val = int32(1)]; + tensor var_1003_cast_fp16_0, tensor var_1003_cast_fp16_1 = split(axis = var_1003_axis_0, split_sizes = var_1003_split_sizes_0, x = out_21_cast_fp16)[name = string("op_1003_cast_fp16")]; + string query_states_21_pad_type_0 = const()[name = string("query_states_21_pad_type_0"), val = string("valid")]; + tensor query_states_21_strides_0 = const()[name = string("query_states_21_strides_0"), val = tensor([1, 1])]; + tensor query_states_21_pad_0 = const()[name = string("query_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_21_dilations_0 = const()[name = string("query_states_21_dilations_0"), val = tensor([1, 1])]; + int32 query_states_21_groups_0 = const()[name = string("query_states_21_groups_0"), val = int32(1)]; + tensor query_states_21_cast_fp16 = conv(dilations = query_states_21_dilations_0, groups = query_states_21_groups_0, pad = query_states_21_pad_0, pad_type = query_states_21_pad_type_0, strides = query_states_21_strides_0, weight = locdit_decoder_layers_5_self_attn_q_proj_weight_cast_fp16, x = var_1003_cast_fp16_0)[name = string("query_states_21_cast_fp16")]; + string key_states_21_pad_type_0 = const()[name = string("key_states_21_pad_type_0"), val = string("valid")]; + tensor key_states_21_strides_0 = const()[name = string("key_states_21_strides_0"), val = tensor([1, 1])]; + tensor key_states_21_pad_0 = const()[name = string("key_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_21_dilations_0 = const()[name = string("key_states_21_dilations_0"), val = tensor([1, 1])]; + int32 key_states_21_groups_0 = const()[name = string("key_states_21_groups_0"), val = int32(1)]; + tensor key_states_21_cast_fp16 = conv(dilations = key_states_21_dilations_0, groups = key_states_21_groups_0, pad = key_states_21_pad_0, pad_type = key_states_21_pad_type_0, strides = key_states_21_strides_0, weight = locdit_decoder_layers_5_self_attn_k_proj_weight_cast_fp16, x = var_1003_cast_fp16_0)[name = string("key_states_21_cast_fp16")]; + string value_states_21_pad_type_0 = const()[name = string("value_states_21_pad_type_0"), val = string("valid")]; + tensor value_states_21_strides_0 = const()[name = string("value_states_21_strides_0"), val = tensor([1, 1])]; + tensor value_states_21_pad_0 = const()[name = string("value_states_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_21_dilations_0 = const()[name = string("value_states_21_dilations_0"), val = tensor([1, 1])]; + int32 value_states_21_groups_0 = const()[name = string("value_states_21_groups_0"), val = int32(1)]; + tensor value_states_21_cast_fp16 = conv(dilations = value_states_21_dilations_0, groups = value_states_21_groups_0, pad = value_states_21_pad_0, pad_type = value_states_21_pad_type_0, strides = value_states_21_strides_0, weight = locdit_decoder_layers_5_self_attn_v_proj_weight_cast_fp16, x = var_1003_cast_fp16_0)[name = string("value_states_21_cast_fp16")]; + tensor var_1030 = const()[name = string("op_1030"), val = tensor([2, 16, 128, 11])]; + tensor x_57_cast_fp16 = reshape(shape = var_1030, x = query_states_21_cast_fp16)[name = string("x_57_cast_fp16")]; + tensor var_1032 = const()[name = string("op_1032"), val = tensor([2, 2, 128, 11])]; + tensor x_59_cast_fp16 = reshape(shape = var_1032, x = key_states_21_cast_fp16)[name = string("x_59_cast_fp16")]; + tensor var_1034 = const()[name = string("op_1034"), val = tensor([2, 2, 128, 11])]; + tensor value_states_23_cast_fp16 = reshape(shape = var_1034, x = value_states_21_cast_fp16)[name = string("value_states_23_cast_fp16")]; + tensor var_1036_cast_fp16 = mul(x = x_57_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1036_cast_fp16")]; + tensor var_1037_split_sizes_0 = const()[name = string("op_1037_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1037_axis_0 = const()[name = string("op_1037_axis_0"), val = int32(-2)]; + tensor var_1037_cast_fp16_0, tensor var_1037_cast_fp16_1 = split(axis = var_1037_axis_0, split_sizes = var_1037_split_sizes_0, x = x_57_cast_fp16)[name = string("op_1037_cast_fp16")]; + bool var_1040_interleave_0 = const()[name = string("op_1040_interleave_0"), val = bool(false)]; + tensor var_1040_cast_fp16 = concat(axis = var_967, interleave = var_1040_interleave_0, values = (var_1037_cast_fp16_1, var_1037_cast_fp16_0))[name = string("op_1040_cast_fp16")]; + tensor var_1041_cast_fp16 = mul(x = var_1040_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1041_cast_fp16")]; + tensor query_states_23_cast_fp16 = add(x = var_1036_cast_fp16, y = var_1041_cast_fp16)[name = string("query_states_23_cast_fp16")]; + tensor var_1043_cast_fp16 = mul(x = x_59_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1043_cast_fp16")]; + tensor var_1044_split_sizes_0 = const()[name = string("op_1044_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1044_axis_0 = const()[name = string("op_1044_axis_0"), val = int32(-2)]; + tensor var_1044_cast_fp16_0, tensor var_1044_cast_fp16_1 = split(axis = var_1044_axis_0, split_sizes = var_1044_split_sizes_0, x = x_59_cast_fp16)[name = string("op_1044_cast_fp16")]; + bool var_1047_interleave_0 = const()[name = string("op_1047_interleave_0"), val = bool(false)]; + tensor var_1047_cast_fp16 = concat(axis = var_967, interleave = var_1047_interleave_0, values = (var_1044_cast_fp16_1, var_1044_cast_fp16_0))[name = string("op_1047_cast_fp16")]; + tensor var_1048_cast_fp16 = mul(x = var_1047_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1048_cast_fp16")]; + tensor key_states_23_cast_fp16 = add(x = var_1043_cast_fp16, y = var_1048_cast_fp16)[name = string("key_states_23_cast_fp16")]; + tensor var_1050_split_sizes_0 = const()[name = string("op_1050_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1050_axis_0 = const()[name = string("op_1050_axis_0"), val = int32(1)]; + tensor var_1050_cast_fp16_0, tensor var_1050_cast_fp16_1 = split(axis = var_1050_axis_0, split_sizes = var_1050_split_sizes_0, x = query_states_23_cast_fp16)[name = string("op_1050_cast_fp16")]; + tensor var_1052_split_sizes_0 = const()[name = string("op_1052_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1052_axis_0 = const()[name = string("op_1052_axis_0"), val = int32(1)]; + tensor var_1052_cast_fp16_0, tensor var_1052_cast_fp16_1 = split(axis = var_1052_axis_0, split_sizes = var_1052_split_sizes_0, x = key_states_23_cast_fp16)[name = string("op_1052_cast_fp16")]; + tensor var_1054_split_sizes_0 = const()[name = string("op_1054_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1054_axis_0 = const()[name = string("op_1054_axis_0"), val = int32(1)]; + tensor var_1054_cast_fp16_0, tensor var_1054_cast_fp16_1 = split(axis = var_1054_axis_0, split_sizes = var_1054_split_sizes_0, x = value_states_23_cast_fp16)[name = string("op_1054_cast_fp16")]; + bool attn_weights_61_transpose_x_1 = const()[name = string("attn_weights_61_transpose_x_1"), val = bool(true)]; + bool attn_weights_61_transpose_y_1 = const()[name = string("attn_weights_61_transpose_y_1"), val = bool(false)]; + tensor attn_weights_61_cast_fp16 = matmul(transpose_x = attn_weights_61_transpose_x_1, transpose_y = attn_weights_61_transpose_y_1, x = var_1052_cast_fp16_0, y = var_1050_cast_fp16_0)[name = string("attn_weights_61_cast_fp16")]; + fp16 var_1058_to_fp16 = const()[name = string("op_1058_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_63_cast_fp16 = mul(x = attn_weights_61_cast_fp16, y = var_1058_to_fp16)[name = string("attn_weights_63_cast_fp16")]; + tensor attn_weights_65_cast_fp16 = softmax(axis = var_967, x = attn_weights_63_cast_fp16)[name = string("attn_weights_65_cast_fp16")]; + bool var_1061_transpose_x_0 = const()[name = string("op_1061_transpose_x_0"), val = bool(false)]; + bool var_1061_transpose_y_0 = const()[name = string("op_1061_transpose_y_0"), val = bool(false)]; + tensor var_1061_cast_fp16 = matmul(transpose_x = var_1061_transpose_x_0, transpose_y = var_1061_transpose_y_0, x = var_1054_cast_fp16_0, y = attn_weights_65_cast_fp16)[name = string("op_1061_cast_fp16")]; + bool attn_weights_67_transpose_x_1 = const()[name = string("attn_weights_67_transpose_x_1"), val = bool(true)]; + bool attn_weights_67_transpose_y_1 = const()[name = string("attn_weights_67_transpose_y_1"), val = bool(false)]; + tensor attn_weights_67_cast_fp16 = matmul(transpose_x = attn_weights_67_transpose_x_1, transpose_y = attn_weights_67_transpose_y_1, x = var_1052_cast_fp16_1, y = var_1050_cast_fp16_1)[name = string("attn_weights_67_cast_fp16")]; + fp16 var_1064_to_fp16 = const()[name = string("op_1064_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_69_cast_fp16 = mul(x = attn_weights_67_cast_fp16, y = var_1064_to_fp16)[name = string("attn_weights_69_cast_fp16")]; + tensor attn_weights_71_cast_fp16 = softmax(axis = var_967, x = attn_weights_69_cast_fp16)[name = string("attn_weights_71_cast_fp16")]; + bool attn_out_11_transpose_x_0 = const()[name = string("attn_out_11_transpose_x_0"), val = bool(false)]; + bool attn_out_11_transpose_y_0 = const()[name = string("attn_out_11_transpose_y_0"), val = bool(false)]; + tensor attn_out_11_cast_fp16 = matmul(transpose_x = attn_out_11_transpose_x_0, transpose_y = attn_out_11_transpose_y_0, x = var_1054_cast_fp16_1, y = attn_weights_71_cast_fp16)[name = string("attn_out_11_cast_fp16")]; + bool attn_output_11_interleave_0 = const()[name = string("attn_output_11_interleave_0"), val = bool(false)]; + tensor attn_output_11_cast_fp16 = concat(axis = var_969, interleave = attn_output_11_interleave_0, values = (var_1061_cast_fp16, attn_out_11_cast_fp16))[name = string("attn_output_11_cast_fp16")]; + tensor var_1070 = const()[name = string("op_1070"), val = tensor([2, 2048, 1, 11])]; + tensor x_61_cast_fp16 = reshape(shape = var_1070, x = attn_output_11_cast_fp16)[name = string("x_61_cast_fp16")]; + string hidden_states_55_pad_type_0 = const()[name = string("hidden_states_55_pad_type_0"), val = string("valid")]; + tensor hidden_states_55_strides_0 = const()[name = string("hidden_states_55_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_55_pad_0 = const()[name = string("hidden_states_55_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_55_dilations_0 = const()[name = string("hidden_states_55_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_55_groups_0 = const()[name = string("hidden_states_55_groups_0"), val = int32(1)]; + tensor hidden_states_55_cast_fp16 = conv(dilations = hidden_states_55_dilations_0, groups = hidden_states_55_groups_0, pad = hidden_states_55_pad_0, pad_type = hidden_states_55_pad_type_0, strides = hidden_states_55_strides_0, weight = locdit_decoder_layers_5_self_attn_o_proj_weight_cast_fp16, x = x_61_cast_fp16)[name = string("hidden_states_55_cast_fp16")]; + tensor hidden_states_57_cast_fp16 = add(x = hidden_states_51_cast_fp16, y = hidden_states_55_cast_fp16)[name = string("hidden_states_57_cast_fp16")]; + fp16 const_34_promoted_to_fp16 = const()[name = string("const_34_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1080_cast_fp16 = mul(x = hidden_states_57_cast_fp16, y = const_34_promoted_to_fp16)[name = string("op_1080_cast_fp16")]; + bool doubled_45_interleave_0 = const()[name = string("doubled_45_interleave_0"), val = bool(false)]; + tensor doubled_45_cast_fp16 = concat(axis = var_969, interleave = doubled_45_interleave_0, values = (hidden_states_57_cast_fp16, var_1080_cast_fp16))[name = string("doubled_45_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + tensor out_23_gamma_0_to_fp16 = const()[name = string("out_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212530560)))]; + fp16 var_1090_to_fp16 = const()[name = string("op_1090_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_1090_to_fp16, gamma = out_23_gamma_0_to_fp16, x = doubled_45_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor var_1101_split_sizes_0 = const()[name = string("op_1101_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1101_axis_0 = const()[name = string("op_1101_axis_0"), val = int32(1)]; + tensor var_1101_cast_fp16_0, tensor var_1101_cast_fp16_1 = split(axis = var_1101_axis_0, split_sizes = var_1101_split_sizes_0, x = out_23_cast_fp16)[name = string("op_1101_cast_fp16")]; + string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")]; + tensor input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor([1, 1])]; + tensor input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_15_dilations_0 = const()[name = string("input_15_dilations_0"), val = tensor([1, 1])]; + int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)]; + tensor input_15_cast_fp16 = conv(dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = locdit_decoder_layers_5_mlp_gate_proj_weight_cast_fp16, x = var_1101_cast_fp16_0)[name = string("input_15_cast_fp16")]; + tensor var_1113_cast_fp16 = silu(x = input_15_cast_fp16)[name = string("op_1113_cast_fp16")]; + string var_1119_pad_type_0 = const()[name = string("op_1119_pad_type_0"), val = string("valid")]; + tensor var_1119_strides_0 = const()[name = string("op_1119_strides_0"), val = tensor([1, 1])]; + tensor var_1119_pad_0 = const()[name = string("op_1119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1119_dilations_0 = const()[name = string("op_1119_dilations_0"), val = tensor([1, 1])]; + int32 var_1119_groups_0 = const()[name = string("op_1119_groups_0"), val = int32(1)]; + tensor var_1119_cast_fp16 = conv(dilations = var_1119_dilations_0, groups = var_1119_groups_0, pad = var_1119_pad_0, pad_type = var_1119_pad_type_0, strides = var_1119_strides_0, weight = locdit_decoder_layers_5_mlp_up_proj_weight_cast_fp16, x = var_1101_cast_fp16_0)[name = string("op_1119_cast_fp16")]; + tensor x_65_cast_fp16 = mul(x = var_1113_cast_fp16, y = var_1119_cast_fp16)[name = string("x_65_cast_fp16")]; + string hidden_states_59_pad_type_0 = const()[name = string("hidden_states_59_pad_type_0"), val = string("valid")]; + tensor hidden_states_59_strides_0 = const()[name = string("hidden_states_59_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_59_pad_0 = const()[name = string("hidden_states_59_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_59_dilations_0 = const()[name = string("hidden_states_59_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_59_groups_0 = const()[name = string("hidden_states_59_groups_0"), val = int32(1)]; + tensor hidden_states_59_cast_fp16 = conv(dilations = hidden_states_59_dilations_0, groups = hidden_states_59_groups_0, pad = hidden_states_59_pad_0, pad_type = hidden_states_59_pad_type_0, strides = hidden_states_59_strides_0, weight = locdit_decoder_layers_5_mlp_down_proj_weight_cast_fp16, x = x_65_cast_fp16)[name = string("hidden_states_59_cast_fp16")]; + tensor hidden_states_61_cast_fp16 = add(x = hidden_states_57_cast_fp16, y = hidden_states_59_cast_fp16)[name = string("hidden_states_61_cast_fp16")]; + int32 var_1130 = const()[name = string("op_1130"), val = int32(-2)]; + int32 var_1132 = const()[name = string("op_1132"), val = int32(1)]; + fp16 const_36_promoted_to_fp16 = const()[name = string("const_36_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1145_cast_fp16 = mul(x = hidden_states_61_cast_fp16, y = const_36_promoted_to_fp16)[name = string("op_1145_cast_fp16")]; + bool doubled_49_interleave_0 = const()[name = string("doubled_49_interleave_0"), val = bool(false)]; + tensor doubled_49_cast_fp16 = concat(axis = var_1132, interleave = doubled_49_interleave_0, values = (hidden_states_61_cast_fp16, var_1145_cast_fp16))[name = string("doubled_49_cast_fp16")]; + tensor out_25_axes_0 = const()[name = string("out_25_axes_0"), val = tensor([1])]; + tensor out_25_gamma_0_to_fp16 = const()[name = string("out_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212534720)))]; + fp16 var_1155_to_fp16 = const()[name = string("op_1155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1155_to_fp16, gamma = out_25_gamma_0_to_fp16, x = doubled_49_cast_fp16)[name = string("out_25_cast_fp16")]; + tensor var_1166_split_sizes_0 = const()[name = string("op_1166_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1166_axis_0 = const()[name = string("op_1166_axis_0"), val = int32(1)]; + tensor var_1166_cast_fp16_0, tensor var_1166_cast_fp16_1 = split(axis = var_1166_axis_0, split_sizes = var_1166_split_sizes_0, x = out_25_cast_fp16)[name = string("op_1166_cast_fp16")]; + string query_states_25_pad_type_0 = const()[name = string("query_states_25_pad_type_0"), val = string("valid")]; + tensor query_states_25_strides_0 = const()[name = string("query_states_25_strides_0"), val = tensor([1, 1])]; + tensor query_states_25_pad_0 = const()[name = string("query_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_25_dilations_0 = const()[name = string("query_states_25_dilations_0"), val = tensor([1, 1])]; + int32 query_states_25_groups_0 = const()[name = string("query_states_25_groups_0"), val = int32(1)]; + tensor query_states_25_cast_fp16 = conv(dilations = query_states_25_dilations_0, groups = query_states_25_groups_0, pad = query_states_25_pad_0, pad_type = query_states_25_pad_type_0, strides = query_states_25_strides_0, weight = locdit_decoder_layers_6_self_attn_q_proj_weight_cast_fp16, x = var_1166_cast_fp16_0)[name = string("query_states_25_cast_fp16")]; + string key_states_25_pad_type_0 = const()[name = string("key_states_25_pad_type_0"), val = string("valid")]; + tensor key_states_25_strides_0 = const()[name = string("key_states_25_strides_0"), val = tensor([1, 1])]; + tensor key_states_25_pad_0 = const()[name = string("key_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_25_dilations_0 = const()[name = string("key_states_25_dilations_0"), val = tensor([1, 1])]; + int32 key_states_25_groups_0 = const()[name = string("key_states_25_groups_0"), val = int32(1)]; + tensor key_states_25_cast_fp16 = conv(dilations = key_states_25_dilations_0, groups = key_states_25_groups_0, pad = key_states_25_pad_0, pad_type = key_states_25_pad_type_0, strides = key_states_25_strides_0, weight = locdit_decoder_layers_6_self_attn_k_proj_weight_cast_fp16, x = var_1166_cast_fp16_0)[name = string("key_states_25_cast_fp16")]; + string value_states_25_pad_type_0 = const()[name = string("value_states_25_pad_type_0"), val = string("valid")]; + tensor value_states_25_strides_0 = const()[name = string("value_states_25_strides_0"), val = tensor([1, 1])]; + tensor value_states_25_pad_0 = const()[name = string("value_states_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_25_dilations_0 = const()[name = string("value_states_25_dilations_0"), val = tensor([1, 1])]; + int32 value_states_25_groups_0 = const()[name = string("value_states_25_groups_0"), val = int32(1)]; + tensor value_states_25_cast_fp16 = conv(dilations = value_states_25_dilations_0, groups = value_states_25_groups_0, pad = value_states_25_pad_0, pad_type = value_states_25_pad_type_0, strides = value_states_25_strides_0, weight = locdit_decoder_layers_6_self_attn_v_proj_weight_cast_fp16, x = var_1166_cast_fp16_0)[name = string("value_states_25_cast_fp16")]; + tensor var_1193 = const()[name = string("op_1193"), val = tensor([2, 16, 128, 11])]; + tensor x_67_cast_fp16 = reshape(shape = var_1193, x = query_states_25_cast_fp16)[name = string("x_67_cast_fp16")]; + tensor var_1195 = const()[name = string("op_1195"), val = tensor([2, 2, 128, 11])]; + tensor x_69_cast_fp16 = reshape(shape = var_1195, x = key_states_25_cast_fp16)[name = string("x_69_cast_fp16")]; + tensor var_1197 = const()[name = string("op_1197"), val = tensor([2, 2, 128, 11])]; + tensor value_states_27_cast_fp16 = reshape(shape = var_1197, x = value_states_25_cast_fp16)[name = string("value_states_27_cast_fp16")]; + tensor var_1199_cast_fp16 = mul(x = x_67_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1199_cast_fp16")]; + tensor var_1200_split_sizes_0 = const()[name = string("op_1200_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1200_axis_0 = const()[name = string("op_1200_axis_0"), val = int32(-2)]; + tensor var_1200_cast_fp16_0, tensor var_1200_cast_fp16_1 = split(axis = var_1200_axis_0, split_sizes = var_1200_split_sizes_0, x = x_67_cast_fp16)[name = string("op_1200_cast_fp16")]; + bool var_1203_interleave_0 = const()[name = string("op_1203_interleave_0"), val = bool(false)]; + tensor var_1203_cast_fp16 = concat(axis = var_1130, interleave = var_1203_interleave_0, values = (var_1200_cast_fp16_1, var_1200_cast_fp16_0))[name = string("op_1203_cast_fp16")]; + tensor var_1204_cast_fp16 = mul(x = var_1203_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1204_cast_fp16")]; + tensor query_states_27_cast_fp16 = add(x = var_1199_cast_fp16, y = var_1204_cast_fp16)[name = string("query_states_27_cast_fp16")]; + tensor var_1206_cast_fp16 = mul(x = x_69_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1206_cast_fp16")]; + tensor var_1207_split_sizes_0 = const()[name = string("op_1207_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1207_axis_0 = const()[name = string("op_1207_axis_0"), val = int32(-2)]; + tensor var_1207_cast_fp16_0, tensor var_1207_cast_fp16_1 = split(axis = var_1207_axis_0, split_sizes = var_1207_split_sizes_0, x = x_69_cast_fp16)[name = string("op_1207_cast_fp16")]; + bool var_1210_interleave_0 = const()[name = string("op_1210_interleave_0"), val = bool(false)]; + tensor var_1210_cast_fp16 = concat(axis = var_1130, interleave = var_1210_interleave_0, values = (var_1207_cast_fp16_1, var_1207_cast_fp16_0))[name = string("op_1210_cast_fp16")]; + tensor var_1211_cast_fp16 = mul(x = var_1210_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1211_cast_fp16")]; + tensor key_states_27_cast_fp16 = add(x = var_1206_cast_fp16, y = var_1211_cast_fp16)[name = string("key_states_27_cast_fp16")]; + tensor var_1213_split_sizes_0 = const()[name = string("op_1213_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1213_axis_0 = const()[name = string("op_1213_axis_0"), val = int32(1)]; + tensor var_1213_cast_fp16_0, tensor var_1213_cast_fp16_1 = split(axis = var_1213_axis_0, split_sizes = var_1213_split_sizes_0, x = query_states_27_cast_fp16)[name = string("op_1213_cast_fp16")]; + tensor var_1215_split_sizes_0 = const()[name = string("op_1215_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1215_axis_0 = const()[name = string("op_1215_axis_0"), val = int32(1)]; + tensor var_1215_cast_fp16_0, tensor var_1215_cast_fp16_1 = split(axis = var_1215_axis_0, split_sizes = var_1215_split_sizes_0, x = key_states_27_cast_fp16)[name = string("op_1215_cast_fp16")]; + tensor var_1217_split_sizes_0 = const()[name = string("op_1217_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1217_axis_0 = const()[name = string("op_1217_axis_0"), val = int32(1)]; + tensor var_1217_cast_fp16_0, tensor var_1217_cast_fp16_1 = split(axis = var_1217_axis_0, split_sizes = var_1217_split_sizes_0, x = value_states_27_cast_fp16)[name = string("op_1217_cast_fp16")]; + bool attn_weights_73_transpose_x_1 = const()[name = string("attn_weights_73_transpose_x_1"), val = bool(true)]; + bool attn_weights_73_transpose_y_1 = const()[name = string("attn_weights_73_transpose_y_1"), val = bool(false)]; + tensor attn_weights_73_cast_fp16 = matmul(transpose_x = attn_weights_73_transpose_x_1, transpose_y = attn_weights_73_transpose_y_1, x = var_1215_cast_fp16_0, y = var_1213_cast_fp16_0)[name = string("attn_weights_73_cast_fp16")]; + fp16 var_1221_to_fp16 = const()[name = string("op_1221_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_75_cast_fp16 = mul(x = attn_weights_73_cast_fp16, y = var_1221_to_fp16)[name = string("attn_weights_75_cast_fp16")]; + tensor attn_weights_77_cast_fp16 = softmax(axis = var_1130, x = attn_weights_75_cast_fp16)[name = string("attn_weights_77_cast_fp16")]; + bool var_1224_transpose_x_0 = const()[name = string("op_1224_transpose_x_0"), val = bool(false)]; + bool var_1224_transpose_y_0 = const()[name = string("op_1224_transpose_y_0"), val = bool(false)]; + tensor var_1224_cast_fp16 = matmul(transpose_x = var_1224_transpose_x_0, transpose_y = var_1224_transpose_y_0, x = var_1217_cast_fp16_0, y = attn_weights_77_cast_fp16)[name = string("op_1224_cast_fp16")]; + bool attn_weights_79_transpose_x_1 = const()[name = string("attn_weights_79_transpose_x_1"), val = bool(true)]; + bool attn_weights_79_transpose_y_1 = const()[name = string("attn_weights_79_transpose_y_1"), val = bool(false)]; + tensor attn_weights_79_cast_fp16 = matmul(transpose_x = attn_weights_79_transpose_x_1, transpose_y = attn_weights_79_transpose_y_1, x = var_1215_cast_fp16_1, y = var_1213_cast_fp16_1)[name = string("attn_weights_79_cast_fp16")]; + fp16 var_1227_to_fp16 = const()[name = string("op_1227_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_81_cast_fp16 = mul(x = attn_weights_79_cast_fp16, y = var_1227_to_fp16)[name = string("attn_weights_81_cast_fp16")]; + tensor attn_weights_83_cast_fp16 = softmax(axis = var_1130, x = attn_weights_81_cast_fp16)[name = string("attn_weights_83_cast_fp16")]; + bool attn_out_13_transpose_x_0 = const()[name = string("attn_out_13_transpose_x_0"), val = bool(false)]; + bool attn_out_13_transpose_y_0 = const()[name = string("attn_out_13_transpose_y_0"), val = bool(false)]; + tensor attn_out_13_cast_fp16 = matmul(transpose_x = attn_out_13_transpose_x_0, transpose_y = attn_out_13_transpose_y_0, x = var_1217_cast_fp16_1, y = attn_weights_83_cast_fp16)[name = string("attn_out_13_cast_fp16")]; + bool attn_output_13_interleave_0 = const()[name = string("attn_output_13_interleave_0"), val = bool(false)]; + tensor attn_output_13_cast_fp16 = concat(axis = var_1132, interleave = attn_output_13_interleave_0, values = (var_1224_cast_fp16, attn_out_13_cast_fp16))[name = string("attn_output_13_cast_fp16")]; + tensor var_1233 = const()[name = string("op_1233"), val = tensor([2, 2048, 1, 11])]; + tensor x_71_cast_fp16 = reshape(shape = var_1233, x = attn_output_13_cast_fp16)[name = string("x_71_cast_fp16")]; + string hidden_states_65_pad_type_0 = const()[name = string("hidden_states_65_pad_type_0"), val = string("valid")]; + tensor hidden_states_65_strides_0 = const()[name = string("hidden_states_65_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_65_pad_0 = const()[name = string("hidden_states_65_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_65_dilations_0 = const()[name = string("hidden_states_65_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_65_groups_0 = const()[name = string("hidden_states_65_groups_0"), val = int32(1)]; + tensor hidden_states_65_cast_fp16 = conv(dilations = hidden_states_65_dilations_0, groups = hidden_states_65_groups_0, pad = hidden_states_65_pad_0, pad_type = hidden_states_65_pad_type_0, strides = hidden_states_65_strides_0, weight = locdit_decoder_layers_6_self_attn_o_proj_weight_cast_fp16, x = x_71_cast_fp16)[name = string("hidden_states_65_cast_fp16")]; + tensor hidden_states_67_cast_fp16 = add(x = hidden_states_61_cast_fp16, y = hidden_states_65_cast_fp16)[name = string("hidden_states_67_cast_fp16")]; + fp16 const_40_promoted_to_fp16 = const()[name = string("const_40_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1243_cast_fp16 = mul(x = hidden_states_67_cast_fp16, y = const_40_promoted_to_fp16)[name = string("op_1243_cast_fp16")]; + bool doubled_53_interleave_0 = const()[name = string("doubled_53_interleave_0"), val = bool(false)]; + tensor doubled_53_cast_fp16 = concat(axis = var_1132, interleave = doubled_53_interleave_0, values = (hidden_states_67_cast_fp16, var_1243_cast_fp16))[name = string("doubled_53_cast_fp16")]; + tensor out_27_axes_0 = const()[name = string("out_27_axes_0"), val = tensor([1])]; + tensor out_27_gamma_0_to_fp16 = const()[name = string("out_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212538880)))]; + fp16 var_1253_to_fp16 = const()[name = string("op_1253_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1253_to_fp16, gamma = out_27_gamma_0_to_fp16, x = doubled_53_cast_fp16)[name = string("out_27_cast_fp16")]; + tensor var_1264_split_sizes_0 = const()[name = string("op_1264_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1264_axis_0 = const()[name = string("op_1264_axis_0"), val = int32(1)]; + tensor var_1264_cast_fp16_0, tensor var_1264_cast_fp16_1 = split(axis = var_1264_axis_0, split_sizes = var_1264_split_sizes_0, x = out_27_cast_fp16)[name = string("op_1264_cast_fp16")]; + string input_17_pad_type_0 = const()[name = string("input_17_pad_type_0"), val = string("valid")]; + tensor input_17_strides_0 = const()[name = string("input_17_strides_0"), val = tensor([1, 1])]; + tensor input_17_pad_0 = const()[name = string("input_17_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_17_dilations_0 = const()[name = string("input_17_dilations_0"), val = tensor([1, 1])]; + int32 input_17_groups_0 = const()[name = string("input_17_groups_0"), val = int32(1)]; + tensor input_17_cast_fp16 = conv(dilations = input_17_dilations_0, groups = input_17_groups_0, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = input_17_strides_0, weight = locdit_decoder_layers_6_mlp_gate_proj_weight_cast_fp16, x = var_1264_cast_fp16_0)[name = string("input_17_cast_fp16")]; + tensor var_1276_cast_fp16 = silu(x = input_17_cast_fp16)[name = string("op_1276_cast_fp16")]; + string var_1282_pad_type_0 = const()[name = string("op_1282_pad_type_0"), val = string("valid")]; + tensor var_1282_strides_0 = const()[name = string("op_1282_strides_0"), val = tensor([1, 1])]; + tensor var_1282_pad_0 = const()[name = string("op_1282_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1282_dilations_0 = const()[name = string("op_1282_dilations_0"), val = tensor([1, 1])]; + int32 var_1282_groups_0 = const()[name = string("op_1282_groups_0"), val = int32(1)]; + tensor var_1282_cast_fp16 = conv(dilations = var_1282_dilations_0, groups = var_1282_groups_0, pad = var_1282_pad_0, pad_type = var_1282_pad_type_0, strides = var_1282_strides_0, weight = locdit_decoder_layers_6_mlp_up_proj_weight_cast_fp16, x = var_1264_cast_fp16_0)[name = string("op_1282_cast_fp16")]; + tensor x_75_cast_fp16 = mul(x = var_1276_cast_fp16, y = var_1282_cast_fp16)[name = string("x_75_cast_fp16")]; + string hidden_states_69_pad_type_0 = const()[name = string("hidden_states_69_pad_type_0"), val = string("valid")]; + tensor hidden_states_69_strides_0 = const()[name = string("hidden_states_69_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_69_pad_0 = const()[name = string("hidden_states_69_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_69_dilations_0 = const()[name = string("hidden_states_69_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_69_groups_0 = const()[name = string("hidden_states_69_groups_0"), val = int32(1)]; + tensor hidden_states_69_cast_fp16 = conv(dilations = hidden_states_69_dilations_0, groups = hidden_states_69_groups_0, pad = hidden_states_69_pad_0, pad_type = hidden_states_69_pad_type_0, strides = hidden_states_69_strides_0, weight = locdit_decoder_layers_6_mlp_down_proj_weight_cast_fp16, x = x_75_cast_fp16)[name = string("hidden_states_69_cast_fp16")]; + tensor hidden_states_71_cast_fp16 = add(x = hidden_states_67_cast_fp16, y = hidden_states_69_cast_fp16)[name = string("hidden_states_71_cast_fp16")]; + int32 var_1293 = const()[name = string("op_1293"), val = int32(-2)]; + int32 var_1295 = const()[name = string("op_1295"), val = int32(1)]; + fp16 const_42_promoted_to_fp16 = const()[name = string("const_42_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1308_cast_fp16 = mul(x = hidden_states_71_cast_fp16, y = const_42_promoted_to_fp16)[name = string("op_1308_cast_fp16")]; + bool doubled_57_interleave_0 = const()[name = string("doubled_57_interleave_0"), val = bool(false)]; + tensor doubled_57_cast_fp16 = concat(axis = var_1295, interleave = doubled_57_interleave_0, values = (hidden_states_71_cast_fp16, var_1308_cast_fp16))[name = string("doubled_57_cast_fp16")]; + tensor out_29_axes_0 = const()[name = string("out_29_axes_0"), val = tensor([1])]; + tensor out_29_gamma_0_to_fp16 = const()[name = string("out_29_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212543040)))]; + fp16 var_1318_to_fp16 = const()[name = string("op_1318_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1318_to_fp16, gamma = out_29_gamma_0_to_fp16, x = doubled_57_cast_fp16)[name = string("out_29_cast_fp16")]; + tensor var_1329_split_sizes_0 = const()[name = string("op_1329_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1329_axis_0 = const()[name = string("op_1329_axis_0"), val = int32(1)]; + tensor var_1329_cast_fp16_0, tensor var_1329_cast_fp16_1 = split(axis = var_1329_axis_0, split_sizes = var_1329_split_sizes_0, x = out_29_cast_fp16)[name = string("op_1329_cast_fp16")]; + string query_states_29_pad_type_0 = const()[name = string("query_states_29_pad_type_0"), val = string("valid")]; + tensor query_states_29_strides_0 = const()[name = string("query_states_29_strides_0"), val = tensor([1, 1])]; + tensor query_states_29_pad_0 = const()[name = string("query_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_29_dilations_0 = const()[name = string("query_states_29_dilations_0"), val = tensor([1, 1])]; + int32 query_states_29_groups_0 = const()[name = string("query_states_29_groups_0"), val = int32(1)]; + tensor query_states_29_cast_fp16 = conv(dilations = query_states_29_dilations_0, groups = query_states_29_groups_0, pad = query_states_29_pad_0, pad_type = query_states_29_pad_type_0, strides = query_states_29_strides_0, weight = locdit_decoder_layers_7_self_attn_q_proj_weight_cast_fp16, x = var_1329_cast_fp16_0)[name = string("query_states_29_cast_fp16")]; + string key_states_29_pad_type_0 = const()[name = string("key_states_29_pad_type_0"), val = string("valid")]; + tensor key_states_29_strides_0 = const()[name = string("key_states_29_strides_0"), val = tensor([1, 1])]; + tensor key_states_29_pad_0 = const()[name = string("key_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_29_dilations_0 = const()[name = string("key_states_29_dilations_0"), val = tensor([1, 1])]; + int32 key_states_29_groups_0 = const()[name = string("key_states_29_groups_0"), val = int32(1)]; + tensor key_states_29_cast_fp16 = conv(dilations = key_states_29_dilations_0, groups = key_states_29_groups_0, pad = key_states_29_pad_0, pad_type = key_states_29_pad_type_0, strides = key_states_29_strides_0, weight = locdit_decoder_layers_7_self_attn_k_proj_weight_cast_fp16, x = var_1329_cast_fp16_0)[name = string("key_states_29_cast_fp16")]; + string value_states_29_pad_type_0 = const()[name = string("value_states_29_pad_type_0"), val = string("valid")]; + tensor value_states_29_strides_0 = const()[name = string("value_states_29_strides_0"), val = tensor([1, 1])]; + tensor value_states_29_pad_0 = const()[name = string("value_states_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_29_dilations_0 = const()[name = string("value_states_29_dilations_0"), val = tensor([1, 1])]; + int32 value_states_29_groups_0 = const()[name = string("value_states_29_groups_0"), val = int32(1)]; + tensor value_states_29_cast_fp16 = conv(dilations = value_states_29_dilations_0, groups = value_states_29_groups_0, pad = value_states_29_pad_0, pad_type = value_states_29_pad_type_0, strides = value_states_29_strides_0, weight = locdit_decoder_layers_7_self_attn_v_proj_weight_cast_fp16, x = var_1329_cast_fp16_0)[name = string("value_states_29_cast_fp16")]; + tensor var_1356 = const()[name = string("op_1356"), val = tensor([2, 16, 128, 11])]; + tensor x_77_cast_fp16 = reshape(shape = var_1356, x = query_states_29_cast_fp16)[name = string("x_77_cast_fp16")]; + tensor var_1358 = const()[name = string("op_1358"), val = tensor([2, 2, 128, 11])]; + tensor x_79_cast_fp16 = reshape(shape = var_1358, x = key_states_29_cast_fp16)[name = string("x_79_cast_fp16")]; + tensor var_1360 = const()[name = string("op_1360"), val = tensor([2, 2, 128, 11])]; + tensor value_states_31_cast_fp16 = reshape(shape = var_1360, x = value_states_29_cast_fp16)[name = string("value_states_31_cast_fp16")]; + tensor var_1362_cast_fp16 = mul(x = x_77_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1362_cast_fp16")]; + tensor var_1363_split_sizes_0 = const()[name = string("op_1363_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1363_axis_0 = const()[name = string("op_1363_axis_0"), val = int32(-2)]; + tensor var_1363_cast_fp16_0, tensor var_1363_cast_fp16_1 = split(axis = var_1363_axis_0, split_sizes = var_1363_split_sizes_0, x = x_77_cast_fp16)[name = string("op_1363_cast_fp16")]; + bool var_1366_interleave_0 = const()[name = string("op_1366_interleave_0"), val = bool(false)]; + tensor var_1366_cast_fp16 = concat(axis = var_1293, interleave = var_1366_interleave_0, values = (var_1363_cast_fp16_1, var_1363_cast_fp16_0))[name = string("op_1366_cast_fp16")]; + tensor var_1367_cast_fp16 = mul(x = var_1366_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1367_cast_fp16")]; + tensor query_states_31_cast_fp16 = add(x = var_1362_cast_fp16, y = var_1367_cast_fp16)[name = string("query_states_31_cast_fp16")]; + tensor var_1369_cast_fp16 = mul(x = x_79_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1369_cast_fp16")]; + tensor var_1370_split_sizes_0 = const()[name = string("op_1370_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1370_axis_0 = const()[name = string("op_1370_axis_0"), val = int32(-2)]; + tensor var_1370_cast_fp16_0, tensor var_1370_cast_fp16_1 = split(axis = var_1370_axis_0, split_sizes = var_1370_split_sizes_0, x = x_79_cast_fp16)[name = string("op_1370_cast_fp16")]; + bool var_1373_interleave_0 = const()[name = string("op_1373_interleave_0"), val = bool(false)]; + tensor var_1373_cast_fp16 = concat(axis = var_1293, interleave = var_1373_interleave_0, values = (var_1370_cast_fp16_1, var_1370_cast_fp16_0))[name = string("op_1373_cast_fp16")]; + tensor var_1374_cast_fp16 = mul(x = var_1373_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1374_cast_fp16")]; + tensor key_states_31_cast_fp16 = add(x = var_1369_cast_fp16, y = var_1374_cast_fp16)[name = string("key_states_31_cast_fp16")]; + tensor var_1376_split_sizes_0 = const()[name = string("op_1376_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1376_axis_0 = const()[name = string("op_1376_axis_0"), val = int32(1)]; + tensor var_1376_cast_fp16_0, tensor var_1376_cast_fp16_1 = split(axis = var_1376_axis_0, split_sizes = var_1376_split_sizes_0, x = query_states_31_cast_fp16)[name = string("op_1376_cast_fp16")]; + tensor var_1378_split_sizes_0 = const()[name = string("op_1378_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1378_axis_0 = const()[name = string("op_1378_axis_0"), val = int32(1)]; + tensor var_1378_cast_fp16_0, tensor var_1378_cast_fp16_1 = split(axis = var_1378_axis_0, split_sizes = var_1378_split_sizes_0, x = key_states_31_cast_fp16)[name = string("op_1378_cast_fp16")]; + tensor var_1380_split_sizes_0 = const()[name = string("op_1380_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1380_axis_0 = const()[name = string("op_1380_axis_0"), val = int32(1)]; + tensor var_1380_cast_fp16_0, tensor var_1380_cast_fp16_1 = split(axis = var_1380_axis_0, split_sizes = var_1380_split_sizes_0, x = value_states_31_cast_fp16)[name = string("op_1380_cast_fp16")]; + bool attn_weights_85_transpose_x_1 = const()[name = string("attn_weights_85_transpose_x_1"), val = bool(true)]; + bool attn_weights_85_transpose_y_1 = const()[name = string("attn_weights_85_transpose_y_1"), val = bool(false)]; + tensor attn_weights_85_cast_fp16 = matmul(transpose_x = attn_weights_85_transpose_x_1, transpose_y = attn_weights_85_transpose_y_1, x = var_1378_cast_fp16_0, y = var_1376_cast_fp16_0)[name = string("attn_weights_85_cast_fp16")]; + fp16 var_1384_to_fp16 = const()[name = string("op_1384_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_87_cast_fp16 = mul(x = attn_weights_85_cast_fp16, y = var_1384_to_fp16)[name = string("attn_weights_87_cast_fp16")]; + tensor attn_weights_89_cast_fp16 = softmax(axis = var_1293, x = attn_weights_87_cast_fp16)[name = string("attn_weights_89_cast_fp16")]; + bool var_1387_transpose_x_0 = const()[name = string("op_1387_transpose_x_0"), val = bool(false)]; + bool var_1387_transpose_y_0 = const()[name = string("op_1387_transpose_y_0"), val = bool(false)]; + tensor var_1387_cast_fp16 = matmul(transpose_x = var_1387_transpose_x_0, transpose_y = var_1387_transpose_y_0, x = var_1380_cast_fp16_0, y = attn_weights_89_cast_fp16)[name = string("op_1387_cast_fp16")]; + bool attn_weights_91_transpose_x_1 = const()[name = string("attn_weights_91_transpose_x_1"), val = bool(true)]; + bool attn_weights_91_transpose_y_1 = const()[name = string("attn_weights_91_transpose_y_1"), val = bool(false)]; + tensor attn_weights_91_cast_fp16 = matmul(transpose_x = attn_weights_91_transpose_x_1, transpose_y = attn_weights_91_transpose_y_1, x = var_1378_cast_fp16_1, y = var_1376_cast_fp16_1)[name = string("attn_weights_91_cast_fp16")]; + fp16 var_1390_to_fp16 = const()[name = string("op_1390_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_93_cast_fp16 = mul(x = attn_weights_91_cast_fp16, y = var_1390_to_fp16)[name = string("attn_weights_93_cast_fp16")]; + tensor attn_weights_95_cast_fp16 = softmax(axis = var_1293, x = attn_weights_93_cast_fp16)[name = string("attn_weights_95_cast_fp16")]; + bool attn_out_15_transpose_x_0 = const()[name = string("attn_out_15_transpose_x_0"), val = bool(false)]; + bool attn_out_15_transpose_y_0 = const()[name = string("attn_out_15_transpose_y_0"), val = bool(false)]; + tensor attn_out_15_cast_fp16 = matmul(transpose_x = attn_out_15_transpose_x_0, transpose_y = attn_out_15_transpose_y_0, x = var_1380_cast_fp16_1, y = attn_weights_95_cast_fp16)[name = string("attn_out_15_cast_fp16")]; + bool attn_output_15_interleave_0 = const()[name = string("attn_output_15_interleave_0"), val = bool(false)]; + tensor attn_output_15_cast_fp16 = concat(axis = var_1295, interleave = attn_output_15_interleave_0, values = (var_1387_cast_fp16, attn_out_15_cast_fp16))[name = string("attn_output_15_cast_fp16")]; + tensor var_1396 = const()[name = string("op_1396"), val = tensor([2, 2048, 1, 11])]; + tensor x_81_cast_fp16 = reshape(shape = var_1396, x = attn_output_15_cast_fp16)[name = string("x_81_cast_fp16")]; + string hidden_states_75_pad_type_0 = const()[name = string("hidden_states_75_pad_type_0"), val = string("valid")]; + tensor hidden_states_75_strides_0 = const()[name = string("hidden_states_75_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_75_pad_0 = const()[name = string("hidden_states_75_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_75_dilations_0 = const()[name = string("hidden_states_75_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_75_groups_0 = const()[name = string("hidden_states_75_groups_0"), val = int32(1)]; + tensor hidden_states_75_cast_fp16 = conv(dilations = hidden_states_75_dilations_0, groups = hidden_states_75_groups_0, pad = hidden_states_75_pad_0, pad_type = hidden_states_75_pad_type_0, strides = hidden_states_75_strides_0, weight = locdit_decoder_layers_7_self_attn_o_proj_weight_cast_fp16, x = x_81_cast_fp16)[name = string("hidden_states_75_cast_fp16")]; + tensor hidden_states_77_cast_fp16 = add(x = hidden_states_71_cast_fp16, y = hidden_states_75_cast_fp16)[name = string("hidden_states_77_cast_fp16")]; + fp16 const_46_promoted_to_fp16 = const()[name = string("const_46_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1406_cast_fp16 = mul(x = hidden_states_77_cast_fp16, y = const_46_promoted_to_fp16)[name = string("op_1406_cast_fp16")]; + bool doubled_61_interleave_0 = const()[name = string("doubled_61_interleave_0"), val = bool(false)]; + tensor doubled_61_cast_fp16 = concat(axis = var_1295, interleave = doubled_61_interleave_0, values = (hidden_states_77_cast_fp16, var_1406_cast_fp16))[name = string("doubled_61_cast_fp16")]; + tensor out_31_axes_0 = const()[name = string("out_31_axes_0"), val = tensor([1])]; + tensor out_31_gamma_0_to_fp16 = const()[name = string("out_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212547200)))]; + fp16 var_1416_to_fp16 = const()[name = string("op_1416_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1416_to_fp16, gamma = out_31_gamma_0_to_fp16, x = doubled_61_cast_fp16)[name = string("out_31_cast_fp16")]; + tensor var_1427_split_sizes_0 = const()[name = string("op_1427_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1427_axis_0 = const()[name = string("op_1427_axis_0"), val = int32(1)]; + tensor var_1427_cast_fp16_0, tensor var_1427_cast_fp16_1 = split(axis = var_1427_axis_0, split_sizes = var_1427_split_sizes_0, x = out_31_cast_fp16)[name = string("op_1427_cast_fp16")]; + string input_19_pad_type_0 = const()[name = string("input_19_pad_type_0"), val = string("valid")]; + tensor input_19_strides_0 = const()[name = string("input_19_strides_0"), val = tensor([1, 1])]; + tensor input_19_pad_0 = const()[name = string("input_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_19_dilations_0 = const()[name = string("input_19_dilations_0"), val = tensor([1, 1])]; + int32 input_19_groups_0 = const()[name = string("input_19_groups_0"), val = int32(1)]; + tensor input_19_cast_fp16 = conv(dilations = input_19_dilations_0, groups = input_19_groups_0, pad = input_19_pad_0, pad_type = input_19_pad_type_0, strides = input_19_strides_0, weight = locdit_decoder_layers_7_mlp_gate_proj_weight_cast_fp16, x = var_1427_cast_fp16_0)[name = string("input_19_cast_fp16")]; + tensor var_1439_cast_fp16 = silu(x = input_19_cast_fp16)[name = string("op_1439_cast_fp16")]; + string var_1445_pad_type_0 = const()[name = string("op_1445_pad_type_0"), val = string("valid")]; + tensor var_1445_strides_0 = const()[name = string("op_1445_strides_0"), val = tensor([1, 1])]; + tensor var_1445_pad_0 = const()[name = string("op_1445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1445_dilations_0 = const()[name = string("op_1445_dilations_0"), val = tensor([1, 1])]; + int32 var_1445_groups_0 = const()[name = string("op_1445_groups_0"), val = int32(1)]; + tensor var_1445_cast_fp16 = conv(dilations = var_1445_dilations_0, groups = var_1445_groups_0, pad = var_1445_pad_0, pad_type = var_1445_pad_type_0, strides = var_1445_strides_0, weight = locdit_decoder_layers_7_mlp_up_proj_weight_cast_fp16, x = var_1427_cast_fp16_0)[name = string("op_1445_cast_fp16")]; + tensor x_85_cast_fp16 = mul(x = var_1439_cast_fp16, y = var_1445_cast_fp16)[name = string("x_85_cast_fp16")]; + string hidden_states_79_pad_type_0 = const()[name = string("hidden_states_79_pad_type_0"), val = string("valid")]; + tensor hidden_states_79_strides_0 = const()[name = string("hidden_states_79_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_79_pad_0 = const()[name = string("hidden_states_79_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_79_dilations_0 = const()[name = string("hidden_states_79_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_79_groups_0 = const()[name = string("hidden_states_79_groups_0"), val = int32(1)]; + tensor hidden_states_79_cast_fp16 = conv(dilations = hidden_states_79_dilations_0, groups = hidden_states_79_groups_0, pad = hidden_states_79_pad_0, pad_type = hidden_states_79_pad_type_0, strides = hidden_states_79_strides_0, weight = locdit_decoder_layers_7_mlp_down_proj_weight_cast_fp16, x = x_85_cast_fp16)[name = string("hidden_states_79_cast_fp16")]; + tensor hidden_states_81_cast_fp16 = add(x = hidden_states_77_cast_fp16, y = hidden_states_79_cast_fp16)[name = string("hidden_states_81_cast_fp16")]; + int32 var_1456 = const()[name = string("op_1456"), val = int32(-2)]; + int32 var_1458 = const()[name = string("op_1458"), val = int32(1)]; + fp16 const_48_promoted_to_fp16 = const()[name = string("const_48_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1471_cast_fp16 = mul(x = hidden_states_81_cast_fp16, y = const_48_promoted_to_fp16)[name = string("op_1471_cast_fp16")]; + bool doubled_65_interleave_0 = const()[name = string("doubled_65_interleave_0"), val = bool(false)]; + tensor doubled_65_cast_fp16 = concat(axis = var_1458, interleave = doubled_65_interleave_0, values = (hidden_states_81_cast_fp16, var_1471_cast_fp16))[name = string("doubled_65_cast_fp16")]; + tensor out_33_axes_0 = const()[name = string("out_33_axes_0"), val = tensor([1])]; + tensor out_33_gamma_0_to_fp16 = const()[name = string("out_33_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212551360)))]; + fp16 var_1481_to_fp16 = const()[name = string("op_1481_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1481_to_fp16, gamma = out_33_gamma_0_to_fp16, x = doubled_65_cast_fp16)[name = string("out_33_cast_fp16")]; + tensor var_1492_split_sizes_0 = const()[name = string("op_1492_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1492_axis_0 = const()[name = string("op_1492_axis_0"), val = int32(1)]; + tensor var_1492_cast_fp16_0, tensor var_1492_cast_fp16_1 = split(axis = var_1492_axis_0, split_sizes = var_1492_split_sizes_0, x = out_33_cast_fp16)[name = string("op_1492_cast_fp16")]; + string query_states_33_pad_type_0 = const()[name = string("query_states_33_pad_type_0"), val = string("valid")]; + tensor query_states_33_strides_0 = const()[name = string("query_states_33_strides_0"), val = tensor([1, 1])]; + tensor query_states_33_pad_0 = const()[name = string("query_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_33_dilations_0 = const()[name = string("query_states_33_dilations_0"), val = tensor([1, 1])]; + int32 query_states_33_groups_0 = const()[name = string("query_states_33_groups_0"), val = int32(1)]; + tensor query_states_33_cast_fp16 = conv(dilations = query_states_33_dilations_0, groups = query_states_33_groups_0, pad = query_states_33_pad_0, pad_type = query_states_33_pad_type_0, strides = query_states_33_strides_0, weight = locdit_decoder_layers_8_self_attn_q_proj_weight_cast_fp16, x = var_1492_cast_fp16_0)[name = string("query_states_33_cast_fp16")]; + string key_states_33_pad_type_0 = const()[name = string("key_states_33_pad_type_0"), val = string("valid")]; + tensor key_states_33_strides_0 = const()[name = string("key_states_33_strides_0"), val = tensor([1, 1])]; + tensor key_states_33_pad_0 = const()[name = string("key_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_33_dilations_0 = const()[name = string("key_states_33_dilations_0"), val = tensor([1, 1])]; + int32 key_states_33_groups_0 = const()[name = string("key_states_33_groups_0"), val = int32(1)]; + tensor key_states_33_cast_fp16 = conv(dilations = key_states_33_dilations_0, groups = key_states_33_groups_0, pad = key_states_33_pad_0, pad_type = key_states_33_pad_type_0, strides = key_states_33_strides_0, weight = locdit_decoder_layers_8_self_attn_k_proj_weight_cast_fp16, x = var_1492_cast_fp16_0)[name = string("key_states_33_cast_fp16")]; + string value_states_33_pad_type_0 = const()[name = string("value_states_33_pad_type_0"), val = string("valid")]; + tensor value_states_33_strides_0 = const()[name = string("value_states_33_strides_0"), val = tensor([1, 1])]; + tensor value_states_33_pad_0 = const()[name = string("value_states_33_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_33_dilations_0 = const()[name = string("value_states_33_dilations_0"), val = tensor([1, 1])]; + int32 value_states_33_groups_0 = const()[name = string("value_states_33_groups_0"), val = int32(1)]; + tensor value_states_33_cast_fp16 = conv(dilations = value_states_33_dilations_0, groups = value_states_33_groups_0, pad = value_states_33_pad_0, pad_type = value_states_33_pad_type_0, strides = value_states_33_strides_0, weight = locdit_decoder_layers_8_self_attn_v_proj_weight_cast_fp16, x = var_1492_cast_fp16_0)[name = string("value_states_33_cast_fp16")]; + tensor var_1519 = const()[name = string("op_1519"), val = tensor([2, 16, 128, 11])]; + tensor x_87_cast_fp16 = reshape(shape = var_1519, x = query_states_33_cast_fp16)[name = string("x_87_cast_fp16")]; + tensor var_1521 = const()[name = string("op_1521"), val = tensor([2, 2, 128, 11])]; + tensor x_89_cast_fp16 = reshape(shape = var_1521, x = key_states_33_cast_fp16)[name = string("x_89_cast_fp16")]; + tensor var_1523 = const()[name = string("op_1523"), val = tensor([2, 2, 128, 11])]; + tensor value_states_35_cast_fp16 = reshape(shape = var_1523, x = value_states_33_cast_fp16)[name = string("value_states_35_cast_fp16")]; + tensor var_1525_cast_fp16 = mul(x = x_87_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1525_cast_fp16")]; + tensor var_1526_split_sizes_0 = const()[name = string("op_1526_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1526_axis_0 = const()[name = string("op_1526_axis_0"), val = int32(-2)]; + tensor var_1526_cast_fp16_0, tensor var_1526_cast_fp16_1 = split(axis = var_1526_axis_0, split_sizes = var_1526_split_sizes_0, x = x_87_cast_fp16)[name = string("op_1526_cast_fp16")]; + bool var_1529_interleave_0 = const()[name = string("op_1529_interleave_0"), val = bool(false)]; + tensor var_1529_cast_fp16 = concat(axis = var_1456, interleave = var_1529_interleave_0, values = (var_1526_cast_fp16_1, var_1526_cast_fp16_0))[name = string("op_1529_cast_fp16")]; + tensor var_1530_cast_fp16 = mul(x = var_1529_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1530_cast_fp16")]; + tensor query_states_35_cast_fp16 = add(x = var_1525_cast_fp16, y = var_1530_cast_fp16)[name = string("query_states_35_cast_fp16")]; + tensor var_1532_cast_fp16 = mul(x = x_89_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1532_cast_fp16")]; + tensor var_1533_split_sizes_0 = const()[name = string("op_1533_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1533_axis_0 = const()[name = string("op_1533_axis_0"), val = int32(-2)]; + tensor var_1533_cast_fp16_0, tensor var_1533_cast_fp16_1 = split(axis = var_1533_axis_0, split_sizes = var_1533_split_sizes_0, x = x_89_cast_fp16)[name = string("op_1533_cast_fp16")]; + bool var_1536_interleave_0 = const()[name = string("op_1536_interleave_0"), val = bool(false)]; + tensor var_1536_cast_fp16 = concat(axis = var_1456, interleave = var_1536_interleave_0, values = (var_1533_cast_fp16_1, var_1533_cast_fp16_0))[name = string("op_1536_cast_fp16")]; + tensor var_1537_cast_fp16 = mul(x = var_1536_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1537_cast_fp16")]; + tensor key_states_35_cast_fp16 = add(x = var_1532_cast_fp16, y = var_1537_cast_fp16)[name = string("key_states_35_cast_fp16")]; + tensor var_1539_split_sizes_0 = const()[name = string("op_1539_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1539_axis_0 = const()[name = string("op_1539_axis_0"), val = int32(1)]; + tensor var_1539_cast_fp16_0, tensor var_1539_cast_fp16_1 = split(axis = var_1539_axis_0, split_sizes = var_1539_split_sizes_0, x = query_states_35_cast_fp16)[name = string("op_1539_cast_fp16")]; + tensor var_1541_split_sizes_0 = const()[name = string("op_1541_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1541_axis_0 = const()[name = string("op_1541_axis_0"), val = int32(1)]; + tensor var_1541_cast_fp16_0, tensor var_1541_cast_fp16_1 = split(axis = var_1541_axis_0, split_sizes = var_1541_split_sizes_0, x = key_states_35_cast_fp16)[name = string("op_1541_cast_fp16")]; + tensor var_1543_split_sizes_0 = const()[name = string("op_1543_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1543_axis_0 = const()[name = string("op_1543_axis_0"), val = int32(1)]; + tensor var_1543_cast_fp16_0, tensor var_1543_cast_fp16_1 = split(axis = var_1543_axis_0, split_sizes = var_1543_split_sizes_0, x = value_states_35_cast_fp16)[name = string("op_1543_cast_fp16")]; + bool attn_weights_97_transpose_x_1 = const()[name = string("attn_weights_97_transpose_x_1"), val = bool(true)]; + bool attn_weights_97_transpose_y_1 = const()[name = string("attn_weights_97_transpose_y_1"), val = bool(false)]; + tensor attn_weights_97_cast_fp16 = matmul(transpose_x = attn_weights_97_transpose_x_1, transpose_y = attn_weights_97_transpose_y_1, x = var_1541_cast_fp16_0, y = var_1539_cast_fp16_0)[name = string("attn_weights_97_cast_fp16")]; + fp16 var_1547_to_fp16 = const()[name = string("op_1547_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_99_cast_fp16 = mul(x = attn_weights_97_cast_fp16, y = var_1547_to_fp16)[name = string("attn_weights_99_cast_fp16")]; + tensor attn_weights_101_cast_fp16 = softmax(axis = var_1456, x = attn_weights_99_cast_fp16)[name = string("attn_weights_101_cast_fp16")]; + bool var_1550_transpose_x_0 = const()[name = string("op_1550_transpose_x_0"), val = bool(false)]; + bool var_1550_transpose_y_0 = const()[name = string("op_1550_transpose_y_0"), val = bool(false)]; + tensor var_1550_cast_fp16 = matmul(transpose_x = var_1550_transpose_x_0, transpose_y = var_1550_transpose_y_0, x = var_1543_cast_fp16_0, y = attn_weights_101_cast_fp16)[name = string("op_1550_cast_fp16")]; + bool attn_weights_103_transpose_x_1 = const()[name = string("attn_weights_103_transpose_x_1"), val = bool(true)]; + bool attn_weights_103_transpose_y_1 = const()[name = string("attn_weights_103_transpose_y_1"), val = bool(false)]; + tensor attn_weights_103_cast_fp16 = matmul(transpose_x = attn_weights_103_transpose_x_1, transpose_y = attn_weights_103_transpose_y_1, x = var_1541_cast_fp16_1, y = var_1539_cast_fp16_1)[name = string("attn_weights_103_cast_fp16")]; + fp16 var_1553_to_fp16 = const()[name = string("op_1553_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_105_cast_fp16 = mul(x = attn_weights_103_cast_fp16, y = var_1553_to_fp16)[name = string("attn_weights_105_cast_fp16")]; + tensor attn_weights_107_cast_fp16 = softmax(axis = var_1456, x = attn_weights_105_cast_fp16)[name = string("attn_weights_107_cast_fp16")]; + bool attn_out_17_transpose_x_0 = const()[name = string("attn_out_17_transpose_x_0"), val = bool(false)]; + bool attn_out_17_transpose_y_0 = const()[name = string("attn_out_17_transpose_y_0"), val = bool(false)]; + tensor attn_out_17_cast_fp16 = matmul(transpose_x = attn_out_17_transpose_x_0, transpose_y = attn_out_17_transpose_y_0, x = var_1543_cast_fp16_1, y = attn_weights_107_cast_fp16)[name = string("attn_out_17_cast_fp16")]; + bool attn_output_17_interleave_0 = const()[name = string("attn_output_17_interleave_0"), val = bool(false)]; + tensor attn_output_17_cast_fp16 = concat(axis = var_1458, interleave = attn_output_17_interleave_0, values = (var_1550_cast_fp16, attn_out_17_cast_fp16))[name = string("attn_output_17_cast_fp16")]; + tensor var_1559 = const()[name = string("op_1559"), val = tensor([2, 2048, 1, 11])]; + tensor x_91_cast_fp16 = reshape(shape = var_1559, x = attn_output_17_cast_fp16)[name = string("x_91_cast_fp16")]; + string hidden_states_85_pad_type_0 = const()[name = string("hidden_states_85_pad_type_0"), val = string("valid")]; + tensor hidden_states_85_strides_0 = const()[name = string("hidden_states_85_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_85_pad_0 = const()[name = string("hidden_states_85_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_85_dilations_0 = const()[name = string("hidden_states_85_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_85_groups_0 = const()[name = string("hidden_states_85_groups_0"), val = int32(1)]; + tensor hidden_states_85_cast_fp16 = conv(dilations = hidden_states_85_dilations_0, groups = hidden_states_85_groups_0, pad = hidden_states_85_pad_0, pad_type = hidden_states_85_pad_type_0, strides = hidden_states_85_strides_0, weight = locdit_decoder_layers_8_self_attn_o_proj_weight_cast_fp16, x = x_91_cast_fp16)[name = string("hidden_states_85_cast_fp16")]; + tensor hidden_states_87_cast_fp16 = add(x = hidden_states_81_cast_fp16, y = hidden_states_85_cast_fp16)[name = string("hidden_states_87_cast_fp16")]; + fp16 const_52_promoted_to_fp16 = const()[name = string("const_52_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1569_cast_fp16 = mul(x = hidden_states_87_cast_fp16, y = const_52_promoted_to_fp16)[name = string("op_1569_cast_fp16")]; + bool doubled_69_interleave_0 = const()[name = string("doubled_69_interleave_0"), val = bool(false)]; + tensor doubled_69_cast_fp16 = concat(axis = var_1458, interleave = doubled_69_interleave_0, values = (hidden_states_87_cast_fp16, var_1569_cast_fp16))[name = string("doubled_69_cast_fp16")]; + tensor out_35_axes_0 = const()[name = string("out_35_axes_0"), val = tensor([1])]; + tensor out_35_gamma_0_to_fp16 = const()[name = string("out_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212555520)))]; + fp16 var_1579_to_fp16 = const()[name = string("op_1579_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1579_to_fp16, gamma = out_35_gamma_0_to_fp16, x = doubled_69_cast_fp16)[name = string("out_35_cast_fp16")]; + tensor var_1590_split_sizes_0 = const()[name = string("op_1590_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1590_axis_0 = const()[name = string("op_1590_axis_0"), val = int32(1)]; + tensor var_1590_cast_fp16_0, tensor var_1590_cast_fp16_1 = split(axis = var_1590_axis_0, split_sizes = var_1590_split_sizes_0, x = out_35_cast_fp16)[name = string("op_1590_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor input_21_cast_fp16 = conv(dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = locdit_decoder_layers_8_mlp_gate_proj_weight_cast_fp16, x = var_1590_cast_fp16_0)[name = string("input_21_cast_fp16")]; + tensor var_1602_cast_fp16 = silu(x = input_21_cast_fp16)[name = string("op_1602_cast_fp16")]; + string var_1608_pad_type_0 = const()[name = string("op_1608_pad_type_0"), val = string("valid")]; + tensor var_1608_strides_0 = const()[name = string("op_1608_strides_0"), val = tensor([1, 1])]; + tensor var_1608_pad_0 = const()[name = string("op_1608_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1608_dilations_0 = const()[name = string("op_1608_dilations_0"), val = tensor([1, 1])]; + int32 var_1608_groups_0 = const()[name = string("op_1608_groups_0"), val = int32(1)]; + tensor var_1608_cast_fp16 = conv(dilations = var_1608_dilations_0, groups = var_1608_groups_0, pad = var_1608_pad_0, pad_type = var_1608_pad_type_0, strides = var_1608_strides_0, weight = locdit_decoder_layers_8_mlp_up_proj_weight_cast_fp16, x = var_1590_cast_fp16_0)[name = string("op_1608_cast_fp16")]; + tensor x_95_cast_fp16 = mul(x = var_1602_cast_fp16, y = var_1608_cast_fp16)[name = string("x_95_cast_fp16")]; + string hidden_states_89_pad_type_0 = const()[name = string("hidden_states_89_pad_type_0"), val = string("valid")]; + tensor hidden_states_89_strides_0 = const()[name = string("hidden_states_89_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_89_pad_0 = const()[name = string("hidden_states_89_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_89_dilations_0 = const()[name = string("hidden_states_89_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_89_groups_0 = const()[name = string("hidden_states_89_groups_0"), val = int32(1)]; + tensor hidden_states_89_cast_fp16 = conv(dilations = hidden_states_89_dilations_0, groups = hidden_states_89_groups_0, pad = hidden_states_89_pad_0, pad_type = hidden_states_89_pad_type_0, strides = hidden_states_89_strides_0, weight = locdit_decoder_layers_8_mlp_down_proj_weight_cast_fp16, x = x_95_cast_fp16)[name = string("hidden_states_89_cast_fp16")]; + tensor hidden_states_91_cast_fp16 = add(x = hidden_states_87_cast_fp16, y = hidden_states_89_cast_fp16)[name = string("hidden_states_91_cast_fp16")]; + int32 var_1619 = const()[name = string("op_1619"), val = int32(-2)]; + int32 var_1621 = const()[name = string("op_1621"), val = int32(1)]; + fp16 const_54_promoted_to_fp16 = const()[name = string("const_54_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1634_cast_fp16 = mul(x = hidden_states_91_cast_fp16, y = const_54_promoted_to_fp16)[name = string("op_1634_cast_fp16")]; + bool doubled_73_interleave_0 = const()[name = string("doubled_73_interleave_0"), val = bool(false)]; + tensor doubled_73_cast_fp16 = concat(axis = var_1621, interleave = doubled_73_interleave_0, values = (hidden_states_91_cast_fp16, var_1634_cast_fp16))[name = string("doubled_73_cast_fp16")]; + tensor out_37_axes_0 = const()[name = string("out_37_axes_0"), val = tensor([1])]; + tensor out_37_gamma_0_to_fp16 = const()[name = string("out_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212559680)))]; + fp16 var_1644_to_fp16 = const()[name = string("op_1644_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1644_to_fp16, gamma = out_37_gamma_0_to_fp16, x = doubled_73_cast_fp16)[name = string("out_37_cast_fp16")]; + tensor var_1655_split_sizes_0 = const()[name = string("op_1655_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1655_axis_0 = const()[name = string("op_1655_axis_0"), val = int32(1)]; + tensor var_1655_cast_fp16_0, tensor var_1655_cast_fp16_1 = split(axis = var_1655_axis_0, split_sizes = var_1655_split_sizes_0, x = out_37_cast_fp16)[name = string("op_1655_cast_fp16")]; + string query_states_37_pad_type_0 = const()[name = string("query_states_37_pad_type_0"), val = string("valid")]; + tensor query_states_37_strides_0 = const()[name = string("query_states_37_strides_0"), val = tensor([1, 1])]; + tensor query_states_37_pad_0 = const()[name = string("query_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_37_dilations_0 = const()[name = string("query_states_37_dilations_0"), val = tensor([1, 1])]; + int32 query_states_37_groups_0 = const()[name = string("query_states_37_groups_0"), val = int32(1)]; + tensor query_states_37_cast_fp16 = conv(dilations = query_states_37_dilations_0, groups = query_states_37_groups_0, pad = query_states_37_pad_0, pad_type = query_states_37_pad_type_0, strides = query_states_37_strides_0, weight = locdit_decoder_layers_9_self_attn_q_proj_weight_cast_fp16, x = var_1655_cast_fp16_0)[name = string("query_states_37_cast_fp16")]; + string key_states_37_pad_type_0 = const()[name = string("key_states_37_pad_type_0"), val = string("valid")]; + tensor key_states_37_strides_0 = const()[name = string("key_states_37_strides_0"), val = tensor([1, 1])]; + tensor key_states_37_pad_0 = const()[name = string("key_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_37_dilations_0 = const()[name = string("key_states_37_dilations_0"), val = tensor([1, 1])]; + int32 key_states_37_groups_0 = const()[name = string("key_states_37_groups_0"), val = int32(1)]; + tensor key_states_37_cast_fp16 = conv(dilations = key_states_37_dilations_0, groups = key_states_37_groups_0, pad = key_states_37_pad_0, pad_type = key_states_37_pad_type_0, strides = key_states_37_strides_0, weight = locdit_decoder_layers_9_self_attn_k_proj_weight_cast_fp16, x = var_1655_cast_fp16_0)[name = string("key_states_37_cast_fp16")]; + string value_states_37_pad_type_0 = const()[name = string("value_states_37_pad_type_0"), val = string("valid")]; + tensor value_states_37_strides_0 = const()[name = string("value_states_37_strides_0"), val = tensor([1, 1])]; + tensor value_states_37_pad_0 = const()[name = string("value_states_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_37_dilations_0 = const()[name = string("value_states_37_dilations_0"), val = tensor([1, 1])]; + int32 value_states_37_groups_0 = const()[name = string("value_states_37_groups_0"), val = int32(1)]; + tensor value_states_37_cast_fp16 = conv(dilations = value_states_37_dilations_0, groups = value_states_37_groups_0, pad = value_states_37_pad_0, pad_type = value_states_37_pad_type_0, strides = value_states_37_strides_0, weight = locdit_decoder_layers_9_self_attn_v_proj_weight_cast_fp16, x = var_1655_cast_fp16_0)[name = string("value_states_37_cast_fp16")]; + tensor var_1682 = const()[name = string("op_1682"), val = tensor([2, 16, 128, 11])]; + tensor x_97_cast_fp16 = reshape(shape = var_1682, x = query_states_37_cast_fp16)[name = string("x_97_cast_fp16")]; + tensor var_1684 = const()[name = string("op_1684"), val = tensor([2, 2, 128, 11])]; + tensor x_99_cast_fp16 = reshape(shape = var_1684, x = key_states_37_cast_fp16)[name = string("x_99_cast_fp16")]; + tensor var_1686 = const()[name = string("op_1686"), val = tensor([2, 2, 128, 11])]; + tensor value_states_39_cast_fp16 = reshape(shape = var_1686, x = value_states_37_cast_fp16)[name = string("value_states_39_cast_fp16")]; + tensor var_1688_cast_fp16 = mul(x = x_97_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1688_cast_fp16")]; + tensor var_1689_split_sizes_0 = const()[name = string("op_1689_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1689_axis_0 = const()[name = string("op_1689_axis_0"), val = int32(-2)]; + tensor var_1689_cast_fp16_0, tensor var_1689_cast_fp16_1 = split(axis = var_1689_axis_0, split_sizes = var_1689_split_sizes_0, x = x_97_cast_fp16)[name = string("op_1689_cast_fp16")]; + bool var_1692_interleave_0 = const()[name = string("op_1692_interleave_0"), val = bool(false)]; + tensor var_1692_cast_fp16 = concat(axis = var_1619, interleave = var_1692_interleave_0, values = (var_1689_cast_fp16_1, var_1689_cast_fp16_0))[name = string("op_1692_cast_fp16")]; + tensor var_1693_cast_fp16 = mul(x = var_1692_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1693_cast_fp16")]; + tensor query_states_39_cast_fp16 = add(x = var_1688_cast_fp16, y = var_1693_cast_fp16)[name = string("query_states_39_cast_fp16")]; + tensor var_1695_cast_fp16 = mul(x = x_99_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1695_cast_fp16")]; + tensor var_1696_split_sizes_0 = const()[name = string("op_1696_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1696_axis_0 = const()[name = string("op_1696_axis_0"), val = int32(-2)]; + tensor var_1696_cast_fp16_0, tensor var_1696_cast_fp16_1 = split(axis = var_1696_axis_0, split_sizes = var_1696_split_sizes_0, x = x_99_cast_fp16)[name = string("op_1696_cast_fp16")]; + bool var_1699_interleave_0 = const()[name = string("op_1699_interleave_0"), val = bool(false)]; + tensor var_1699_cast_fp16 = concat(axis = var_1619, interleave = var_1699_interleave_0, values = (var_1696_cast_fp16_1, var_1696_cast_fp16_0))[name = string("op_1699_cast_fp16")]; + tensor var_1700_cast_fp16 = mul(x = var_1699_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1700_cast_fp16")]; + tensor key_states_39_cast_fp16 = add(x = var_1695_cast_fp16, y = var_1700_cast_fp16)[name = string("key_states_39_cast_fp16")]; + tensor var_1702_split_sizes_0 = const()[name = string("op_1702_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1702_axis_0 = const()[name = string("op_1702_axis_0"), val = int32(1)]; + tensor var_1702_cast_fp16_0, tensor var_1702_cast_fp16_1 = split(axis = var_1702_axis_0, split_sizes = var_1702_split_sizes_0, x = query_states_39_cast_fp16)[name = string("op_1702_cast_fp16")]; + tensor var_1704_split_sizes_0 = const()[name = string("op_1704_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1704_axis_0 = const()[name = string("op_1704_axis_0"), val = int32(1)]; + tensor var_1704_cast_fp16_0, tensor var_1704_cast_fp16_1 = split(axis = var_1704_axis_0, split_sizes = var_1704_split_sizes_0, x = key_states_39_cast_fp16)[name = string("op_1704_cast_fp16")]; + tensor var_1706_split_sizes_0 = const()[name = string("op_1706_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1706_axis_0 = const()[name = string("op_1706_axis_0"), val = int32(1)]; + tensor var_1706_cast_fp16_0, tensor var_1706_cast_fp16_1 = split(axis = var_1706_axis_0, split_sizes = var_1706_split_sizes_0, x = value_states_39_cast_fp16)[name = string("op_1706_cast_fp16")]; + bool attn_weights_109_transpose_x_1 = const()[name = string("attn_weights_109_transpose_x_1"), val = bool(true)]; + bool attn_weights_109_transpose_y_1 = const()[name = string("attn_weights_109_transpose_y_1"), val = bool(false)]; + tensor attn_weights_109_cast_fp16 = matmul(transpose_x = attn_weights_109_transpose_x_1, transpose_y = attn_weights_109_transpose_y_1, x = var_1704_cast_fp16_0, y = var_1702_cast_fp16_0)[name = string("attn_weights_109_cast_fp16")]; + fp16 var_1710_to_fp16 = const()[name = string("op_1710_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_111_cast_fp16 = mul(x = attn_weights_109_cast_fp16, y = var_1710_to_fp16)[name = string("attn_weights_111_cast_fp16")]; + tensor attn_weights_113_cast_fp16 = softmax(axis = var_1619, x = attn_weights_111_cast_fp16)[name = string("attn_weights_113_cast_fp16")]; + bool var_1713_transpose_x_0 = const()[name = string("op_1713_transpose_x_0"), val = bool(false)]; + bool var_1713_transpose_y_0 = const()[name = string("op_1713_transpose_y_0"), val = bool(false)]; + tensor var_1713_cast_fp16 = matmul(transpose_x = var_1713_transpose_x_0, transpose_y = var_1713_transpose_y_0, x = var_1706_cast_fp16_0, y = attn_weights_113_cast_fp16)[name = string("op_1713_cast_fp16")]; + bool attn_weights_115_transpose_x_1 = const()[name = string("attn_weights_115_transpose_x_1"), val = bool(true)]; + bool attn_weights_115_transpose_y_1 = const()[name = string("attn_weights_115_transpose_y_1"), val = bool(false)]; + tensor attn_weights_115_cast_fp16 = matmul(transpose_x = attn_weights_115_transpose_x_1, transpose_y = attn_weights_115_transpose_y_1, x = var_1704_cast_fp16_1, y = var_1702_cast_fp16_1)[name = string("attn_weights_115_cast_fp16")]; + fp16 var_1716_to_fp16 = const()[name = string("op_1716_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_117_cast_fp16 = mul(x = attn_weights_115_cast_fp16, y = var_1716_to_fp16)[name = string("attn_weights_117_cast_fp16")]; + tensor attn_weights_119_cast_fp16 = softmax(axis = var_1619, x = attn_weights_117_cast_fp16)[name = string("attn_weights_119_cast_fp16")]; + bool attn_out_19_transpose_x_0 = const()[name = string("attn_out_19_transpose_x_0"), val = bool(false)]; + bool attn_out_19_transpose_y_0 = const()[name = string("attn_out_19_transpose_y_0"), val = bool(false)]; + tensor attn_out_19_cast_fp16 = matmul(transpose_x = attn_out_19_transpose_x_0, transpose_y = attn_out_19_transpose_y_0, x = var_1706_cast_fp16_1, y = attn_weights_119_cast_fp16)[name = string("attn_out_19_cast_fp16")]; + bool attn_output_19_interleave_0 = const()[name = string("attn_output_19_interleave_0"), val = bool(false)]; + tensor attn_output_19_cast_fp16 = concat(axis = var_1621, interleave = attn_output_19_interleave_0, values = (var_1713_cast_fp16, attn_out_19_cast_fp16))[name = string("attn_output_19_cast_fp16")]; + tensor var_1722 = const()[name = string("op_1722"), val = tensor([2, 2048, 1, 11])]; + tensor x_101_cast_fp16 = reshape(shape = var_1722, x = attn_output_19_cast_fp16)[name = string("x_101_cast_fp16")]; + string hidden_states_95_pad_type_0 = const()[name = string("hidden_states_95_pad_type_0"), val = string("valid")]; + tensor hidden_states_95_strides_0 = const()[name = string("hidden_states_95_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_95_pad_0 = const()[name = string("hidden_states_95_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_95_dilations_0 = const()[name = string("hidden_states_95_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_95_groups_0 = const()[name = string("hidden_states_95_groups_0"), val = int32(1)]; + tensor hidden_states_95_cast_fp16 = conv(dilations = hidden_states_95_dilations_0, groups = hidden_states_95_groups_0, pad = hidden_states_95_pad_0, pad_type = hidden_states_95_pad_type_0, strides = hidden_states_95_strides_0, weight = locdit_decoder_layers_9_self_attn_o_proj_weight_cast_fp16, x = x_101_cast_fp16)[name = string("hidden_states_95_cast_fp16")]; + tensor hidden_states_97_cast_fp16 = add(x = hidden_states_91_cast_fp16, y = hidden_states_95_cast_fp16)[name = string("hidden_states_97_cast_fp16")]; + fp16 const_58_promoted_to_fp16 = const()[name = string("const_58_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1732_cast_fp16 = mul(x = hidden_states_97_cast_fp16, y = const_58_promoted_to_fp16)[name = string("op_1732_cast_fp16")]; + bool doubled_77_interleave_0 = const()[name = string("doubled_77_interleave_0"), val = bool(false)]; + tensor doubled_77_cast_fp16 = concat(axis = var_1621, interleave = doubled_77_interleave_0, values = (hidden_states_97_cast_fp16, var_1732_cast_fp16))[name = string("doubled_77_cast_fp16")]; + tensor out_39_axes_0 = const()[name = string("out_39_axes_0"), val = tensor([1])]; + tensor out_39_gamma_0_to_fp16 = const()[name = string("out_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212563840)))]; + fp16 var_1742_to_fp16 = const()[name = string("op_1742_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1742_to_fp16, gamma = out_39_gamma_0_to_fp16, x = doubled_77_cast_fp16)[name = string("out_39_cast_fp16")]; + tensor var_1753_split_sizes_0 = const()[name = string("op_1753_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1753_axis_0 = const()[name = string("op_1753_axis_0"), val = int32(1)]; + tensor var_1753_cast_fp16_0, tensor var_1753_cast_fp16_1 = split(axis = var_1753_axis_0, split_sizes = var_1753_split_sizes_0, x = out_39_cast_fp16)[name = string("op_1753_cast_fp16")]; + string input_23_pad_type_0 = const()[name = string("input_23_pad_type_0"), val = string("valid")]; + tensor input_23_strides_0 = const()[name = string("input_23_strides_0"), val = tensor([1, 1])]; + tensor input_23_pad_0 = const()[name = string("input_23_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_23_dilations_0 = const()[name = string("input_23_dilations_0"), val = tensor([1, 1])]; + int32 input_23_groups_0 = const()[name = string("input_23_groups_0"), val = int32(1)]; + tensor input_23_cast_fp16 = conv(dilations = input_23_dilations_0, groups = input_23_groups_0, pad = input_23_pad_0, pad_type = input_23_pad_type_0, strides = input_23_strides_0, weight = locdit_decoder_layers_9_mlp_gate_proj_weight_cast_fp16, x = var_1753_cast_fp16_0)[name = string("input_23_cast_fp16")]; + tensor var_1765_cast_fp16 = silu(x = input_23_cast_fp16)[name = string("op_1765_cast_fp16")]; + string var_1771_pad_type_0 = const()[name = string("op_1771_pad_type_0"), val = string("valid")]; + tensor var_1771_strides_0 = const()[name = string("op_1771_strides_0"), val = tensor([1, 1])]; + tensor var_1771_pad_0 = const()[name = string("op_1771_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1771_dilations_0 = const()[name = string("op_1771_dilations_0"), val = tensor([1, 1])]; + int32 var_1771_groups_0 = const()[name = string("op_1771_groups_0"), val = int32(1)]; + tensor var_1771_cast_fp16 = conv(dilations = var_1771_dilations_0, groups = var_1771_groups_0, pad = var_1771_pad_0, pad_type = var_1771_pad_type_0, strides = var_1771_strides_0, weight = locdit_decoder_layers_9_mlp_up_proj_weight_cast_fp16, x = var_1753_cast_fp16_0)[name = string("op_1771_cast_fp16")]; + tensor x_105_cast_fp16 = mul(x = var_1765_cast_fp16, y = var_1771_cast_fp16)[name = string("x_105_cast_fp16")]; + string hidden_states_99_pad_type_0 = const()[name = string("hidden_states_99_pad_type_0"), val = string("valid")]; + tensor hidden_states_99_strides_0 = const()[name = string("hidden_states_99_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_99_pad_0 = const()[name = string("hidden_states_99_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_99_dilations_0 = const()[name = string("hidden_states_99_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_99_groups_0 = const()[name = string("hidden_states_99_groups_0"), val = int32(1)]; + tensor hidden_states_99_cast_fp16 = conv(dilations = hidden_states_99_dilations_0, groups = hidden_states_99_groups_0, pad = hidden_states_99_pad_0, pad_type = hidden_states_99_pad_type_0, strides = hidden_states_99_strides_0, weight = locdit_decoder_layers_9_mlp_down_proj_weight_cast_fp16, x = x_105_cast_fp16)[name = string("hidden_states_99_cast_fp16")]; + tensor hidden_states_101_cast_fp16 = add(x = hidden_states_97_cast_fp16, y = hidden_states_99_cast_fp16)[name = string("hidden_states_101_cast_fp16")]; + int32 var_1782 = const()[name = string("op_1782"), val = int32(-2)]; + int32 var_1784 = const()[name = string("op_1784"), val = int32(1)]; + fp16 const_60_promoted_to_fp16 = const()[name = string("const_60_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1797_cast_fp16 = mul(x = hidden_states_101_cast_fp16, y = const_60_promoted_to_fp16)[name = string("op_1797_cast_fp16")]; + bool doubled_81_interleave_0 = const()[name = string("doubled_81_interleave_0"), val = bool(false)]; + tensor doubled_81_cast_fp16 = concat(axis = var_1784, interleave = doubled_81_interleave_0, values = (hidden_states_101_cast_fp16, var_1797_cast_fp16))[name = string("doubled_81_cast_fp16")]; + tensor out_41_axes_0 = const()[name = string("out_41_axes_0"), val = tensor([1])]; + tensor out_41_gamma_0_to_fp16 = const()[name = string("out_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212568000)))]; + fp16 var_1807_to_fp16 = const()[name = string("op_1807_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1807_to_fp16, gamma = out_41_gamma_0_to_fp16, x = doubled_81_cast_fp16)[name = string("out_41_cast_fp16")]; + tensor var_1818_split_sizes_0 = const()[name = string("op_1818_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1818_axis_0 = const()[name = string("op_1818_axis_0"), val = int32(1)]; + tensor var_1818_cast_fp16_0, tensor var_1818_cast_fp16_1 = split(axis = var_1818_axis_0, split_sizes = var_1818_split_sizes_0, x = out_41_cast_fp16)[name = string("op_1818_cast_fp16")]; + string query_states_41_pad_type_0 = const()[name = string("query_states_41_pad_type_0"), val = string("valid")]; + tensor query_states_41_strides_0 = const()[name = string("query_states_41_strides_0"), val = tensor([1, 1])]; + tensor query_states_41_pad_0 = const()[name = string("query_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_41_dilations_0 = const()[name = string("query_states_41_dilations_0"), val = tensor([1, 1])]; + int32 query_states_41_groups_0 = const()[name = string("query_states_41_groups_0"), val = int32(1)]; + tensor query_states_41_cast_fp16 = conv(dilations = query_states_41_dilations_0, groups = query_states_41_groups_0, pad = query_states_41_pad_0, pad_type = query_states_41_pad_type_0, strides = query_states_41_strides_0, weight = locdit_decoder_layers_10_self_attn_q_proj_weight_cast_fp16, x = var_1818_cast_fp16_0)[name = string("query_states_41_cast_fp16")]; + string key_states_41_pad_type_0 = const()[name = string("key_states_41_pad_type_0"), val = string("valid")]; + tensor key_states_41_strides_0 = const()[name = string("key_states_41_strides_0"), val = tensor([1, 1])]; + tensor key_states_41_pad_0 = const()[name = string("key_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_41_dilations_0 = const()[name = string("key_states_41_dilations_0"), val = tensor([1, 1])]; + int32 key_states_41_groups_0 = const()[name = string("key_states_41_groups_0"), val = int32(1)]; + tensor key_states_41_cast_fp16 = conv(dilations = key_states_41_dilations_0, groups = key_states_41_groups_0, pad = key_states_41_pad_0, pad_type = key_states_41_pad_type_0, strides = key_states_41_strides_0, weight = locdit_decoder_layers_10_self_attn_k_proj_weight_cast_fp16, x = var_1818_cast_fp16_0)[name = string("key_states_41_cast_fp16")]; + string value_states_41_pad_type_0 = const()[name = string("value_states_41_pad_type_0"), val = string("valid")]; + tensor value_states_41_strides_0 = const()[name = string("value_states_41_strides_0"), val = tensor([1, 1])]; + tensor value_states_41_pad_0 = const()[name = string("value_states_41_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_41_dilations_0 = const()[name = string("value_states_41_dilations_0"), val = tensor([1, 1])]; + int32 value_states_41_groups_0 = const()[name = string("value_states_41_groups_0"), val = int32(1)]; + tensor value_states_41_cast_fp16 = conv(dilations = value_states_41_dilations_0, groups = value_states_41_groups_0, pad = value_states_41_pad_0, pad_type = value_states_41_pad_type_0, strides = value_states_41_strides_0, weight = locdit_decoder_layers_10_self_attn_v_proj_weight_cast_fp16, x = var_1818_cast_fp16_0)[name = string("value_states_41_cast_fp16")]; + tensor var_1845 = const()[name = string("op_1845"), val = tensor([2, 16, 128, 11])]; + tensor x_107_cast_fp16 = reshape(shape = var_1845, x = query_states_41_cast_fp16)[name = string("x_107_cast_fp16")]; + tensor var_1847 = const()[name = string("op_1847"), val = tensor([2, 2, 128, 11])]; + tensor x_109_cast_fp16 = reshape(shape = var_1847, x = key_states_41_cast_fp16)[name = string("x_109_cast_fp16")]; + tensor var_1849 = const()[name = string("op_1849"), val = tensor([2, 2, 128, 11])]; + tensor value_states_43_cast_fp16 = reshape(shape = var_1849, x = value_states_41_cast_fp16)[name = string("value_states_43_cast_fp16")]; + tensor var_1851_cast_fp16 = mul(x = x_107_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1851_cast_fp16")]; + tensor var_1852_split_sizes_0 = const()[name = string("op_1852_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1852_axis_0 = const()[name = string("op_1852_axis_0"), val = int32(-2)]; + tensor var_1852_cast_fp16_0, tensor var_1852_cast_fp16_1 = split(axis = var_1852_axis_0, split_sizes = var_1852_split_sizes_0, x = x_107_cast_fp16)[name = string("op_1852_cast_fp16")]; + bool var_1855_interleave_0 = const()[name = string("op_1855_interleave_0"), val = bool(false)]; + tensor var_1855_cast_fp16 = concat(axis = var_1782, interleave = var_1855_interleave_0, values = (var_1852_cast_fp16_1, var_1852_cast_fp16_0))[name = string("op_1855_cast_fp16")]; + tensor var_1856_cast_fp16 = mul(x = var_1855_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1856_cast_fp16")]; + tensor query_states_43_cast_fp16 = add(x = var_1851_cast_fp16, y = var_1856_cast_fp16)[name = string("query_states_43_cast_fp16")]; + tensor var_1858_cast_fp16 = mul(x = x_109_cast_fp16, y = rope_cos_to_fp16)[name = string("op_1858_cast_fp16")]; + tensor var_1859_split_sizes_0 = const()[name = string("op_1859_split_sizes_0"), val = tensor([64, 64])]; + int32 var_1859_axis_0 = const()[name = string("op_1859_axis_0"), val = int32(-2)]; + tensor var_1859_cast_fp16_0, tensor var_1859_cast_fp16_1 = split(axis = var_1859_axis_0, split_sizes = var_1859_split_sizes_0, x = x_109_cast_fp16)[name = string("op_1859_cast_fp16")]; + bool var_1862_interleave_0 = const()[name = string("op_1862_interleave_0"), val = bool(false)]; + tensor var_1862_cast_fp16 = concat(axis = var_1782, interleave = var_1862_interleave_0, values = (var_1859_cast_fp16_1, var_1859_cast_fp16_0))[name = string("op_1862_cast_fp16")]; + tensor var_1863_cast_fp16 = mul(x = var_1862_cast_fp16, y = rope_sin_to_fp16)[name = string("op_1863_cast_fp16")]; + tensor key_states_43_cast_fp16 = add(x = var_1858_cast_fp16, y = var_1863_cast_fp16)[name = string("key_states_43_cast_fp16")]; + tensor var_1865_split_sizes_0 = const()[name = string("op_1865_split_sizes_0"), val = tensor([8, 8])]; + int32 var_1865_axis_0 = const()[name = string("op_1865_axis_0"), val = int32(1)]; + tensor var_1865_cast_fp16_0, tensor var_1865_cast_fp16_1 = split(axis = var_1865_axis_0, split_sizes = var_1865_split_sizes_0, x = query_states_43_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1867_split_sizes_0 = const()[name = string("op_1867_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1867_axis_0 = const()[name = string("op_1867_axis_0"), val = int32(1)]; + tensor var_1867_cast_fp16_0, tensor var_1867_cast_fp16_1 = split(axis = var_1867_axis_0, split_sizes = var_1867_split_sizes_0, x = key_states_43_cast_fp16)[name = string("op_1867_cast_fp16")]; + tensor var_1869_split_sizes_0 = const()[name = string("op_1869_split_sizes_0"), val = tensor([1, 1])]; + int32 var_1869_axis_0 = const()[name = string("op_1869_axis_0"), val = int32(1)]; + tensor var_1869_cast_fp16_0, tensor var_1869_cast_fp16_1 = split(axis = var_1869_axis_0, split_sizes = var_1869_split_sizes_0, x = value_states_43_cast_fp16)[name = string("op_1869_cast_fp16")]; + bool attn_weights_121_transpose_x_1 = const()[name = string("attn_weights_121_transpose_x_1"), val = bool(true)]; + bool attn_weights_121_transpose_y_1 = const()[name = string("attn_weights_121_transpose_y_1"), val = bool(false)]; + tensor attn_weights_121_cast_fp16 = matmul(transpose_x = attn_weights_121_transpose_x_1, transpose_y = attn_weights_121_transpose_y_1, x = var_1867_cast_fp16_0, y = var_1865_cast_fp16_0)[name = string("attn_weights_121_cast_fp16")]; + fp16 var_1873_to_fp16 = const()[name = string("op_1873_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_123_cast_fp16 = mul(x = attn_weights_121_cast_fp16, y = var_1873_to_fp16)[name = string("attn_weights_123_cast_fp16")]; + tensor attn_weights_125_cast_fp16 = softmax(axis = var_1782, x = attn_weights_123_cast_fp16)[name = string("attn_weights_125_cast_fp16")]; + bool var_1876_transpose_x_0 = const()[name = string("op_1876_transpose_x_0"), val = bool(false)]; + bool var_1876_transpose_y_0 = const()[name = string("op_1876_transpose_y_0"), val = bool(false)]; + tensor var_1876_cast_fp16 = matmul(transpose_x = var_1876_transpose_x_0, transpose_y = var_1876_transpose_y_0, x = var_1869_cast_fp16_0, y = attn_weights_125_cast_fp16)[name = string("op_1876_cast_fp16")]; + bool attn_weights_127_transpose_x_1 = const()[name = string("attn_weights_127_transpose_x_1"), val = bool(true)]; + bool attn_weights_127_transpose_y_1 = const()[name = string("attn_weights_127_transpose_y_1"), val = bool(false)]; + tensor attn_weights_127_cast_fp16 = matmul(transpose_x = attn_weights_127_transpose_x_1, transpose_y = attn_weights_127_transpose_y_1, x = var_1867_cast_fp16_1, y = var_1865_cast_fp16_1)[name = string("attn_weights_127_cast_fp16")]; + fp16 var_1879_to_fp16 = const()[name = string("op_1879_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_129_cast_fp16 = mul(x = attn_weights_127_cast_fp16, y = var_1879_to_fp16)[name = string("attn_weights_129_cast_fp16")]; + tensor attn_weights_131_cast_fp16 = softmax(axis = var_1782, x = attn_weights_129_cast_fp16)[name = string("attn_weights_131_cast_fp16")]; + bool attn_out_21_transpose_x_0 = const()[name = string("attn_out_21_transpose_x_0"), val = bool(false)]; + bool attn_out_21_transpose_y_0 = const()[name = string("attn_out_21_transpose_y_0"), val = bool(false)]; + tensor attn_out_21_cast_fp16 = matmul(transpose_x = attn_out_21_transpose_x_0, transpose_y = attn_out_21_transpose_y_0, x = var_1869_cast_fp16_1, y = attn_weights_131_cast_fp16)[name = string("attn_out_21_cast_fp16")]; + bool attn_output_21_interleave_0 = const()[name = string("attn_output_21_interleave_0"), val = bool(false)]; + tensor attn_output_21_cast_fp16 = concat(axis = var_1784, interleave = attn_output_21_interleave_0, values = (var_1876_cast_fp16, attn_out_21_cast_fp16))[name = string("attn_output_21_cast_fp16")]; + tensor var_1885 = const()[name = string("op_1885"), val = tensor([2, 2048, 1, 11])]; + tensor x_111_cast_fp16 = reshape(shape = var_1885, x = attn_output_21_cast_fp16)[name = string("x_111_cast_fp16")]; + string hidden_states_105_pad_type_0 = const()[name = string("hidden_states_105_pad_type_0"), val = string("valid")]; + tensor hidden_states_105_strides_0 = const()[name = string("hidden_states_105_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_105_pad_0 = const()[name = string("hidden_states_105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_105_dilations_0 = const()[name = string("hidden_states_105_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_105_groups_0 = const()[name = string("hidden_states_105_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212572160)))]; + tensor hidden_states_105_cast_fp16 = conv(dilations = hidden_states_105_dilations_0, groups = hidden_states_105_groups_0, pad = hidden_states_105_pad_0, pad_type = hidden_states_105_pad_type_0, strides = hidden_states_105_strides_0, weight = locdit_decoder_layers_10_self_attn_o_proj_weight_to_fp16, x = x_111_cast_fp16)[name = string("hidden_states_105_cast_fp16")]; + tensor hidden_states_107_cast_fp16 = add(x = hidden_states_101_cast_fp16, y = hidden_states_105_cast_fp16)[name = string("hidden_states_107_cast_fp16")]; + fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1895_cast_fp16 = mul(x = hidden_states_107_cast_fp16, y = const_64_promoted_to_fp16)[name = string("op_1895_cast_fp16")]; + bool doubled_85_interleave_0 = const()[name = string("doubled_85_interleave_0"), val = bool(false)]; + tensor doubled_85_cast_fp16 = concat(axis = var_1784, interleave = doubled_85_interleave_0, values = (hidden_states_107_cast_fp16, var_1895_cast_fp16))[name = string("doubled_85_cast_fp16")]; + tensor out_43_axes_0 = const()[name = string("out_43_axes_0"), val = tensor([1])]; + tensor out_43_gamma_0_to_fp16 = const()[name = string("out_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216766528)))]; + fp16 var_1905_to_fp16 = const()[name = string("op_1905_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1905_to_fp16, gamma = out_43_gamma_0_to_fp16, x = doubled_85_cast_fp16)[name = string("out_43_cast_fp16")]; + tensor var_1916_split_sizes_0 = const()[name = string("op_1916_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1916_axis_0 = const()[name = string("op_1916_axis_0"), val = int32(1)]; + tensor var_1916_cast_fp16_0, tensor var_1916_cast_fp16_1 = split(axis = var_1916_axis_0, split_sizes = var_1916_split_sizes_0, x = out_43_cast_fp16)[name = string("op_1916_cast_fp16")]; + string input_25_pad_type_0 = const()[name = string("input_25_pad_type_0"), val = string("valid")]; + tensor input_25_strides_0 = const()[name = string("input_25_strides_0"), val = tensor([1, 1])]; + tensor input_25_pad_0 = const()[name = string("input_25_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_25_dilations_0 = const()[name = string("input_25_dilations_0"), val = tensor([1, 1])]; + int32 input_25_groups_0 = const()[name = string("input_25_groups_0"), val = int32(1)]; + tensor input_25_cast_fp16 = conv(dilations = input_25_dilations_0, groups = input_25_groups_0, pad = input_25_pad_0, pad_type = input_25_pad_type_0, strides = input_25_strides_0, weight = locdit_decoder_layers_10_mlp_gate_proj_weight_cast_fp16, x = var_1916_cast_fp16_0)[name = string("input_25_cast_fp16")]; + tensor var_1928_cast_fp16 = silu(x = input_25_cast_fp16)[name = string("op_1928_cast_fp16")]; + string var_1934_pad_type_0 = const()[name = string("op_1934_pad_type_0"), val = string("valid")]; + tensor var_1934_strides_0 = const()[name = string("op_1934_strides_0"), val = tensor([1, 1])]; + tensor var_1934_pad_0 = const()[name = string("op_1934_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1934_dilations_0 = const()[name = string("op_1934_dilations_0"), val = tensor([1, 1])]; + int32 var_1934_groups_0 = const()[name = string("op_1934_groups_0"), val = int32(1)]; + tensor var_1934_cast_fp16 = conv(dilations = var_1934_dilations_0, groups = var_1934_groups_0, pad = var_1934_pad_0, pad_type = var_1934_pad_type_0, strides = var_1934_strides_0, weight = locdit_decoder_layers_10_mlp_up_proj_weight_cast_fp16, x = var_1916_cast_fp16_0)[name = string("op_1934_cast_fp16")]; + tensor x_115_cast_fp16 = mul(x = var_1928_cast_fp16, y = var_1934_cast_fp16)[name = string("x_115_cast_fp16")]; + string hidden_states_109_pad_type_0 = const()[name = string("hidden_states_109_pad_type_0"), val = string("valid")]; + tensor hidden_states_109_strides_0 = const()[name = string("hidden_states_109_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_109_pad_0 = const()[name = string("hidden_states_109_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_109_dilations_0 = const()[name = string("hidden_states_109_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_109_groups_0 = const()[name = string("hidden_states_109_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_10_mlp_down_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_10_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216770688)))]; + tensor hidden_states_109_cast_fp16 = conv(dilations = hidden_states_109_dilations_0, groups = hidden_states_109_groups_0, pad = hidden_states_109_pad_0, pad_type = hidden_states_109_pad_type_0, strides = hidden_states_109_strides_0, weight = locdit_decoder_layers_10_mlp_down_proj_weight_to_fp16, x = x_115_cast_fp16)[name = string("hidden_states_109_cast_fp16")]; + tensor hidden_states_111_cast_fp16 = add(x = hidden_states_107_cast_fp16, y = hidden_states_109_cast_fp16)[name = string("hidden_states_111_cast_fp16")]; + int32 var_1945 = const()[name = string("op_1945"), val = int32(-2)]; + int32 var_1947 = const()[name = string("op_1947"), val = int32(1)]; + fp16 const_66_promoted_to_fp16 = const()[name = string("const_66_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_1960_cast_fp16 = mul(x = hidden_states_111_cast_fp16, y = const_66_promoted_to_fp16)[name = string("op_1960_cast_fp16")]; + bool doubled_89_interleave_0 = const()[name = string("doubled_89_interleave_0"), val = bool(false)]; + tensor doubled_89_cast_fp16 = concat(axis = var_1947, interleave = doubled_89_interleave_0, values = (hidden_states_111_cast_fp16, var_1960_cast_fp16))[name = string("doubled_89_cast_fp16")]; + tensor out_45_axes_0 = const()[name = string("out_45_axes_0"), val = tensor([1])]; + tensor out_45_gamma_0_to_fp16 = const()[name = string("out_45_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225159360)))]; + fp16 var_1970_to_fp16 = const()[name = string("op_1970_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1970_to_fp16, gamma = out_45_gamma_0_to_fp16, x = doubled_89_cast_fp16)[name = string("out_45_cast_fp16")]; + tensor var_1981_split_sizes_0 = const()[name = string("op_1981_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_1981_axis_0 = const()[name = string("op_1981_axis_0"), val = int32(1)]; + tensor var_1981_cast_fp16_0, tensor var_1981_cast_fp16_1 = split(axis = var_1981_axis_0, split_sizes = var_1981_split_sizes_0, x = out_45_cast_fp16)[name = string("op_1981_cast_fp16")]; + string query_states_45_pad_type_0 = const()[name = string("query_states_45_pad_type_0"), val = string("valid")]; + tensor query_states_45_strides_0 = const()[name = string("query_states_45_strides_0"), val = tensor([1, 1])]; + tensor query_states_45_pad_0 = const()[name = string("query_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_states_45_dilations_0 = const()[name = string("query_states_45_dilations_0"), val = tensor([1, 1])]; + int32 query_states_45_groups_0 = const()[name = string("query_states_45_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225163520)))]; + tensor query_states_45_cast_fp16 = conv(dilations = query_states_45_dilations_0, groups = query_states_45_groups_0, pad = query_states_45_pad_0, pad_type = query_states_45_pad_type_0, strides = query_states_45_strides_0, weight = locdit_decoder_layers_11_self_attn_q_proj_weight_to_fp16, x = var_1981_cast_fp16_0)[name = string("query_states_45_cast_fp16")]; + string key_states_45_pad_type_0 = const()[name = string("key_states_45_pad_type_0"), val = string("valid")]; + tensor key_states_45_strides_0 = const()[name = string("key_states_45_strides_0"), val = tensor([1, 1])]; + tensor key_states_45_pad_0 = const()[name = string("key_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_states_45_dilations_0 = const()[name = string("key_states_45_dilations_0"), val = tensor([1, 1])]; + int32 key_states_45_groups_0 = const()[name = string("key_states_45_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229357888)))]; + tensor key_states_45_cast_fp16 = conv(dilations = key_states_45_dilations_0, groups = key_states_45_groups_0, pad = key_states_45_pad_0, pad_type = key_states_45_pad_type_0, strides = key_states_45_strides_0, weight = locdit_decoder_layers_11_self_attn_k_proj_weight_to_fp16, x = var_1981_cast_fp16_0)[name = string("key_states_45_cast_fp16")]; + string value_states_45_pad_type_0 = const()[name = string("value_states_45_pad_type_0"), val = string("valid")]; + tensor value_states_45_strides_0 = const()[name = string("value_states_45_strides_0"), val = tensor([1, 1])]; + tensor value_states_45_pad_0 = const()[name = string("value_states_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_states_45_dilations_0 = const()[name = string("value_states_45_dilations_0"), val = tensor([1, 1])]; + int32 value_states_45_groups_0 = const()[name = string("value_states_45_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229882240)))]; + tensor value_states_45_cast_fp16 = conv(dilations = value_states_45_dilations_0, groups = value_states_45_groups_0, pad = value_states_45_pad_0, pad_type = value_states_45_pad_type_0, strides = value_states_45_strides_0, weight = locdit_decoder_layers_11_self_attn_v_proj_weight_to_fp16, x = var_1981_cast_fp16_0)[name = string("value_states_45_cast_fp16")]; + tensor var_2008 = const()[name = string("op_2008"), val = tensor([2, 16, 128, 11])]; + tensor x_117_cast_fp16 = reshape(shape = var_2008, x = query_states_45_cast_fp16)[name = string("x_117_cast_fp16")]; + tensor var_2010 = const()[name = string("op_2010"), val = tensor([2, 2, 128, 11])]; + tensor x_119_cast_fp16 = reshape(shape = var_2010, x = key_states_45_cast_fp16)[name = string("x_119_cast_fp16")]; + tensor var_2012 = const()[name = string("op_2012"), val = tensor([2, 2, 128, 11])]; + tensor value_states_cast_fp16 = reshape(shape = var_2012, x = value_states_45_cast_fp16)[name = string("value_states_cast_fp16")]; + tensor var_2014_cast_fp16 = mul(x = x_117_cast_fp16, y = rope_cos_to_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2015_split_sizes_0 = const()[name = string("op_2015_split_sizes_0"), val = tensor([64, 64])]; + int32 var_2015_axis_0 = const()[name = string("op_2015_axis_0"), val = int32(-2)]; + tensor var_2015_cast_fp16_0, tensor var_2015_cast_fp16_1 = split(axis = var_2015_axis_0, split_sizes = var_2015_split_sizes_0, x = x_117_cast_fp16)[name = string("op_2015_cast_fp16")]; + bool var_2018_interleave_0 = const()[name = string("op_2018_interleave_0"), val = bool(false)]; + tensor var_2018_cast_fp16 = concat(axis = var_1945, interleave = var_2018_interleave_0, values = (var_2015_cast_fp16_1, var_2015_cast_fp16_0))[name = string("op_2018_cast_fp16")]; + tensor var_2019_cast_fp16 = mul(x = var_2018_cast_fp16, y = rope_sin_to_fp16)[name = string("op_2019_cast_fp16")]; + tensor query_states_cast_fp16 = add(x = var_2014_cast_fp16, y = var_2019_cast_fp16)[name = string("query_states_cast_fp16")]; + tensor var_2021_cast_fp16 = mul(x = x_119_cast_fp16, y = rope_cos_to_fp16)[name = string("op_2021_cast_fp16")]; + tensor var_2022_split_sizes_0 = const()[name = string("op_2022_split_sizes_0"), val = tensor([64, 64])]; + int32 var_2022_axis_0 = const()[name = string("op_2022_axis_0"), val = int32(-2)]; + tensor var_2022_cast_fp16_0, tensor var_2022_cast_fp16_1 = split(axis = var_2022_axis_0, split_sizes = var_2022_split_sizes_0, x = x_119_cast_fp16)[name = string("op_2022_cast_fp16")]; + bool var_2025_interleave_0 = const()[name = string("op_2025_interleave_0"), val = bool(false)]; + tensor var_2025_cast_fp16 = concat(axis = var_1945, interleave = var_2025_interleave_0, values = (var_2022_cast_fp16_1, var_2022_cast_fp16_0))[name = string("op_2025_cast_fp16")]; + tensor var_2026_cast_fp16 = mul(x = var_2025_cast_fp16, y = rope_sin_to_fp16)[name = string("op_2026_cast_fp16")]; + tensor key_states_cast_fp16 = add(x = var_2021_cast_fp16, y = var_2026_cast_fp16)[name = string("key_states_cast_fp16")]; + tensor var_2028_split_sizes_0 = const()[name = string("op_2028_split_sizes_0"), val = tensor([8, 8])]; + int32 var_2028_axis_0 = const()[name = string("op_2028_axis_0"), val = int32(1)]; + tensor var_2028_cast_fp16_0, tensor var_2028_cast_fp16_1 = split(axis = var_2028_axis_0, split_sizes = var_2028_split_sizes_0, x = query_states_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2030_split_sizes_0 = const()[name = string("op_2030_split_sizes_0"), val = tensor([1, 1])]; + int32 var_2030_axis_0 = const()[name = string("op_2030_axis_0"), val = int32(1)]; + tensor var_2030_cast_fp16_0, tensor var_2030_cast_fp16_1 = split(axis = var_2030_axis_0, split_sizes = var_2030_split_sizes_0, x = key_states_cast_fp16)[name = string("op_2030_cast_fp16")]; + tensor var_2032_split_sizes_0 = const()[name = string("op_2032_split_sizes_0"), val = tensor([1, 1])]; + int32 var_2032_axis_0 = const()[name = string("op_2032_axis_0"), val = int32(1)]; + tensor var_2032_cast_fp16_0, tensor var_2032_cast_fp16_1 = split(axis = var_2032_axis_0, split_sizes = var_2032_split_sizes_0, x = value_states_cast_fp16)[name = string("op_2032_cast_fp16")]; + bool attn_weights_133_transpose_x_1 = const()[name = string("attn_weights_133_transpose_x_1"), val = bool(true)]; + bool attn_weights_133_transpose_y_1 = const()[name = string("attn_weights_133_transpose_y_1"), val = bool(false)]; + tensor attn_weights_133_cast_fp16 = matmul(transpose_x = attn_weights_133_transpose_x_1, transpose_y = attn_weights_133_transpose_y_1, x = var_2030_cast_fp16_0, y = var_2028_cast_fp16_0)[name = string("attn_weights_133_cast_fp16")]; + fp16 var_2036_to_fp16 = const()[name = string("op_2036_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_135_cast_fp16 = mul(x = attn_weights_133_cast_fp16, y = var_2036_to_fp16)[name = string("attn_weights_135_cast_fp16")]; + tensor attn_weights_137_cast_fp16 = softmax(axis = var_1945, x = attn_weights_135_cast_fp16)[name = string("attn_weights_137_cast_fp16")]; + bool var_2039_transpose_x_0 = const()[name = string("op_2039_transpose_x_0"), val = bool(false)]; + bool var_2039_transpose_y_0 = const()[name = string("op_2039_transpose_y_0"), val = bool(false)]; + tensor var_2039_cast_fp16 = matmul(transpose_x = var_2039_transpose_x_0, transpose_y = var_2039_transpose_y_0, x = var_2032_cast_fp16_0, y = attn_weights_137_cast_fp16)[name = string("op_2039_cast_fp16")]; + bool attn_weights_139_transpose_x_1 = const()[name = string("attn_weights_139_transpose_x_1"), val = bool(true)]; + bool attn_weights_139_transpose_y_1 = const()[name = string("attn_weights_139_transpose_y_1"), val = bool(false)]; + tensor attn_weights_139_cast_fp16 = matmul(transpose_x = attn_weights_139_transpose_x_1, transpose_y = attn_weights_139_transpose_y_1, x = var_2030_cast_fp16_1, y = var_2028_cast_fp16_1)[name = string("attn_weights_139_cast_fp16")]; + fp16 var_2042_to_fp16 = const()[name = string("op_2042_to_fp16"), val = fp16(0x1.6ap-4)]; + tensor attn_weights_141_cast_fp16 = mul(x = attn_weights_139_cast_fp16, y = var_2042_to_fp16)[name = string("attn_weights_141_cast_fp16")]; + tensor attn_weights_cast_fp16 = softmax(axis = var_1945, x = attn_weights_141_cast_fp16)[name = string("attn_weights_cast_fp16")]; + bool attn_out_transpose_x_0 = const()[name = string("attn_out_transpose_x_0"), val = bool(false)]; + bool attn_out_transpose_y_0 = const()[name = string("attn_out_transpose_y_0"), val = bool(false)]; + tensor attn_out_cast_fp16 = matmul(transpose_x = attn_out_transpose_x_0, transpose_y = attn_out_transpose_y_0, x = var_2032_cast_fp16_1, y = attn_weights_cast_fp16)[name = string("attn_out_cast_fp16")]; + bool attn_output_interleave_0 = const()[name = string("attn_output_interleave_0"), val = bool(false)]; + tensor attn_output_cast_fp16 = concat(axis = var_1947, interleave = attn_output_interleave_0, values = (var_2039_cast_fp16, attn_out_cast_fp16))[name = string("attn_output_cast_fp16")]; + tensor var_2048 = const()[name = string("op_2048"), val = tensor([2, 2048, 1, 11])]; + tensor x_121_cast_fp16 = reshape(shape = var_2048, x = attn_output_cast_fp16)[name = string("x_121_cast_fp16")]; + string hidden_states_115_pad_type_0 = const()[name = string("hidden_states_115_pad_type_0"), val = string("valid")]; + tensor hidden_states_115_strides_0 = const()[name = string("hidden_states_115_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_115_pad_0 = const()[name = string("hidden_states_115_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_115_dilations_0 = const()[name = string("hidden_states_115_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_115_groups_0 = const()[name = string("hidden_states_115_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230406592)))]; + tensor hidden_states_115_cast_fp16 = conv(dilations = hidden_states_115_dilations_0, groups = hidden_states_115_groups_0, pad = hidden_states_115_pad_0, pad_type = hidden_states_115_pad_type_0, strides = hidden_states_115_strides_0, weight = locdit_decoder_layers_11_self_attn_o_proj_weight_to_fp16, x = x_121_cast_fp16)[name = string("hidden_states_115_cast_fp16")]; + tensor hidden_states_117_cast_fp16 = add(x = hidden_states_111_cast_fp16, y = hidden_states_115_cast_fp16)[name = string("hidden_states_117_cast_fp16")]; + fp16 const_70_promoted_to_fp16 = const()[name = string("const_70_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_2058_cast_fp16 = mul(x = hidden_states_117_cast_fp16, y = const_70_promoted_to_fp16)[name = string("op_2058_cast_fp16")]; + bool doubled_93_interleave_0 = const()[name = string("doubled_93_interleave_0"), val = bool(false)]; + tensor doubled_93_cast_fp16 = concat(axis = var_1947, interleave = doubled_93_interleave_0, values = (hidden_states_117_cast_fp16, var_2058_cast_fp16))[name = string("doubled_93_cast_fp16")]; + tensor out_47_axes_0 = const()[name = string("out_47_axes_0"), val = tensor([1])]; + tensor out_47_gamma_0_to_fp16 = const()[name = string("out_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234600960)))]; + fp16 var_2068_to_fp16 = const()[name = string("op_2068_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_2068_to_fp16, gamma = out_47_gamma_0_to_fp16, x = doubled_93_cast_fp16)[name = string("out_47_cast_fp16")]; + tensor var_2079_split_sizes_0 = const()[name = string("op_2079_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_2079_axis_0 = const()[name = string("op_2079_axis_0"), val = int32(1)]; + tensor var_2079_cast_fp16_0, tensor var_2079_cast_fp16_1 = split(axis = var_2079_axis_0, split_sizes = var_2079_split_sizes_0, x = out_47_cast_fp16)[name = string("op_2079_cast_fp16")]; + string input_pad_type_0 = const()[name = string("input_pad_type_0"), val = string("valid")]; + tensor input_strides_0 = const()[name = string("input_strides_0"), val = tensor([1, 1])]; + tensor input_pad_0 = const()[name = string("input_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_dilations_0 = const()[name = string("input_dilations_0"), val = tensor([1, 1])]; + int32 input_groups_0 = const()[name = string("input_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_mlp_gate_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_mlp_gate_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234605120)))]; + tensor input_cast_fp16 = conv(dilations = input_dilations_0, groups = input_groups_0, pad = input_pad_0, pad_type = input_pad_type_0, strides = input_strides_0, weight = locdit_decoder_layers_11_mlp_gate_proj_weight_to_fp16, x = var_2079_cast_fp16_0)[name = string("input_cast_fp16")]; + tensor var_2091_cast_fp16 = silu(x = input_cast_fp16)[name = string("op_2091_cast_fp16")]; + string var_2097_pad_type_0 = const()[name = string("op_2097_pad_type_0"), val = string("valid")]; + tensor var_2097_strides_0 = const()[name = string("op_2097_strides_0"), val = tensor([1, 1])]; + tensor var_2097_pad_0 = const()[name = string("op_2097_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2097_dilations_0 = const()[name = string("op_2097_dilations_0"), val = tensor([1, 1])]; + int32 var_2097_groups_0 = const()[name = string("op_2097_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_mlp_up_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_mlp_up_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242993792)))]; + tensor var_2097_cast_fp16 = conv(dilations = var_2097_dilations_0, groups = var_2097_groups_0, pad = var_2097_pad_0, pad_type = var_2097_pad_type_0, strides = var_2097_strides_0, weight = locdit_decoder_layers_11_mlp_up_proj_weight_to_fp16, x = var_2079_cast_fp16_0)[name = string("op_2097_cast_fp16")]; + tensor x_125_cast_fp16 = mul(x = var_2091_cast_fp16, y = var_2097_cast_fp16)[name = string("x_125_cast_fp16")]; + string hidden_states_119_pad_type_0 = const()[name = string("hidden_states_119_pad_type_0"), val = string("valid")]; + tensor hidden_states_119_strides_0 = const()[name = string("hidden_states_119_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_119_pad_0 = const()[name = string("hidden_states_119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_119_dilations_0 = const()[name = string("hidden_states_119_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_119_groups_0 = const()[name = string("hidden_states_119_groups_0"), val = int32(1)]; + tensor locdit_decoder_layers_11_mlp_down_proj_weight_to_fp16 = const()[name = string("locdit_decoder_layers_11_mlp_down_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251382464)))]; + tensor hidden_states_119_cast_fp16 = conv(dilations = hidden_states_119_dilations_0, groups = hidden_states_119_groups_0, pad = hidden_states_119_pad_0, pad_type = hidden_states_119_pad_type_0, strides = hidden_states_119_strides_0, weight = locdit_decoder_layers_11_mlp_down_proj_weight_to_fp16, x = x_125_cast_fp16)[name = string("hidden_states_119_cast_fp16")]; + tensor hidden_states_121_cast_fp16 = add(x = hidden_states_117_cast_fp16, y = hidden_states_119_cast_fp16)[name = string("hidden_states_121_cast_fp16")]; + int32 var_2112 = const()[name = string("op_2112"), val = int32(1)]; + fp16 const_72_promoted_to_fp16 = const()[name = string("const_72_promoted_to_fp16"), val = fp16(-0x1p+0)]; + tensor var_2114_cast_fp16 = mul(x = hidden_states_121_cast_fp16, y = const_72_promoted_to_fp16)[name = string("op_2114_cast_fp16")]; + bool doubled_97_interleave_0 = const()[name = string("doubled_97_interleave_0"), val = bool(false)]; + tensor doubled_97_cast_fp16 = concat(axis = var_2112, interleave = doubled_97_interleave_0, values = (hidden_states_121_cast_fp16, var_2114_cast_fp16))[name = string("doubled_97_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + tensor out_gamma_0_to_fp16 = const()[name = string("out_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259771136)))]; + fp16 var_2124_to_fp16 = const()[name = string("op_2124_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2124_to_fp16, gamma = out_gamma_0_to_fp16, x = doubled_97_cast_fp16)[name = string("out_cast_fp16")]; + tensor var_2135_split_sizes_0 = const()[name = string("op_2135_split_sizes_0"), val = tensor([1024, 1024])]; + int32 var_2135_axis_0 = const()[name = string("op_2135_axis_0"), val = int32(1)]; + tensor var_2135_cast_fp16_0, tensor var_2135_cast_fp16_1 = split(axis = var_2135_axis_0, split_sizes = var_2135_split_sizes_0, x = out_cast_fp16)[name = string("op_2135_cast_fp16")]; + tensor x_begin_0 = const()[name = string("x_begin_0"), val = tensor([0, 0, 0, 7])]; + tensor x_end_0 = const()[name = string("x_end_0"), val = tensor([2, 1024, 1, 11])]; + tensor x_end_mask_0 = const()[name = string("x_end_mask_0"), val = tensor([true, true, true, true])]; + tensor x_cast_fp16 = slice_by_index(begin = x_begin_0, end = x_end_0, end_mask = x_end_mask_0, x = var_2135_cast_fp16_0)[name = string("x_cast_fp16")]; + string var_2153_pad_type_0 = const()[name = string("op_2153_pad_type_0"), val = string("valid")]; + tensor var_2153_strides_0 = const()[name = string("op_2153_strides_0"), val = tensor([1, 1])]; + tensor var_2153_pad_0 = const()[name = string("op_2153_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2153_dilations_0 = const()[name = string("op_2153_dilations_0"), val = tensor([1, 1])]; + int32 var_2153_groups_0 = const()[name = string("op_2153_groups_0"), val = int32(1)]; + tensor locdit_out_proj_weight_to_fp16 = const()[name = string("locdit_out_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259775296)))]; + tensor locdit_out_proj_bias_to_fp16 = const()[name = string("locdit_out_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259906432)))]; + tensor var_2153_cast_fp16 = conv(bias = locdit_out_proj_bias_to_fp16, dilations = var_2153_dilations_0, groups = var_2153_groups_0, pad = var_2153_pad_0, pad_type = var_2153_pad_type_0, strides = var_2153_strides_0, weight = locdit_out_proj_weight_to_fp16, x = x_cast_fp16)[name = string("op_2153_cast_fp16")]; + string var_2153_cast_fp16_to_fp32_dtype_0 = const()[name = string("op_2153_cast_fp16_to_fp32_dtype_0"), val = string("fp32")]; + tensor output = cast(dtype = var_2153_cast_fp16_to_fp32_dtype_0, x = var_2153_cast_fp16)[name = string("cast_73")]; + } -> (output); +} \ No newline at end of file