aoiandroid/nllb200-coreml-1024-float32-ios
050
1program(1.0)2[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.8.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]3{4 func main<ios16>(tensor<int32, [1, ?]> attention_mask, tensor<int32, [1, ?]> input_ids) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"attention_mask", [1, 1]}, {"input_ids", [1, 1]}}), ("RangeDims", {{"attention_mask", [[1, 1], [1, 1024]]}, {"input_ids", [[1, 1], [1, 1024]]}})))] {5 tensor<fp32, [256206, 1024]> encoder_embed_tokens_weight = const()[name = tensor<string, []>("encoder_embed_tokens_weight"), val = tensor<fp32, [256206, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];6 tensor<fp32, [1026, 1024]> encoder_embed_positions_weights = const()[name = tensor<string, []>("encoder_embed_positions_weights"), val = tensor<fp32, [1026, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1049419904)))];7 tensor<fp32, [1024]> encoder_layers_0_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053622464)))];8 tensor<fp32, [1024]> encoder_layers_0_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_0_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053626624)))];9 tensor<fp32, [1024]> encoder_layers_0_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053630784)))];10 tensor<fp32, [1024, 1024]> encoder_layers_0_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053634944)))];11 tensor<fp32, [1024]> encoder_layers_0_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1057829312)))];12 tensor<fp32, [1024, 1024]> encoder_layers_0_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1057833472)))];13 tensor<fp32, [1024]> encoder_layers_0_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1062027840)))];14 tensor<fp32, [1024, 1024]> encoder_layers_0_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1062032000)))];15 tensor<fp32, [1024]> encoder_layers_0_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1066226368)))];16 tensor<fp32, [1024, 1024]> encoder_layers_0_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1066230528)))];17 tensor<fp32, [1024]> encoder_layers_0_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070424896)))];18 tensor<fp32, [1024]> encoder_layers_0_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_0_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070429056)))];19 tensor<fp32, [4096]> encoder_layers_0_fc1_bias = const()[name = tensor<string, []>("encoder_layers_0_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070433216)))];20 tensor<fp32, [4096, 1024]> encoder_layers_0_fc1_weight = const()[name = tensor<string, []>("encoder_layers_0_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070449664)))];21 tensor<fp32, [1024]> encoder_layers_0_fc2_bias = const()[name = tensor<string, []>("encoder_layers_0_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087226944)))];22 tensor<fp32, [1024, 4096]> encoder_layers_0_fc2_weight = const()[name = tensor<string, []>("encoder_layers_0_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1087231104)))];23 tensor<fp32, [1024]> encoder_layers_1_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1104008384)))];24 tensor<fp32, [1024]> encoder_layers_1_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_1_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1104012544)))];25 tensor<fp32, [1024]> encoder_layers_1_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1104016704)))];26 tensor<fp32, [1024, 1024]> encoder_layers_1_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1104020864)))];27 tensor<fp32, [1024]> encoder_layers_1_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1108215232)))];28 tensor<fp32, [1024, 1024]> encoder_layers_1_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1108219392)))];29 tensor<fp32, [1024]> encoder_layers_1_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112413760)))];30 tensor<fp32, [1024, 1024]> encoder_layers_1_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112417920)))];31 tensor<fp32, [1024]> encoder_layers_1_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116612288)))];32 tensor<fp32, [1024, 1024]> encoder_layers_1_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116616448)))];33 tensor<fp32, [1024]> encoder_layers_1_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120810816)))];34 tensor<fp32, [1024]> encoder_layers_1_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_1_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120814976)))];35 tensor<fp32, [4096]> encoder_layers_1_fc1_bias = const()[name = tensor<string, []>("encoder_layers_1_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120819136)))];36 tensor<fp32, [4096, 1024]> encoder_layers_1_fc1_weight = const()[name = tensor<string, []>("encoder_layers_1_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120835584)))];37 tensor<fp32, [1024]> encoder_layers_1_fc2_bias = const()[name = tensor<string, []>("encoder_layers_1_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137612864)))];38 tensor<fp32, [1024, 4096]> encoder_layers_1_fc2_weight = const()[name = tensor<string, []>("encoder_layers_1_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137617024)))];39 tensor<fp32, [1024]> encoder_layers_2_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154394304)))];40 tensor<fp32, [1024]> encoder_layers_2_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_2_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154398464)))];41 tensor<fp32, [1024]> encoder_layers_2_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154402624)))];42 tensor<fp32, [1024, 1024]> encoder_layers_2_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154406784)))];43 tensor<fp32, [1024]> encoder_layers_2_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158601152)))];44 tensor<fp32, [1024, 1024]> encoder_layers_2_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158605312)))];45 tensor<fp32, [1024]> encoder_layers_2_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162799680)))];46 tensor<fp32, [1024, 1024]> encoder_layers_2_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162803840)))];47 tensor<fp32, [1024]> encoder_layers_2_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1166998208)))];48 tensor<fp32, [1024, 1024]> encoder_layers_2_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1167002368)))];49 tensor<fp32, [1024]> encoder_layers_2_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171196736)))];50 tensor<fp32, [1024]> encoder_layers_2_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_2_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171200896)))];51 tensor<fp32, [4096]> encoder_layers_2_fc1_bias = const()[name = tensor<string, []>("encoder_layers_2_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171205056)))];52 tensor<fp32, [4096, 1024]> encoder_layers_2_fc1_weight = const()[name = tensor<string, []>("encoder_layers_2_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171221504)))];53 tensor<fp32, [1024]> encoder_layers_2_fc2_bias = const()[name = tensor<string, []>("encoder_layers_2_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1187998784)))];54 tensor<fp32, [1024, 4096]> encoder_layers_2_fc2_weight = const()[name = tensor<string, []>("encoder_layers_2_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188002944)))];55 tensor<fp32, [1024]> encoder_layers_3_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204780224)))];56 tensor<fp32, [1024]> encoder_layers_3_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_3_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204784384)))];57 tensor<fp32, [1024]> encoder_layers_3_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204788544)))];58 tensor<fp32, [1024, 1024]> encoder_layers_3_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204792704)))];59 tensor<fp32, [1024]> encoder_layers_3_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208987072)))];60 tensor<fp32, [1024, 1024]> encoder_layers_3_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208991232)))];61 tensor<fp32, [1024]> encoder_layers_3_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213185600)))];62 tensor<fp32, [1024, 1024]> encoder_layers_3_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213189760)))];63 tensor<fp32, [1024]> encoder_layers_3_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1217384128)))];64 tensor<fp32, [1024, 1024]> encoder_layers_3_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1217388288)))];65 tensor<fp32, [1024]> encoder_layers_3_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221582656)))];66 tensor<fp32, [1024]> encoder_layers_3_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_3_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221586816)))];67 tensor<fp32, [4096]> encoder_layers_3_fc1_bias = const()[name = tensor<string, []>("encoder_layers_3_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221590976)))];68 tensor<fp32, [4096, 1024]> encoder_layers_3_fc1_weight = const()[name = tensor<string, []>("encoder_layers_3_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221607424)))];69 tensor<fp32, [1024]> encoder_layers_3_fc2_bias = const()[name = tensor<string, []>("encoder_layers_3_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238384704)))];70 tensor<fp32, [1024, 4096]> encoder_layers_3_fc2_weight = const()[name = tensor<string, []>("encoder_layers_3_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238388864)))];71 tensor<fp32, [1024]> encoder_layers_4_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255166144)))];72 tensor<fp32, [1024]> encoder_layers_4_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_4_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255170304)))];73 tensor<fp32, [1024]> encoder_layers_4_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255174464)))];74 tensor<fp32, [1024, 1024]> encoder_layers_4_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255178624)))];75 tensor<fp32, [1024]> encoder_layers_4_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259372992)))];76 tensor<fp32, [1024, 1024]> encoder_layers_4_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1259377152)))];77 tensor<fp32, [1024]> encoder_layers_4_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1263571520)))];78 tensor<fp32, [1024, 1024]> encoder_layers_4_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1263575680)))];79 tensor<fp32, [1024]> encoder_layers_4_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1267770048)))];80 tensor<fp32, [1024, 1024]> encoder_layers_4_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1267774208)))];81 tensor<fp32, [1024]> encoder_layers_4_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271968576)))];82 tensor<fp32, [1024]> encoder_layers_4_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_4_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271972736)))];83 tensor<fp32, [4096]> encoder_layers_4_fc1_bias = const()[name = tensor<string, []>("encoder_layers_4_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271976896)))];84 tensor<fp32, [4096, 1024]> encoder_layers_4_fc1_weight = const()[name = tensor<string, []>("encoder_layers_4_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1271993344)))];85 tensor<fp32, [1024]> encoder_layers_4_fc2_bias = const()[name = tensor<string, []>("encoder_layers_4_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288770624)))];86 tensor<fp32, [1024, 4096]> encoder_layers_4_fc2_weight = const()[name = tensor<string, []>("encoder_layers_4_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288774784)))];87 tensor<fp32, [1024]> encoder_layers_5_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305552064)))];88 tensor<fp32, [1024]> encoder_layers_5_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_5_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305556224)))];89 tensor<fp32, [1024]> encoder_layers_5_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305560384)))];90 tensor<fp32, [1024, 1024]> encoder_layers_5_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305564544)))];91 tensor<fp32, [1024]> encoder_layers_5_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309758912)))];92 tensor<fp32, [1024, 1024]> encoder_layers_5_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309763072)))];93 tensor<fp32, [1024]> encoder_layers_5_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313957440)))];94 tensor<fp32, [1024, 1024]> encoder_layers_5_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313961600)))];95 tensor<fp32, [1024]> encoder_layers_5_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1318155968)))];96 tensor<fp32, [1024, 1024]> encoder_layers_5_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1318160128)))];97 tensor<fp32, [1024]> encoder_layers_5_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322354496)))];98 tensor<fp32, [1024]> encoder_layers_5_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_5_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322358656)))];99 tensor<fp32, [4096]> encoder_layers_5_fc1_bias = const()[name = tensor<string, []>("encoder_layers_5_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322362816)))];100 tensor<fp32, [4096, 1024]> encoder_layers_5_fc1_weight = const()[name = tensor<string, []>("encoder_layers_5_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1322379264)))];101 tensor<fp32, [1024]> encoder_layers_5_fc2_bias = const()[name = tensor<string, []>("encoder_layers_5_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339156544)))];102 tensor<fp32, [1024, 4096]> encoder_layers_5_fc2_weight = const()[name = tensor<string, []>("encoder_layers_5_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1339160704)))];103 tensor<fp32, [1024]> encoder_layers_6_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355937984)))];104 tensor<fp32, [1024]> encoder_layers_6_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_6_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355942144)))];105 tensor<fp32, [1024]> encoder_layers_6_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355946304)))];106 tensor<fp32, [1024, 1024]> encoder_layers_6_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355950464)))];107 tensor<fp32, [1024]> encoder_layers_6_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360144832)))];108 tensor<fp32, [1024, 1024]> encoder_layers_6_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360148992)))];109 tensor<fp32, [1024]> encoder_layers_6_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364343360)))];110 tensor<fp32, [1024, 1024]> encoder_layers_6_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364347520)))];111 tensor<fp32, [1024]> encoder_layers_6_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1368541888)))];112 tensor<fp32, [1024, 1024]> encoder_layers_6_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1368546048)))];113 tensor<fp32, [1024]> encoder_layers_6_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372740416)))];114 tensor<fp32, [1024]> encoder_layers_6_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_6_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372744576)))];115 tensor<fp32, [4096]> encoder_layers_6_fc1_bias = const()[name = tensor<string, []>("encoder_layers_6_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372748736)))];116 tensor<fp32, [4096, 1024]> encoder_layers_6_fc1_weight = const()[name = tensor<string, []>("encoder_layers_6_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372765184)))];117 tensor<fp32, [1024]> encoder_layers_6_fc2_bias = const()[name = tensor<string, []>("encoder_layers_6_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389542464)))];118 tensor<fp32, [1024, 4096]> encoder_layers_6_fc2_weight = const()[name = tensor<string, []>("encoder_layers_6_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389546624)))];119 tensor<fp32, [1024]> encoder_layers_7_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406323904)))];120 tensor<fp32, [1024]> encoder_layers_7_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_7_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406328064)))];121 tensor<fp32, [1024]> encoder_layers_7_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406332224)))];122 tensor<fp32, [1024, 1024]> encoder_layers_7_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406336384)))];123 tensor<fp32, [1024]> encoder_layers_7_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410530752)))];124 tensor<fp32, [1024, 1024]> encoder_layers_7_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410534912)))];125 tensor<fp32, [1024]> encoder_layers_7_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414729280)))];126 tensor<fp32, [1024, 1024]> encoder_layers_7_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414733440)))];127 tensor<fp32, [1024]> encoder_layers_7_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1418927808)))];128 tensor<fp32, [1024, 1024]> encoder_layers_7_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1418931968)))];129 tensor<fp32, [1024]> encoder_layers_7_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423126336)))];130 tensor<fp32, [1024]> encoder_layers_7_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_7_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423130496)))];131 tensor<fp32, [4096]> encoder_layers_7_fc1_bias = const()[name = tensor<string, []>("encoder_layers_7_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423134656)))];132 tensor<fp32, [4096, 1024]> encoder_layers_7_fc1_weight = const()[name = tensor<string, []>("encoder_layers_7_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423151104)))];133 tensor<fp32, [1024]> encoder_layers_7_fc2_bias = const()[name = tensor<string, []>("encoder_layers_7_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1439928384)))];134 tensor<fp32, [1024, 4096]> encoder_layers_7_fc2_weight = const()[name = tensor<string, []>("encoder_layers_7_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1439932544)))];135 tensor<fp32, [1024]> encoder_layers_8_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1456709824)))];136 tensor<fp32, [1024]> encoder_layers_8_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_8_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1456713984)))];137 tensor<fp32, [1024]> encoder_layers_8_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1456718144)))];138 tensor<fp32, [1024, 1024]> encoder_layers_8_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1456722304)))];139 tensor<fp32, [1024]> encoder_layers_8_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1460916672)))];140 tensor<fp32, [1024, 1024]> encoder_layers_8_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1460920832)))];141 tensor<fp32, [1024]> encoder_layers_8_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465115200)))];142 tensor<fp32, [1024, 1024]> encoder_layers_8_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465119360)))];143 tensor<fp32, [1024]> encoder_layers_8_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1469313728)))];144 tensor<fp32, [1024, 1024]> encoder_layers_8_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1469317888)))];145 tensor<fp32, [1024]> encoder_layers_8_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473512256)))];146 tensor<fp32, [1024]> encoder_layers_8_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_8_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473516416)))];147 tensor<fp32, [4096]> encoder_layers_8_fc1_bias = const()[name = tensor<string, []>("encoder_layers_8_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473520576)))];148 tensor<fp32, [4096, 1024]> encoder_layers_8_fc1_weight = const()[name = tensor<string, []>("encoder_layers_8_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473537024)))];149 tensor<fp32, [1024]> encoder_layers_8_fc2_bias = const()[name = tensor<string, []>("encoder_layers_8_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490314304)))];150 tensor<fp32, [1024, 4096]> encoder_layers_8_fc2_weight = const()[name = tensor<string, []>("encoder_layers_8_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490318464)))];151 tensor<fp32, [1024]> encoder_layers_9_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507095744)))];152 tensor<fp32, [1024]> encoder_layers_9_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_9_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507099904)))];153 tensor<fp32, [1024]> encoder_layers_9_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507104064)))];154 tensor<fp32, [1024, 1024]> encoder_layers_9_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507108224)))];155 tensor<fp32, [1024]> encoder_layers_9_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1511302592)))];156 tensor<fp32, [1024, 1024]> encoder_layers_9_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1511306752)))];157 tensor<fp32, [1024]> encoder_layers_9_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515501120)))];158 tensor<fp32, [1024, 1024]> encoder_layers_9_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1515505280)))];159 tensor<fp32, [1024]> encoder_layers_9_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519699648)))];160 tensor<fp32, [1024, 1024]> encoder_layers_9_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1519703808)))];161 tensor<fp32, [1024]> encoder_layers_9_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523898176)))];162 tensor<fp32, [1024]> encoder_layers_9_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_9_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523902336)))];163 tensor<fp32, [4096]> encoder_layers_9_fc1_bias = const()[name = tensor<string, []>("encoder_layers_9_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523906496)))];164 tensor<fp32, [4096, 1024]> encoder_layers_9_fc1_weight = const()[name = tensor<string, []>("encoder_layers_9_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523922944)))];165 tensor<fp32, [1024]> encoder_layers_9_fc2_bias = const()[name = tensor<string, []>("encoder_layers_9_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540700224)))];166 tensor<fp32, [1024, 4096]> encoder_layers_9_fc2_weight = const()[name = tensor<string, []>("encoder_layers_9_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540704384)))];167 tensor<fp32, [1024]> encoder_layers_10_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1557481664)))];168 tensor<fp32, [1024]> encoder_layers_10_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_10_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1557485824)))];169 tensor<fp32, [1024]> encoder_layers_10_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1557489984)))];170 tensor<fp32, [1024, 1024]> encoder_layers_10_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1557494144)))];171 tensor<fp32, [1024]> encoder_layers_10_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1561688512)))];172 tensor<fp32, [1024, 1024]> encoder_layers_10_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1561692672)))];173 tensor<fp32, [1024]> encoder_layers_10_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565887040)))];174 tensor<fp32, [1024, 1024]> encoder_layers_10_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565891200)))];175 tensor<fp32, [1024]> encoder_layers_10_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570085568)))];176 tensor<fp32, [1024, 1024]> encoder_layers_10_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1570089728)))];177 tensor<fp32, [1024]> encoder_layers_10_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574284096)))];178 tensor<fp32, [1024]> encoder_layers_10_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_10_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574288256)))];179 tensor<fp32, [4096]> encoder_layers_10_fc1_bias = const()[name = tensor<string, []>("encoder_layers_10_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574292416)))];180 tensor<fp32, [4096, 1024]> encoder_layers_10_fc1_weight = const()[name = tensor<string, []>("encoder_layers_10_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1574308864)))];181 tensor<fp32, [1024]> encoder_layers_10_fc2_bias = const()[name = tensor<string, []>("encoder_layers_10_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1591086144)))];182 tensor<fp32, [1024, 4096]> encoder_layers_10_fc2_weight = const()[name = tensor<string, []>("encoder_layers_10_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1591090304)))];183 tensor<fp32, [1024]> encoder_layers_11_self_attn_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607867584)))];184 tensor<fp32, [1024]> encoder_layers_11_self_attn_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_11_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607871744)))];185 tensor<fp32, [1024]> encoder_layers_11_self_attn_q_proj_bias = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607875904)))];186 tensor<fp32, [1024, 1024]> encoder_layers_11_self_attn_q_proj_weight = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607880064)))];187 tensor<fp32, [1024]> encoder_layers_11_self_attn_k_proj_bias = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612074432)))];188 tensor<fp32, [1024, 1024]> encoder_layers_11_self_attn_k_proj_weight = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612078592)))];189 tensor<fp32, [1024]> encoder_layers_11_self_attn_v_proj_bias = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616272960)))];190 tensor<fp32, [1024, 1024]> encoder_layers_11_self_attn_v_proj_weight = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616277120)))];191 tensor<fp32, [1024]> encoder_layers_11_self_attn_out_proj_bias = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1620471488)))];192 tensor<fp32, [1024, 1024]> encoder_layers_11_self_attn_out_proj_weight = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1620475648)))];193 tensor<fp32, [1024]> encoder_layers_11_final_layer_norm_bias = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624670016)))];194 tensor<fp32, [1024]> encoder_layers_11_final_layer_norm_weight = const()[name = tensor<string, []>("encoder_layers_11_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624674176)))];195 tensor<fp32, [4096]> encoder_layers_11_fc1_bias = const()[name = tensor<string, []>("encoder_layers_11_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624678336)))];196 tensor<fp32, [4096, 1024]> encoder_layers_11_fc1_weight = const()[name = tensor<string, []>("encoder_layers_11_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624694784)))];197 tensor<fp32, [1024]> encoder_layers_11_fc2_bias = const()[name = tensor<string, []>("encoder_layers_11_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641472064)))];198 tensor<fp32, [1024, 4096]> encoder_layers_11_fc2_weight = const()[name = tensor<string, []>("encoder_layers_11_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641476224)))];199 tensor<fp32, [1024]> encoder_layer_norm_bias = const()[name = tensor<string, []>("encoder_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658253504)))];200 tensor<fp32, [1024]> encoder_layer_norm_weight = const()[name = tensor<string, []>("encoder_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658257664)))];201 tensor<fp32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<fp32, []>(0x1.4f8b58p-17)];202 tensor<fp32, []> var_7 = const()[name = tensor<string, []>("op_7"), val = tensor<fp32, []>(0x1p-3)];203 tensor<fp32, []> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<fp32, []>(-0x1.fffffep+127)];204 tensor<int32, []> var_17 = const()[name = tensor<string, []>("op_17"), val = tensor<int32, []>(0)];205 tensor<int32, []> var_23 = const()[name = tensor<string, []>("op_23"), val = tensor<int32, []>(-1)];206 tensor<int32, []> var_24 = const()[name = tensor<string, []>("op_24"), val = tensor<int32, []>(1)];207 tensor<int32, []> var_56_axis_0 = const()[name = tensor<string, []>("op_56_axis_0"), val = tensor<int32, []>(0)];208 tensor<int32, []> var_56_batch_dims_0 = const()[name = tensor<string, []>("op_56_batch_dims_0"), val = tensor<int32, []>(0)];209 tensor<fp32, [1, ?, 1024]> var_56 = gather(axis = var_56_axis_0, batch_dims = var_56_batch_dims_0, indices = input_ids, x = encoder_embed_tokens_weight)[name = tensor<string, []>("op_56")];210 tensor<fp32, []> var_57 = const()[name = tensor<string, []>("op_57"), val = tensor<fp32, []>(0x1p+5)];211 tensor<fp32, [1, ?, 1024]> inputs_embeds = mul(x = var_56, y = var_57)[name = tensor<string, []>("inputs_embeds")];212 tensor<bool, [1, ?]> var_62 = not_equal(x = input_ids, y = var_24)[name = tensor<string, []>("op_62")];213 tensor<string, []> mask_dtype_0 = const()[name = tensor<string, []>("mask_dtype_0"), val = tensor<string, []>("int32")];214 tensor<bool, []> var_64_exclusive_0 = const()[name = tensor<string, []>("op_64_exclusive_0"), val = tensor<bool, []>(false)];215 tensor<bool, []> var_64_reverse_0 = const()[name = tensor<string, []>("op_64_reverse_0"), val = tensor<bool, []>(false)];216 tensor<int32, [1, ?]> mask = cast(dtype = mask_dtype_0, x = var_62)[name = tensor<string, []>("cast_55")];217 tensor<int32, [1, ?]> var_64 = cumsum(axis = var_24, exclusive = var_64_exclusive_0, reverse = var_64_reverse_0, x = mask)[name = tensor<string, []>("op_64")];218 tensor<int32, [1, ?]> incremental_indices = mul(x = var_64, y = mask)[name = tensor<string, []>("incremental_indices")];219 tensor<int32, []> var_70 = const()[name = tensor<string, []>("op_70"), val = tensor<int32, []>(1)];220 tensor<int32, [1, ?]> var_71 = add(x = incremental_indices, y = var_70)[name = tensor<string, []>("op_71")];221 tensor<int32, [1]> var_73 = const()[name = tensor<string, []>("op_73"), val = tensor<int32, [1]>([-1])];222 tensor<int32, [?]> var_74 = reshape(shape = var_73, x = var_71)[name = tensor<string, []>("op_74")];223 tensor<int32, []> var_75_batch_dims_0 = const()[name = tensor<string, []>("op_75_batch_dims_0"), val = tensor<int32, []>(0)];224 tensor<fp32, [?, 1024]> var_75 = gather(axis = var_17, batch_dims = var_75_batch_dims_0, indices = var_74, x = encoder_embed_positions_weights)[name = tensor<string, []>("op_75")];225 tensor<int32, [3]> concat_1x = const()[name = tensor<string, []>("concat_1x"), val = tensor<int32, [3]>([1, -1, 1024])];226 tensor<fp32, [1, ?, 1024]> var_78 = reshape(shape = concat_1x, x = var_75)[name = tensor<string, []>("op_78")];227 tensor<fp32, [1, ?, 1024]> input_3 = add(x = inputs_embeds, y = var_78)[name = tensor<string, []>("input_3")];228 tensor<int32, [2]> var_83_shape = shape(x = attention_mask)[name = tensor<string, []>("op_83_shape")];229 tensor<int32, []> gather_3_indices_0 = const()[name = tensor<string, []>("gather_3_indices_0"), val = tensor<int32, []>(1)];230 tensor<int32, []> gather_3_axis_0 = const()[name = tensor<string, []>("gather_3_axis_0"), val = tensor<int32, []>(0)];231 tensor<int32, []> gather_3_batch_dims_0 = const()[name = tensor<string, []>("gather_3_batch_dims_0"), val = tensor<int32, []>(0)];232 tensor<int32, []> gather_3 = gather(axis = gather_3_axis_0, batch_dims = gather_3_batch_dims_0, indices = gather_3_indices_0, x = var_83_shape)[name = tensor<string, []>("gather_3")];233 tensor<int32, []> gather_4 = const()[name = tensor<string, []>("gather_4"), val = tensor<int32, []>(1)];234 tensor<int32, [1]> var_87_axes_0 = const()[name = tensor<string, []>("op_87_axes_0"), val = tensor<int32, [1]>([1])];235 tensor<int32, [1, 1, ?]> var_87 = expand_dims(axes = var_87_axes_0, x = attention_mask)[name = tensor<string, []>("op_87")];236 tensor<int32, [1]> var_88_axes_0 = const()[name = tensor<string, []>("op_88_axes_0"), val = tensor<int32, [1]>([2])];237 tensor<int32, [1, 1, 1, ?]> var_88 = expand_dims(axes = var_88_axes_0, x = var_87)[name = tensor<string, []>("op_88")];238 tensor<int32, []> concat_2_axis_0 = const()[name = tensor<string, []>("concat_2_axis_0"), val = tensor<int32, []>(0)];239 tensor<bool, []> concat_2_interleave_0 = const()[name = tensor<string, []>("concat_2_interleave_0"), val = tensor<bool, []>(false)];240 tensor<int32, [4]> concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (gather_4, var_24, gather_3, gather_3))[name = tensor<string, []>("concat_2")];241 tensor<int32, [4]> shape_0 = shape(x = var_88)[name = tensor<string, []>("shape_0")];242 tensor<int32, []> equal_0_y_0 = const()[name = tensor<string, []>("equal_0_y_0"), val = tensor<int32, []>(-1)];243 tensor<bool, [4]> equal_0 = equal(x = concat_2, y = equal_0_y_0)[name = tensor<string, []>("equal_0")];244 tensor<int32, [4]> select_0 = select(a = shape_0, b = concat_2, cond = equal_0)[name = tensor<string, []>("select_0")];245 tensor<int32, [4]> real_div_0 = real_div(x = select_0, y = shape_0)[name = tensor<string, []>("real_div_0")];246 tensor<int32, [?, ?, ?, ?]> var_91 = tile(reps = real_div_0, x = var_88)[name = tensor<string, []>("op_91")];247 tensor<string, []> expanded_mask_dtype_0 = const()[name = tensor<string, []>("expanded_mask_dtype_0"), val = tensor<string, []>("fp32")];248 tensor<fp32, []> const_1 = const()[name = tensor<string, []>("const_1"), val = tensor<fp32, []>(0x1p+0)];249 tensor<fp32, [?, ?, ?, ?]> expanded_mask = cast(dtype = expanded_mask_dtype_0, x = var_91)[name = tensor<string, []>("cast_54")];250 tensor<fp32, [?, ?, ?, ?]> inverted_mask = sub(x = const_1, y = expanded_mask)[name = tensor<string, []>("inverted_mask")];251 tensor<string, []> var_96_dtype_0 = const()[name = tensor<string, []>("op_96_dtype_0"), val = tensor<string, []>("bool")];252 tensor<bool, [?, ?, ?, ?]> var_96 = cast(dtype = var_96_dtype_0, x = inverted_mask)[name = tensor<string, []>("cast_53")];253 tensor<fp32, [?, ?, ?, ?]> attention_mask_3 = select(a = var_10, b = inverted_mask, cond = var_96)[name = tensor<string, []>("attention_mask_3")];254 tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])];255 tensor<fp32, [1, ?, 1024]> hidden_states_1 = layer_norm(axes = hidden_states_1_axes_0, beta = encoder_layers_0_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_0_self_attn_layer_norm_weight, x = input_3)[name = tensor<string, []>("hidden_states_1")];256 tensor<int32, [3]> var_111_shape = shape(x = hidden_states_1)[name = tensor<string, []>("op_111_shape")];257 tensor<int32, []> gather_6 = const()[name = tensor<string, []>("gather_6"), val = tensor<int32, []>(1)];258 tensor<int32, []> gather_7_indices_0 = const()[name = tensor<string, []>("gather_7_indices_0"), val = tensor<int32, []>(1)];259 tensor<int32, []> gather_7_axis_0 = const()[name = tensor<string, []>("gather_7_axis_0"), val = tensor<int32, []>(0)];260 tensor<int32, []> gather_7_batch_dims_0 = const()[name = tensor<string, []>("gather_7_batch_dims_0"), val = tensor<int32, []>(0)];261 tensor<int32, []> gather_7 = gather(axis = gather_7_axis_0, batch_dims = gather_7_batch_dims_0, indices = gather_7_indices_0, x = var_111_shape)[name = tensor<string, []>("gather_7")];262 tensor<fp32, [1, ?, 1024]> var_115 = linear(bias = encoder_layers_0_self_attn_q_proj_bias, weight = encoder_layers_0_self_attn_q_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_0")];263 tensor<int32, [4]> concat_3x = const()[name = tensor<string, []>("concat_3x"), val = tensor<int32, [4]>([1, -1, 16, 64])];264 tensor<fp32, [1, ?, 16, 64]> var_117 = reshape(shape = concat_3x, x = var_115)[name = tensor<string, []>("op_117")];265 tensor<int32, [4]> query_1_perm_0 = const()[name = tensor<string, []>("query_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];266 tensor<fp32, [1, ?, 1024]> key_states_1 = linear(bias = encoder_layers_0_self_attn_k_proj_bias, weight = encoder_layers_0_self_attn_k_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_1")];267 tensor<fp32, [1, ?, 1024]> value_states_1 = linear(bias = encoder_layers_0_self_attn_v_proj_bias, weight = encoder_layers_0_self_attn_v_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_2")];268 tensor<int32, [4]> concat_4x = const()[name = tensor<string, []>("concat_4x"), val = tensor<int32, [4]>([1, -1, 16, 64])];269 tensor<fp32, [1, ?, 16, 64]> var_126 = reshape(shape = concat_4x, x = key_states_1)[name = tensor<string, []>("op_126")];270 tensor<int32, [4]> key_1_perm_0 = const()[name = tensor<string, []>("key_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];271 tensor<int32, [4]> concat_5x = const()[name = tensor<string, []>("concat_5x"), val = tensor<int32, [4]>([1, -1, 16, 64])];272 tensor<fp32, [1, ?, 16, 64]> var_129 = reshape(shape = concat_5x, x = value_states_1)[name = tensor<string, []>("op_129")];273 tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];274 tensor<fp32, [1, 16, ?, 64]> key_1 = transpose(perm = key_1_perm_0, x = var_126)[name = tensor<string, []>("transpose_46")];275 tensor<int32, [4]> var_131_shape = shape(x = key_1)[name = tensor<string, []>("op_131_shape")];276 tensor<int32, []> gather_8_indices_0 = const()[name = tensor<string, []>("gather_8_indices_0"), val = tensor<int32, []>(2)];277 tensor<int32, []> gather_8_axis_0 = const()[name = tensor<string, []>("gather_8_axis_0"), val = tensor<int32, []>(0)];278 tensor<int32, []> gather_8_batch_dims_0 = const()[name = tensor<string, []>("gather_8_batch_dims_0"), val = tensor<int32, []>(0)];279 tensor<int32, []> gather_8 = gather(axis = gather_8_axis_0, batch_dims = gather_8_batch_dims_0, indices = gather_8_indices_0, x = var_131_shape)[name = tensor<string, []>("gather_8")];280 tensor<int32, []> concat_6_values0_0 = const()[name = tensor<string, []>("concat_6_values0_0"), val = tensor<int32, []>(0)];281 tensor<int32, []> concat_6_values1_0 = const()[name = tensor<string, []>("concat_6_values1_0"), val = tensor<int32, []>(0)];282 tensor<int32, []> concat_6_values2_0 = const()[name = tensor<string, []>("concat_6_values2_0"), val = tensor<int32, []>(0)];283 tensor<int32, []> concat_6_axis_0 = const()[name = tensor<string, []>("concat_6_axis_0"), val = tensor<int32, []>(0)];284 tensor<bool, []> concat_6_interleave_0 = const()[name = tensor<string, []>("concat_6_interleave_0"), val = tensor<bool, []>(false)];285 tensor<int32, [4]> concat_6 = concat(axis = concat_6_axis_0, interleave = concat_6_interleave_0, values = (concat_6_values0_0, concat_6_values1_0, concat_6_values2_0, gather_8))[name = tensor<string, []>("concat_6")];286 tensor<int32, [4]> attention_mask_5_begin_0 = const()[name = tensor<string, []>("attention_mask_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];287 tensor<bool, [4]> attention_mask_5_end_mask_0 = const()[name = tensor<string, []>("attention_mask_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];288 tensor<fp32, [?, ?, ?, ?]> attention_mask_5 = slice_by_index(begin = attention_mask_5_begin_0, end = concat_6, end_mask = attention_mask_5_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_5")];289 tensor<fp32, [1, 16, ?, 64]> query_1 = transpose(perm = query_1_perm_0, x = var_117)[name = tensor<string, []>("transpose_47")];290 tensor<fp32, [1, 16, ?, 64]> mul_0 = mul(x = query_1, y = var_7)[name = tensor<string, []>("mul_0")];291 tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)];292 tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)];293 tensor<fp32, [1, 16, ?, ?]> matmul_0 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = mul_0, y = key_1)[name = tensor<string, []>("matmul_0")];294 tensor<fp32, [?, 16, ?, ?]> add_0 = add(x = matmul_0, y = attention_mask_5)[name = tensor<string, []>("add_0")];295 tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)];296 tensor<fp32, [?, 16, ?, ?]> softmax_0 = softmax(axis = softmax_0_axis_0, x = add_0)[name = tensor<string, []>("softmax_0")];297 tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];298 tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];299 tensor<fp32, [1, 16, ?, 64]> value_1 = transpose(perm = value_1_perm_0, x = var_129)[name = tensor<string, []>("transpose_45")];300 tensor<fp32, [?, 16, ?, 64]> attn_output_1 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0, y = value_1)[name = tensor<string, []>("attn_output_1")];301 tensor<int32, [4]> var_137_perm_0 = const()[name = tensor<string, []>("op_137_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];302 tensor<int32, []> concat_7_axis_0 = const()[name = tensor<string, []>("concat_7_axis_0"), val = tensor<int32, []>(0)];303 tensor<bool, []> concat_7_interleave_0 = const()[name = tensor<string, []>("concat_7_interleave_0"), val = tensor<bool, []>(false)];304 tensor<int32, [3]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (gather_6, gather_7, var_23))[name = tensor<string, []>("concat_7")];305 tensor<fp32, [?, ?, 16, 64]> var_137 = transpose(perm = var_137_perm_0, x = attn_output_1)[name = tensor<string, []>("transpose_44")];306 tensor<fp32, [1, ?, ?]> var_140 = reshape(shape = concat_7, x = var_137)[name = tensor<string, []>("op_140")];307 tensor<fp32, [1, ?, 1024]> input_9 = linear(bias = encoder_layers_0_self_attn_out_proj_bias, weight = encoder_layers_0_self_attn_out_proj_weight, x = var_140)[name = tensor<string, []>("linear_3")];308 tensor<fp32, [1, ?, 1024]> input_11 = add(x = input_3, y = input_9)[name = tensor<string, []>("input_11")];309 tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([-1])];310 tensor<fp32, [1, ?, 1024]> input_13 = layer_norm(axes = input_13_axes_0, beta = encoder_layers_0_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_0_final_layer_norm_weight, x = input_11)[name = tensor<string, []>("input_13")];311 tensor<fp32, [1, ?, 4096]> input_15 = linear(bias = encoder_layers_0_fc1_bias, weight = encoder_layers_0_fc1_weight, x = input_13)[name = tensor<string, []>("linear_4")];312 tensor<fp32, [1, ?, 4096]> input_17 = relu(x = input_15)[name = tensor<string, []>("input_17")];313 tensor<fp32, [1, ?, 1024]> input_21 = linear(bias = encoder_layers_0_fc2_bias, weight = encoder_layers_0_fc2_weight, x = input_17)[name = tensor<string, []>("linear_5")];314 tensor<fp32, [1, ?, 1024]> input_23 = add(x = input_11, y = input_21)[name = tensor<string, []>("input_23")];315 tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])];316 tensor<fp32, [1, ?, 1024]> hidden_states_7 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_layers_1_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_1_self_attn_layer_norm_weight, x = input_23)[name = tensor<string, []>("hidden_states_7")];317 tensor<int32, [3]> var_174_shape = shape(x = hidden_states_7)[name = tensor<string, []>("op_174_shape")];318 tensor<int32, []> gather_9 = const()[name = tensor<string, []>("gather_9"), val = tensor<int32, []>(1)];319 tensor<int32, []> gather_10_indices_0 = const()[name = tensor<string, []>("gather_10_indices_0"), val = tensor<int32, []>(1)];320 tensor<int32, []> gather_10_axis_0 = const()[name = tensor<string, []>("gather_10_axis_0"), val = tensor<int32, []>(0)];321 tensor<int32, []> gather_10_batch_dims_0 = const()[name = tensor<string, []>("gather_10_batch_dims_0"), val = tensor<int32, []>(0)];322 tensor<int32, []> gather_10 = gather(axis = gather_10_axis_0, batch_dims = gather_10_batch_dims_0, indices = gather_10_indices_0, x = var_174_shape)[name = tensor<string, []>("gather_10")];323 tensor<fp32, [1, ?, 1024]> var_178 = linear(bias = encoder_layers_1_self_attn_q_proj_bias, weight = encoder_layers_1_self_attn_q_proj_weight, x = hidden_states_7)[name = tensor<string, []>("linear_6")];324 tensor<int32, [4]> concat_8x = const()[name = tensor<string, []>("concat_8x"), val = tensor<int32, [4]>([1, -1, 16, 64])];325 tensor<fp32, [1, ?, 16, 64]> var_180 = reshape(shape = concat_8x, x = var_178)[name = tensor<string, []>("op_180")];326 tensor<int32, [4]> query_3_perm_0 = const()[name = tensor<string, []>("query_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];327 tensor<fp32, [1, ?, 1024]> key_states_3 = linear(bias = encoder_layers_1_self_attn_k_proj_bias, weight = encoder_layers_1_self_attn_k_proj_weight, x = hidden_states_7)[name = tensor<string, []>("linear_7")];328 tensor<fp32, [1, ?, 1024]> value_states_3 = linear(bias = encoder_layers_1_self_attn_v_proj_bias, weight = encoder_layers_1_self_attn_v_proj_weight, x = hidden_states_7)[name = tensor<string, []>("linear_8")];329 tensor<int32, [4]> concat_9x = const()[name = tensor<string, []>("concat_9x"), val = tensor<int32, [4]>([1, -1, 16, 64])];330 tensor<fp32, [1, ?, 16, 64]> var_189 = reshape(shape = concat_9x, x = key_states_3)[name = tensor<string, []>("op_189")];331 tensor<int32, [4]> key_3_perm_0 = const()[name = tensor<string, []>("key_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];332 tensor<int32, [4]> concat_10x = const()[name = tensor<string, []>("concat_10x"), val = tensor<int32, [4]>([1, -1, 16, 64])];333 tensor<fp32, [1, ?, 16, 64]> var_192 = reshape(shape = concat_10x, x = value_states_3)[name = tensor<string, []>("op_192")];334 tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];335 tensor<fp32, [1, 16, ?, 64]> key_3 = transpose(perm = key_3_perm_0, x = var_189)[name = tensor<string, []>("transpose_42")];336 tensor<int32, [4]> var_194_shape = shape(x = key_3)[name = tensor<string, []>("op_194_shape")];337 tensor<int32, []> gather_11_indices_0 = const()[name = tensor<string, []>("gather_11_indices_0"), val = tensor<int32, []>(2)];338 tensor<int32, []> gather_11_axis_0 = const()[name = tensor<string, []>("gather_11_axis_0"), val = tensor<int32, []>(0)];339 tensor<int32, []> gather_11_batch_dims_0 = const()[name = tensor<string, []>("gather_11_batch_dims_0"), val = tensor<int32, []>(0)];340 tensor<int32, []> gather_11 = gather(axis = gather_11_axis_0, batch_dims = gather_11_batch_dims_0, indices = gather_11_indices_0, x = var_194_shape)[name = tensor<string, []>("gather_11")];341 tensor<int32, []> concat_11_values0_0 = const()[name = tensor<string, []>("concat_11_values0_0"), val = tensor<int32, []>(0)];342 tensor<int32, []> concat_11_values1_0 = const()[name = tensor<string, []>("concat_11_values1_0"), val = tensor<int32, []>(0)];343 tensor<int32, []> concat_11_values2_0 = const()[name = tensor<string, []>("concat_11_values2_0"), val = tensor<int32, []>(0)];344 tensor<int32, []> concat_11_axis_0 = const()[name = tensor<string, []>("concat_11_axis_0"), val = tensor<int32, []>(0)];345 tensor<bool, []> concat_11_interleave_0 = const()[name = tensor<string, []>("concat_11_interleave_0"), val = tensor<bool, []>(false)];346 tensor<int32, [4]> concat_11 = concat(axis = concat_11_axis_0, interleave = concat_11_interleave_0, values = (concat_11_values0_0, concat_11_values1_0, concat_11_values2_0, gather_11))[name = tensor<string, []>("concat_11")];347 tensor<int32, [4]> attention_mask_7_begin_0 = const()[name = tensor<string, []>("attention_mask_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];348 tensor<bool, [4]> attention_mask_7_end_mask_0 = const()[name = tensor<string, []>("attention_mask_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];349 tensor<fp32, [?, ?, ?, ?]> attention_mask_7 = slice_by_index(begin = attention_mask_7_begin_0, end = concat_11, end_mask = attention_mask_7_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_7")];350 tensor<fp32, [1, 16, ?, 64]> query_3 = transpose(perm = query_3_perm_0, x = var_180)[name = tensor<string, []>("transpose_43")];351 tensor<fp32, [1, 16, ?, 64]> mul_1 = mul(x = query_3, y = var_7)[name = tensor<string, []>("mul_1")];352 tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)];353 tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)];354 tensor<fp32, [1, 16, ?, ?]> matmul_1 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = mul_1, y = key_3)[name = tensor<string, []>("matmul_1")];355 tensor<fp32, [?, 16, ?, ?]> add_1 = add(x = matmul_1, y = attention_mask_7)[name = tensor<string, []>("add_1")];356 tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)];357 tensor<fp32, [?, 16, ?, ?]> softmax_1 = softmax(axis = softmax_1_axis_0, x = add_1)[name = tensor<string, []>("softmax_1")];358 tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];359 tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];360 tensor<fp32, [1, 16, ?, 64]> value_3 = transpose(perm = value_3_perm_0, x = var_192)[name = tensor<string, []>("transpose_41")];361 tensor<fp32, [?, 16, ?, 64]> attn_output_5 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1, y = value_3)[name = tensor<string, []>("attn_output_5")];362 tensor<int32, [4]> var_200_perm_0 = const()[name = tensor<string, []>("op_200_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];363 tensor<int32, []> concat_12_axis_0 = const()[name = tensor<string, []>("concat_12_axis_0"), val = tensor<int32, []>(0)];364 tensor<bool, []> concat_12_interleave_0 = const()[name = tensor<string, []>("concat_12_interleave_0"), val = tensor<bool, []>(false)];365 tensor<int32, [3]> concat_12 = concat(axis = concat_12_axis_0, interleave = concat_12_interleave_0, values = (gather_9, gather_10, var_23))[name = tensor<string, []>("concat_12")];366 tensor<fp32, [?, ?, 16, 64]> var_200 = transpose(perm = var_200_perm_0, x = attn_output_5)[name = tensor<string, []>("transpose_40")];367 tensor<fp32, [1, ?, ?]> var_203 = reshape(shape = concat_12, x = var_200)[name = tensor<string, []>("op_203")];368 tensor<fp32, [1, ?, 1024]> input_27 = linear(bias = encoder_layers_1_self_attn_out_proj_bias, weight = encoder_layers_1_self_attn_out_proj_weight, x = var_203)[name = tensor<string, []>("linear_9")];369 tensor<fp32, [1, ?, 1024]> input_29 = add(x = input_23, y = input_27)[name = tensor<string, []>("input_29")];370 tensor<int32, [1]> input_31_axes_0 = const()[name = tensor<string, []>("input_31_axes_0"), val = tensor<int32, [1]>([-1])];371 tensor<fp32, [1, ?, 1024]> input_31 = layer_norm(axes = input_31_axes_0, beta = encoder_layers_1_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_1_final_layer_norm_weight, x = input_29)[name = tensor<string, []>("input_31")];372 tensor<fp32, [1, ?, 4096]> input_33 = linear(bias = encoder_layers_1_fc1_bias, weight = encoder_layers_1_fc1_weight, x = input_31)[name = tensor<string, []>("linear_10")];373 tensor<fp32, [1, ?, 4096]> input_35 = relu(x = input_33)[name = tensor<string, []>("input_35")];374 tensor<fp32, [1, ?, 1024]> input_39 = linear(bias = encoder_layers_1_fc2_bias, weight = encoder_layers_1_fc2_weight, x = input_35)[name = tensor<string, []>("linear_11")];375 tensor<fp32, [1, ?, 1024]> input_41 = add(x = input_29, y = input_39)[name = tensor<string, []>("input_41")];376 tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])];377 tensor<fp32, [1, ?, 1024]> hidden_states_13 = layer_norm(axes = hidden_states_13_axes_0, beta = encoder_layers_2_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_2_self_attn_layer_norm_weight, x = input_41)[name = tensor<string, []>("hidden_states_13")];378 tensor<int32, [3]> var_237_shape = shape(x = hidden_states_13)[name = tensor<string, []>("op_237_shape")];379 tensor<int32, []> gather_12 = const()[name = tensor<string, []>("gather_12"), val = tensor<int32, []>(1)];380 tensor<int32, []> gather_13_indices_0 = const()[name = tensor<string, []>("gather_13_indices_0"), val = tensor<int32, []>(1)];381 tensor<int32, []> gather_13_axis_0 = const()[name = tensor<string, []>("gather_13_axis_0"), val = tensor<int32, []>(0)];382 tensor<int32, []> gather_13_batch_dims_0 = const()[name = tensor<string, []>("gather_13_batch_dims_0"), val = tensor<int32, []>(0)];383 tensor<int32, []> gather_13 = gather(axis = gather_13_axis_0, batch_dims = gather_13_batch_dims_0, indices = gather_13_indices_0, x = var_237_shape)[name = tensor<string, []>("gather_13")];384 tensor<fp32, [1, ?, 1024]> var_241 = linear(bias = encoder_layers_2_self_attn_q_proj_bias, weight = encoder_layers_2_self_attn_q_proj_weight, x = hidden_states_13)[name = tensor<string, []>("linear_12")];385 tensor<int32, [4]> concat_13x = const()[name = tensor<string, []>("concat_13x"), val = tensor<int32, [4]>([1, -1, 16, 64])];386 tensor<fp32, [1, ?, 16, 64]> var_243 = reshape(shape = concat_13x, x = var_241)[name = tensor<string, []>("op_243")];387 tensor<int32, [4]> query_5_perm_0 = const()[name = tensor<string, []>("query_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];388 tensor<fp32, [1, ?, 1024]> key_states_5 = linear(bias = encoder_layers_2_self_attn_k_proj_bias, weight = encoder_layers_2_self_attn_k_proj_weight, x = hidden_states_13)[name = tensor<string, []>("linear_13")];389 tensor<fp32, [1, ?, 1024]> value_states_5 = linear(bias = encoder_layers_2_self_attn_v_proj_bias, weight = encoder_layers_2_self_attn_v_proj_weight, x = hidden_states_13)[name = tensor<string, []>("linear_14")];390 tensor<int32, [4]> concat_14x = const()[name = tensor<string, []>("concat_14x"), val = tensor<int32, [4]>([1, -1, 16, 64])];391 tensor<fp32, [1, ?, 16, 64]> var_252 = reshape(shape = concat_14x, x = key_states_5)[name = tensor<string, []>("op_252")];392 tensor<int32, [4]> key_5_perm_0 = const()[name = tensor<string, []>("key_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];393 tensor<int32, [4]> concat_15x = const()[name = tensor<string, []>("concat_15x"), val = tensor<int32, [4]>([1, -1, 16, 64])];394 tensor<fp32, [1, ?, 16, 64]> var_255 = reshape(shape = concat_15x, x = value_states_5)[name = tensor<string, []>("op_255")];395 tensor<int32, [4]> value_5_perm_0 = const()[name = tensor<string, []>("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];396 tensor<fp32, [1, 16, ?, 64]> key_5 = transpose(perm = key_5_perm_0, x = var_252)[name = tensor<string, []>("transpose_38")];397 tensor<int32, [4]> var_257_shape = shape(x = key_5)[name = tensor<string, []>("op_257_shape")];398 tensor<int32, []> gather_14_indices_0 = const()[name = tensor<string, []>("gather_14_indices_0"), val = tensor<int32, []>(2)];399 tensor<int32, []> gather_14_axis_0 = const()[name = tensor<string, []>("gather_14_axis_0"), val = tensor<int32, []>(0)];400 tensor<int32, []> gather_14_batch_dims_0 = const()[name = tensor<string, []>("gather_14_batch_dims_0"), val = tensor<int32, []>(0)];401 tensor<int32, []> gather_14 = gather(axis = gather_14_axis_0, batch_dims = gather_14_batch_dims_0, indices = gather_14_indices_0, x = var_257_shape)[name = tensor<string, []>("gather_14")];402 tensor<int32, []> concat_16_values0_0 = const()[name = tensor<string, []>("concat_16_values0_0"), val = tensor<int32, []>(0)];403 tensor<int32, []> concat_16_values1_0 = const()[name = tensor<string, []>("concat_16_values1_0"), val = tensor<int32, []>(0)];404 tensor<int32, []> concat_16_values2_0 = const()[name = tensor<string, []>("concat_16_values2_0"), val = tensor<int32, []>(0)];405 tensor<int32, []> concat_16_axis_0 = const()[name = tensor<string, []>("concat_16_axis_0"), val = tensor<int32, []>(0)];406 tensor<bool, []> concat_16_interleave_0 = const()[name = tensor<string, []>("concat_16_interleave_0"), val = tensor<bool, []>(false)];407 tensor<int32, [4]> concat_16 = concat(axis = concat_16_axis_0, interleave = concat_16_interleave_0, values = (concat_16_values0_0, concat_16_values1_0, concat_16_values2_0, gather_14))[name = tensor<string, []>("concat_16")];408 tensor<int32, [4]> attention_mask_9_begin_0 = const()[name = tensor<string, []>("attention_mask_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];409 tensor<bool, [4]> attention_mask_9_end_mask_0 = const()[name = tensor<string, []>("attention_mask_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];410 tensor<fp32, [?, ?, ?, ?]> attention_mask_9 = slice_by_index(begin = attention_mask_9_begin_0, end = concat_16, end_mask = attention_mask_9_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_9")];411 tensor<fp32, [1, 16, ?, 64]> query_5 = transpose(perm = query_5_perm_0, x = var_243)[name = tensor<string, []>("transpose_39")];412 tensor<fp32, [1, 16, ?, 64]> mul_2 = mul(x = query_5, y = var_7)[name = tensor<string, []>("mul_2")];413 tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)];414 tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)];415 tensor<fp32, [1, 16, ?, ?]> matmul_2 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = mul_2, y = key_5)[name = tensor<string, []>("matmul_2")];416 tensor<fp32, [?, 16, ?, ?]> add_2 = add(x = matmul_2, y = attention_mask_9)[name = tensor<string, []>("add_2")];417 tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)];418 tensor<fp32, [?, 16, ?, ?]> softmax_2 = softmax(axis = softmax_2_axis_0, x = add_2)[name = tensor<string, []>("softmax_2")];419 tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];420 tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];421 tensor<fp32, [1, 16, ?, 64]> value_5 = transpose(perm = value_5_perm_0, x = var_255)[name = tensor<string, []>("transpose_37")];422 tensor<fp32, [?, 16, ?, 64]> attn_output_9 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2, y = value_5)[name = tensor<string, []>("attn_output_9")];423 tensor<int32, [4]> var_263_perm_0 = const()[name = tensor<string, []>("op_263_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];424 tensor<int32, []> concat_17_axis_0 = const()[name = tensor<string, []>("concat_17_axis_0"), val = tensor<int32, []>(0)];425 tensor<bool, []> concat_17_interleave_0 = const()[name = tensor<string, []>("concat_17_interleave_0"), val = tensor<bool, []>(false)];426 tensor<int32, [3]> concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (gather_12, gather_13, var_23))[name = tensor<string, []>("concat_17")];427 tensor<fp32, [?, ?, 16, 64]> var_263 = transpose(perm = var_263_perm_0, x = attn_output_9)[name = tensor<string, []>("transpose_36")];428 tensor<fp32, [1, ?, ?]> var_266 = reshape(shape = concat_17, x = var_263)[name = tensor<string, []>("op_266")];429 tensor<fp32, [1, ?, 1024]> input_45 = linear(bias = encoder_layers_2_self_attn_out_proj_bias, weight = encoder_layers_2_self_attn_out_proj_weight, x = var_266)[name = tensor<string, []>("linear_15")];430 tensor<fp32, [1, ?, 1024]> input_47 = add(x = input_41, y = input_45)[name = tensor<string, []>("input_47")];431 tensor<int32, [1]> input_49_axes_0 = const()[name = tensor<string, []>("input_49_axes_0"), val = tensor<int32, [1]>([-1])];432 tensor<fp32, [1, ?, 1024]> input_49 = layer_norm(axes = input_49_axes_0, beta = encoder_layers_2_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_2_final_layer_norm_weight, x = input_47)[name = tensor<string, []>("input_49")];433 tensor<fp32, [1, ?, 4096]> input_51 = linear(bias = encoder_layers_2_fc1_bias, weight = encoder_layers_2_fc1_weight, x = input_49)[name = tensor<string, []>("linear_16")];434 tensor<fp32, [1, ?, 4096]> input_53 = relu(x = input_51)[name = tensor<string, []>("input_53")];435 tensor<fp32, [1, ?, 1024]> input_57 = linear(bias = encoder_layers_2_fc2_bias, weight = encoder_layers_2_fc2_weight, x = input_53)[name = tensor<string, []>("linear_17")];436 tensor<fp32, [1, ?, 1024]> input_59 = add(x = input_47, y = input_57)[name = tensor<string, []>("input_59")];437 tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])];438 tensor<fp32, [1, ?, 1024]> hidden_states_19 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_layers_3_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_3_self_attn_layer_norm_weight, x = input_59)[name = tensor<string, []>("hidden_states_19")];439 tensor<int32, [3]> var_300_shape = shape(x = hidden_states_19)[name = tensor<string, []>("op_300_shape")];440 tensor<int32, []> gather_15 = const()[name = tensor<string, []>("gather_15"), val = tensor<int32, []>(1)];441 tensor<int32, []> gather_16_indices_0 = const()[name = tensor<string, []>("gather_16_indices_0"), val = tensor<int32, []>(1)];442 tensor<int32, []> gather_16_axis_0 = const()[name = tensor<string, []>("gather_16_axis_0"), val = tensor<int32, []>(0)];443 tensor<int32, []> gather_16_batch_dims_0 = const()[name = tensor<string, []>("gather_16_batch_dims_0"), val = tensor<int32, []>(0)];444 tensor<int32, []> gather_16 = gather(axis = gather_16_axis_0, batch_dims = gather_16_batch_dims_0, indices = gather_16_indices_0, x = var_300_shape)[name = tensor<string, []>("gather_16")];445 tensor<fp32, [1, ?, 1024]> var_304 = linear(bias = encoder_layers_3_self_attn_q_proj_bias, weight = encoder_layers_3_self_attn_q_proj_weight, x = hidden_states_19)[name = tensor<string, []>("linear_18")];446 tensor<int32, [4]> concat_18x = const()[name = tensor<string, []>("concat_18x"), val = tensor<int32, [4]>([1, -1, 16, 64])];447 tensor<fp32, [1, ?, 16, 64]> var_306 = reshape(shape = concat_18x, x = var_304)[name = tensor<string, []>("op_306")];448 tensor<int32, [4]> query_7_perm_0 = const()[name = tensor<string, []>("query_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];449 tensor<fp32, [1, ?, 1024]> key_states_7 = linear(bias = encoder_layers_3_self_attn_k_proj_bias, weight = encoder_layers_3_self_attn_k_proj_weight, x = hidden_states_19)[name = tensor<string, []>("linear_19")];450 tensor<fp32, [1, ?, 1024]> value_states_7 = linear(bias = encoder_layers_3_self_attn_v_proj_bias, weight = encoder_layers_3_self_attn_v_proj_weight, x = hidden_states_19)[name = tensor<string, []>("linear_20")];451 tensor<int32, [4]> concat_19x = const()[name = tensor<string, []>("concat_19x"), val = tensor<int32, [4]>([1, -1, 16, 64])];452 tensor<fp32, [1, ?, 16, 64]> var_315 = reshape(shape = concat_19x, x = key_states_7)[name = tensor<string, []>("op_315")];453 tensor<int32, [4]> key_7_perm_0 = const()[name = tensor<string, []>("key_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];454 tensor<int32, [4]> concat_20x = const()[name = tensor<string, []>("concat_20x"), val = tensor<int32, [4]>([1, -1, 16, 64])];455 tensor<fp32, [1, ?, 16, 64]> var_318 = reshape(shape = concat_20x, x = value_states_7)[name = tensor<string, []>("op_318")];456 tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];457 tensor<fp32, [1, 16, ?, 64]> key_7 = transpose(perm = key_7_perm_0, x = var_315)[name = tensor<string, []>("transpose_34")];458 tensor<int32, [4]> var_320_shape = shape(x = key_7)[name = tensor<string, []>("op_320_shape")];459 tensor<int32, []> gather_17_indices_0 = const()[name = tensor<string, []>("gather_17_indices_0"), val = tensor<int32, []>(2)];460 tensor<int32, []> gather_17_axis_0 = const()[name = tensor<string, []>("gather_17_axis_0"), val = tensor<int32, []>(0)];461 tensor<int32, []> gather_17_batch_dims_0 = const()[name = tensor<string, []>("gather_17_batch_dims_0"), val = tensor<int32, []>(0)];462 tensor<int32, []> gather_17 = gather(axis = gather_17_axis_0, batch_dims = gather_17_batch_dims_0, indices = gather_17_indices_0, x = var_320_shape)[name = tensor<string, []>("gather_17")];463 tensor<int32, []> concat_21_values0_0 = const()[name = tensor<string, []>("concat_21_values0_0"), val = tensor<int32, []>(0)];464 tensor<int32, []> concat_21_values1_0 = const()[name = tensor<string, []>("concat_21_values1_0"), val = tensor<int32, []>(0)];465 tensor<int32, []> concat_21_values2_0 = const()[name = tensor<string, []>("concat_21_values2_0"), val = tensor<int32, []>(0)];466 tensor<int32, []> concat_21_axis_0 = const()[name = tensor<string, []>("concat_21_axis_0"), val = tensor<int32, []>(0)];467 tensor<bool, []> concat_21_interleave_0 = const()[name = tensor<string, []>("concat_21_interleave_0"), val = tensor<bool, []>(false)];468 tensor<int32, [4]> concat_21 = concat(axis = concat_21_axis_0, interleave = concat_21_interleave_0, values = (concat_21_values0_0, concat_21_values1_0, concat_21_values2_0, gather_17))[name = tensor<string, []>("concat_21")];469 tensor<int32, [4]> attention_mask_11_begin_0 = const()[name = tensor<string, []>("attention_mask_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];470 tensor<bool, [4]> attention_mask_11_end_mask_0 = const()[name = tensor<string, []>("attention_mask_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];471 tensor<fp32, [?, ?, ?, ?]> attention_mask_11 = slice_by_index(begin = attention_mask_11_begin_0, end = concat_21, end_mask = attention_mask_11_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_11")];472 tensor<fp32, [1, 16, ?, 64]> query_7 = transpose(perm = query_7_perm_0, x = var_306)[name = tensor<string, []>("transpose_35")];473 tensor<fp32, [1, 16, ?, 64]> mul_3 = mul(x = query_7, y = var_7)[name = tensor<string, []>("mul_3")];474 tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)];475 tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)];476 tensor<fp32, [1, 16, ?, ?]> matmul_3 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = mul_3, y = key_7)[name = tensor<string, []>("matmul_3")];477 tensor<fp32, [?, 16, ?, ?]> add_3 = add(x = matmul_3, y = attention_mask_11)[name = tensor<string, []>("add_3")];478 tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)];479 tensor<fp32, [?, 16, ?, ?]> softmax_3 = softmax(axis = softmax_3_axis_0, x = add_3)[name = tensor<string, []>("softmax_3")];480 tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];481 tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];482 tensor<fp32, [1, 16, ?, 64]> value_7 = transpose(perm = value_7_perm_0, x = var_318)[name = tensor<string, []>("transpose_33")];483 tensor<fp32, [?, 16, ?, 64]> attn_output_13 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3, y = value_7)[name = tensor<string, []>("attn_output_13")];484 tensor<int32, [4]> var_326_perm_0 = const()[name = tensor<string, []>("op_326_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];485 tensor<int32, []> concat_22_axis_0 = const()[name = tensor<string, []>("concat_22_axis_0"), val = tensor<int32, []>(0)];486 tensor<bool, []> concat_22_interleave_0 = const()[name = tensor<string, []>("concat_22_interleave_0"), val = tensor<bool, []>(false)];487 tensor<int32, [3]> concat_22 = concat(axis = concat_22_axis_0, interleave = concat_22_interleave_0, values = (gather_15, gather_16, var_23))[name = tensor<string, []>("concat_22")];488 tensor<fp32, [?, ?, 16, 64]> var_326 = transpose(perm = var_326_perm_0, x = attn_output_13)[name = tensor<string, []>("transpose_32")];489 tensor<fp32, [1, ?, ?]> var_329 = reshape(shape = concat_22, x = var_326)[name = tensor<string, []>("op_329")];490 tensor<fp32, [1, ?, 1024]> input_63 = linear(bias = encoder_layers_3_self_attn_out_proj_bias, weight = encoder_layers_3_self_attn_out_proj_weight, x = var_329)[name = tensor<string, []>("linear_21")];491 tensor<fp32, [1, ?, 1024]> input_65 = add(x = input_59, y = input_63)[name = tensor<string, []>("input_65")];492 tensor<int32, [1]> input_67_axes_0 = const()[name = tensor<string, []>("input_67_axes_0"), val = tensor<int32, [1]>([-1])];493 tensor<fp32, [1, ?, 1024]> input_67 = layer_norm(axes = input_67_axes_0, beta = encoder_layers_3_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_3_final_layer_norm_weight, x = input_65)[name = tensor<string, []>("input_67")];494 tensor<fp32, [1, ?, 4096]> input_69 = linear(bias = encoder_layers_3_fc1_bias, weight = encoder_layers_3_fc1_weight, x = input_67)[name = tensor<string, []>("linear_22")];495 tensor<fp32, [1, ?, 4096]> input_71 = relu(x = input_69)[name = tensor<string, []>("input_71")];496 tensor<fp32, [1, ?, 1024]> input_75 = linear(bias = encoder_layers_3_fc2_bias, weight = encoder_layers_3_fc2_weight, x = input_71)[name = tensor<string, []>("linear_23")];497 tensor<fp32, [1, ?, 1024]> input_77 = add(x = input_65, y = input_75)[name = tensor<string, []>("input_77")];498 tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])];499 tensor<fp32, [1, ?, 1024]> hidden_states_25 = layer_norm(axes = hidden_states_25_axes_0, beta = encoder_layers_4_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_4_self_attn_layer_norm_weight, x = input_77)[name = tensor<string, []>("hidden_states_25")];500 tensor<int32, [3]> var_363_shape = shape(x = hidden_states_25)[name = tensor<string, []>("op_363_shape")];501 tensor<int32, []> gather_18 = const()[name = tensor<string, []>("gather_18"), val = tensor<int32, []>(1)];502 tensor<int32, []> gather_19_indices_0 = const()[name = tensor<string, []>("gather_19_indices_0"), val = tensor<int32, []>(1)];503 tensor<int32, []> gather_19_axis_0 = const()[name = tensor<string, []>("gather_19_axis_0"), val = tensor<int32, []>(0)];504 tensor<int32, []> gather_19_batch_dims_0 = const()[name = tensor<string, []>("gather_19_batch_dims_0"), val = tensor<int32, []>(0)];505 tensor<int32, []> gather_19 = gather(axis = gather_19_axis_0, batch_dims = gather_19_batch_dims_0, indices = gather_19_indices_0, x = var_363_shape)[name = tensor<string, []>("gather_19")];506 tensor<fp32, [1, ?, 1024]> var_367 = linear(bias = encoder_layers_4_self_attn_q_proj_bias, weight = encoder_layers_4_self_attn_q_proj_weight, x = hidden_states_25)[name = tensor<string, []>("linear_24")];507 tensor<int32, [4]> concat_23x = const()[name = tensor<string, []>("concat_23x"), val = tensor<int32, [4]>([1, -1, 16, 64])];508 tensor<fp32, [1, ?, 16, 64]> var_369 = reshape(shape = concat_23x, x = var_367)[name = tensor<string, []>("op_369")];509 tensor<int32, [4]> query_9_perm_0 = const()[name = tensor<string, []>("query_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];510 tensor<fp32, [1, ?, 1024]> key_states_9 = linear(bias = encoder_layers_4_self_attn_k_proj_bias, weight = encoder_layers_4_self_attn_k_proj_weight, x = hidden_states_25)[name = tensor<string, []>("linear_25")];511 tensor<fp32, [1, ?, 1024]> value_states_9 = linear(bias = encoder_layers_4_self_attn_v_proj_bias, weight = encoder_layers_4_self_attn_v_proj_weight, x = hidden_states_25)[name = tensor<string, []>("linear_26")];512 tensor<int32, [4]> concat_24x = const()[name = tensor<string, []>("concat_24x"), val = tensor<int32, [4]>([1, -1, 16, 64])];513 tensor<fp32, [1, ?, 16, 64]> var_378 = reshape(shape = concat_24x, x = key_states_9)[name = tensor<string, []>("op_378")];514 tensor<int32, [4]> key_9_perm_0 = const()[name = tensor<string, []>("key_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];515 tensor<int32, [4]> concat_25x = const()[name = tensor<string, []>("concat_25x"), val = tensor<int32, [4]>([1, -1, 16, 64])];516 tensor<fp32, [1, ?, 16, 64]> var_381 = reshape(shape = concat_25x, x = value_states_9)[name = tensor<string, []>("op_381")];517 tensor<int32, [4]> value_9_perm_0 = const()[name = tensor<string, []>("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];518 tensor<fp32, [1, 16, ?, 64]> key_9 = transpose(perm = key_9_perm_0, x = var_378)[name = tensor<string, []>("transpose_30")];519 tensor<int32, [4]> var_383_shape = shape(x = key_9)[name = tensor<string, []>("op_383_shape")];520 tensor<int32, []> gather_20_indices_0 = const()[name = tensor<string, []>("gather_20_indices_0"), val = tensor<int32, []>(2)];521 tensor<int32, []> gather_20_axis_0 = const()[name = tensor<string, []>("gather_20_axis_0"), val = tensor<int32, []>(0)];522 tensor<int32, []> gather_20_batch_dims_0 = const()[name = tensor<string, []>("gather_20_batch_dims_0"), val = tensor<int32, []>(0)];523 tensor<int32, []> gather_20 = gather(axis = gather_20_axis_0, batch_dims = gather_20_batch_dims_0, indices = gather_20_indices_0, x = var_383_shape)[name = tensor<string, []>("gather_20")];524 tensor<int32, []> concat_26_values0_0 = const()[name = tensor<string, []>("concat_26_values0_0"), val = tensor<int32, []>(0)];525 tensor<int32, []> concat_26_values1_0 = const()[name = tensor<string, []>("concat_26_values1_0"), val = tensor<int32, []>(0)];526 tensor<int32, []> concat_26_values2_0 = const()[name = tensor<string, []>("concat_26_values2_0"), val = tensor<int32, []>(0)];527 tensor<int32, []> concat_26_axis_0 = const()[name = tensor<string, []>("concat_26_axis_0"), val = tensor<int32, []>(0)];528 tensor<bool, []> concat_26_interleave_0 = const()[name = tensor<string, []>("concat_26_interleave_0"), val = tensor<bool, []>(false)];529 tensor<int32, [4]> concat_26 = concat(axis = concat_26_axis_0, interleave = concat_26_interleave_0, values = (concat_26_values0_0, concat_26_values1_0, concat_26_values2_0, gather_20))[name = tensor<string, []>("concat_26")];530 tensor<int32, [4]> attention_mask_13_begin_0 = const()[name = tensor<string, []>("attention_mask_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];531 tensor<bool, [4]> attention_mask_13_end_mask_0 = const()[name = tensor<string, []>("attention_mask_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];532 tensor<fp32, [?, ?, ?, ?]> attention_mask_13 = slice_by_index(begin = attention_mask_13_begin_0, end = concat_26, end_mask = attention_mask_13_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_13")];533 tensor<fp32, [1, 16, ?, 64]> query_9 = transpose(perm = query_9_perm_0, x = var_369)[name = tensor<string, []>("transpose_31")];534 tensor<fp32, [1, 16, ?, 64]> mul_4 = mul(x = query_9, y = var_7)[name = tensor<string, []>("mul_4")];535 tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)];536 tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)];537 tensor<fp32, [1, 16, ?, ?]> matmul_4 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = mul_4, y = key_9)[name = tensor<string, []>("matmul_4")];538 tensor<fp32, [?, 16, ?, ?]> add_4 = add(x = matmul_4, y = attention_mask_13)[name = tensor<string, []>("add_4")];539 tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)];540 tensor<fp32, [?, 16, ?, ?]> softmax_4 = softmax(axis = softmax_4_axis_0, x = add_4)[name = tensor<string, []>("softmax_4")];541 tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];542 tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];543 tensor<fp32, [1, 16, ?, 64]> value_9 = transpose(perm = value_9_perm_0, x = var_381)[name = tensor<string, []>("transpose_29")];544 tensor<fp32, [?, 16, ?, 64]> attn_output_17 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4, y = value_9)[name = tensor<string, []>("attn_output_17")];545 tensor<int32, [4]> var_389_perm_0 = const()[name = tensor<string, []>("op_389_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];546 tensor<int32, []> concat_27_axis_0 = const()[name = tensor<string, []>("concat_27_axis_0"), val = tensor<int32, []>(0)];547 tensor<bool, []> concat_27_interleave_0 = const()[name = tensor<string, []>("concat_27_interleave_0"), val = tensor<bool, []>(false)];548 tensor<int32, [3]> concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (gather_18, gather_19, var_23))[name = tensor<string, []>("concat_27")];549 tensor<fp32, [?, ?, 16, 64]> var_389 = transpose(perm = var_389_perm_0, x = attn_output_17)[name = tensor<string, []>("transpose_28")];550 tensor<fp32, [1, ?, ?]> var_392 = reshape(shape = concat_27, x = var_389)[name = tensor<string, []>("op_392")];551 tensor<fp32, [1, ?, 1024]> input_81 = linear(bias = encoder_layers_4_self_attn_out_proj_bias, weight = encoder_layers_4_self_attn_out_proj_weight, x = var_392)[name = tensor<string, []>("linear_27")];552 tensor<fp32, [1, ?, 1024]> input_83 = add(x = input_77, y = input_81)[name = tensor<string, []>("input_83")];553 tensor<int32, [1]> input_85_axes_0 = const()[name = tensor<string, []>("input_85_axes_0"), val = tensor<int32, [1]>([-1])];554 tensor<fp32, [1, ?, 1024]> input_85 = layer_norm(axes = input_85_axes_0, beta = encoder_layers_4_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_4_final_layer_norm_weight, x = input_83)[name = tensor<string, []>("input_85")];555 tensor<fp32, [1, ?, 4096]> input_87 = linear(bias = encoder_layers_4_fc1_bias, weight = encoder_layers_4_fc1_weight, x = input_85)[name = tensor<string, []>("linear_28")];556 tensor<fp32, [1, ?, 4096]> input_89 = relu(x = input_87)[name = tensor<string, []>("input_89")];557 tensor<fp32, [1, ?, 1024]> input_93 = linear(bias = encoder_layers_4_fc2_bias, weight = encoder_layers_4_fc2_weight, x = input_89)[name = tensor<string, []>("linear_29")];558 tensor<fp32, [1, ?, 1024]> input_95 = add(x = input_83, y = input_93)[name = tensor<string, []>("input_95")];559 tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])];560 tensor<fp32, [1, ?, 1024]> hidden_states_31 = layer_norm(axes = hidden_states_31_axes_0, beta = encoder_layers_5_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_5_self_attn_layer_norm_weight, x = input_95)[name = tensor<string, []>("hidden_states_31")];561 tensor<int32, [3]> var_426_shape = shape(x = hidden_states_31)[name = tensor<string, []>("op_426_shape")];562 tensor<int32, []> gather_21 = const()[name = tensor<string, []>("gather_21"), val = tensor<int32, []>(1)];563 tensor<int32, []> gather_22_indices_0 = const()[name = tensor<string, []>("gather_22_indices_0"), val = tensor<int32, []>(1)];564 tensor<int32, []> gather_22_axis_0 = const()[name = tensor<string, []>("gather_22_axis_0"), val = tensor<int32, []>(0)];565 tensor<int32, []> gather_22_batch_dims_0 = const()[name = tensor<string, []>("gather_22_batch_dims_0"), val = tensor<int32, []>(0)];566 tensor<int32, []> gather_22 = gather(axis = gather_22_axis_0, batch_dims = gather_22_batch_dims_0, indices = gather_22_indices_0, x = var_426_shape)[name = tensor<string, []>("gather_22")];567 tensor<fp32, [1, ?, 1024]> var_430 = linear(bias = encoder_layers_5_self_attn_q_proj_bias, weight = encoder_layers_5_self_attn_q_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_30")];568 tensor<int32, [4]> concat_28x = const()[name = tensor<string, []>("concat_28x"), val = tensor<int32, [4]>([1, -1, 16, 64])];569 tensor<fp32, [1, ?, 16, 64]> var_432 = reshape(shape = concat_28x, x = var_430)[name = tensor<string, []>("op_432")];570 tensor<int32, [4]> query_11_perm_0 = const()[name = tensor<string, []>("query_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];571 tensor<fp32, [1, ?, 1024]> key_states_11 = linear(bias = encoder_layers_5_self_attn_k_proj_bias, weight = encoder_layers_5_self_attn_k_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_31")];572 tensor<fp32, [1, ?, 1024]> value_states_11 = linear(bias = encoder_layers_5_self_attn_v_proj_bias, weight = encoder_layers_5_self_attn_v_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_32")];573 tensor<int32, [4]> concat_29x = const()[name = tensor<string, []>("concat_29x"), val = tensor<int32, [4]>([1, -1, 16, 64])];574 tensor<fp32, [1, ?, 16, 64]> var_441 = reshape(shape = concat_29x, x = key_states_11)[name = tensor<string, []>("op_441")];575 tensor<int32, [4]> key_11_perm_0 = const()[name = tensor<string, []>("key_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];576 tensor<int32, [4]> concat_30x = const()[name = tensor<string, []>("concat_30x"), val = tensor<int32, [4]>([1, -1, 16, 64])];577 tensor<fp32, [1, ?, 16, 64]> var_444 = reshape(shape = concat_30x, x = value_states_11)[name = tensor<string, []>("op_444")];578 tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];579 tensor<fp32, [1, 16, ?, 64]> key_11 = transpose(perm = key_11_perm_0, x = var_441)[name = tensor<string, []>("transpose_26")];580 tensor<int32, [4]> var_446_shape = shape(x = key_11)[name = tensor<string, []>("op_446_shape")];581 tensor<int32, []> gather_23_indices_0 = const()[name = tensor<string, []>("gather_23_indices_0"), val = tensor<int32, []>(2)];582 tensor<int32, []> gather_23_axis_0 = const()[name = tensor<string, []>("gather_23_axis_0"), val = tensor<int32, []>(0)];583 tensor<int32, []> gather_23_batch_dims_0 = const()[name = tensor<string, []>("gather_23_batch_dims_0"), val = tensor<int32, []>(0)];584 tensor<int32, []> gather_23 = gather(axis = gather_23_axis_0, batch_dims = gather_23_batch_dims_0, indices = gather_23_indices_0, x = var_446_shape)[name = tensor<string, []>("gather_23")];585 tensor<int32, []> concat_31_values0_0 = const()[name = tensor<string, []>("concat_31_values0_0"), val = tensor<int32, []>(0)];586 tensor<int32, []> concat_31_values1_0 = const()[name = tensor<string, []>("concat_31_values1_0"), val = tensor<int32, []>(0)];587 tensor<int32, []> concat_31_values2_0 = const()[name = tensor<string, []>("concat_31_values2_0"), val = tensor<int32, []>(0)];588 tensor<int32, []> concat_31_axis_0 = const()[name = tensor<string, []>("concat_31_axis_0"), val = tensor<int32, []>(0)];589 tensor<bool, []> concat_31_interleave_0 = const()[name = tensor<string, []>("concat_31_interleave_0"), val = tensor<bool, []>(false)];590 tensor<int32, [4]> concat_31 = concat(axis = concat_31_axis_0, interleave = concat_31_interleave_0, values = (concat_31_values0_0, concat_31_values1_0, concat_31_values2_0, gather_23))[name = tensor<string, []>("concat_31")];591 tensor<int32, [4]> attention_mask_15_begin_0 = const()[name = tensor<string, []>("attention_mask_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];592 tensor<bool, [4]> attention_mask_15_end_mask_0 = const()[name = tensor<string, []>("attention_mask_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];593 tensor<fp32, [?, ?, ?, ?]> attention_mask_15 = slice_by_index(begin = attention_mask_15_begin_0, end = concat_31, end_mask = attention_mask_15_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_15")];594 tensor<fp32, [1, 16, ?, 64]> query_11 = transpose(perm = query_11_perm_0, x = var_432)[name = tensor<string, []>("transpose_27")];595 tensor<fp32, [1, 16, ?, 64]> mul_5 = mul(x = query_11, y = var_7)[name = tensor<string, []>("mul_5")];596 tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)];597 tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)];598 tensor<fp32, [1, 16, ?, ?]> matmul_5 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = mul_5, y = key_11)[name = tensor<string, []>("matmul_5")];599 tensor<fp32, [?, 16, ?, ?]> add_5 = add(x = matmul_5, y = attention_mask_15)[name = tensor<string, []>("add_5")];600 tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)];601 tensor<fp32, [?, 16, ?, ?]> softmax_5 = softmax(axis = softmax_5_axis_0, x = add_5)[name = tensor<string, []>("softmax_5")];602 tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];603 tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];604 tensor<fp32, [1, 16, ?, 64]> value_11 = transpose(perm = value_11_perm_0, x = var_444)[name = tensor<string, []>("transpose_25")];605 tensor<fp32, [?, 16, ?, 64]> attn_output_21 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5, y = value_11)[name = tensor<string, []>("attn_output_21")];606 tensor<int32, [4]> var_452_perm_0 = const()[name = tensor<string, []>("op_452_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];607 tensor<int32, []> concat_32_axis_0 = const()[name = tensor<string, []>("concat_32_axis_0"), val = tensor<int32, []>(0)];608 tensor<bool, []> concat_32_interleave_0 = const()[name = tensor<string, []>("concat_32_interleave_0"), val = tensor<bool, []>(false)];609 tensor<int32, [3]> concat_32 = concat(axis = concat_32_axis_0, interleave = concat_32_interleave_0, values = (gather_21, gather_22, var_23))[name = tensor<string, []>("concat_32")];610 tensor<fp32, [?, ?, 16, 64]> var_452 = transpose(perm = var_452_perm_0, x = attn_output_21)[name = tensor<string, []>("transpose_24")];611 tensor<fp32, [1, ?, ?]> var_455 = reshape(shape = concat_32, x = var_452)[name = tensor<string, []>("op_455")];612 tensor<fp32, [1, ?, 1024]> input_99 = linear(bias = encoder_layers_5_self_attn_out_proj_bias, weight = encoder_layers_5_self_attn_out_proj_weight, x = var_455)[name = tensor<string, []>("linear_33")];613 tensor<fp32, [1, ?, 1024]> input_101 = add(x = input_95, y = input_99)[name = tensor<string, []>("input_101")];614 tensor<int32, [1]> input_103_axes_0 = const()[name = tensor<string, []>("input_103_axes_0"), val = tensor<int32, [1]>([-1])];615 tensor<fp32, [1, ?, 1024]> input_103 = layer_norm(axes = input_103_axes_0, beta = encoder_layers_5_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_5_final_layer_norm_weight, x = input_101)[name = tensor<string, []>("input_103")];616 tensor<fp32, [1, ?, 4096]> input_105 = linear(bias = encoder_layers_5_fc1_bias, weight = encoder_layers_5_fc1_weight, x = input_103)[name = tensor<string, []>("linear_34")];617 tensor<fp32, [1, ?, 4096]> input_107 = relu(x = input_105)[name = tensor<string, []>("input_107")];618 tensor<fp32, [1, ?, 1024]> input_111 = linear(bias = encoder_layers_5_fc2_bias, weight = encoder_layers_5_fc2_weight, x = input_107)[name = tensor<string, []>("linear_35")];619 tensor<fp32, [1, ?, 1024]> input_113 = add(x = input_101, y = input_111)[name = tensor<string, []>("input_113")];620 tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])];621 tensor<fp32, [1, ?, 1024]> hidden_states_37 = layer_norm(axes = hidden_states_37_axes_0, beta = encoder_layers_6_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_6_self_attn_layer_norm_weight, x = input_113)[name = tensor<string, []>("hidden_states_37")];622 tensor<int32, [3]> var_489_shape = shape(x = hidden_states_37)[name = tensor<string, []>("op_489_shape")];623 tensor<int32, []> gather_24 = const()[name = tensor<string, []>("gather_24"), val = tensor<int32, []>(1)];624 tensor<int32, []> gather_25_indices_0 = const()[name = tensor<string, []>("gather_25_indices_0"), val = tensor<int32, []>(1)];625 tensor<int32, []> gather_25_axis_0 = const()[name = tensor<string, []>("gather_25_axis_0"), val = tensor<int32, []>(0)];626 tensor<int32, []> gather_25_batch_dims_0 = const()[name = tensor<string, []>("gather_25_batch_dims_0"), val = tensor<int32, []>(0)];627 tensor<int32, []> gather_25 = gather(axis = gather_25_axis_0, batch_dims = gather_25_batch_dims_0, indices = gather_25_indices_0, x = var_489_shape)[name = tensor<string, []>("gather_25")];628 tensor<fp32, [1, ?, 1024]> var_493 = linear(bias = encoder_layers_6_self_attn_q_proj_bias, weight = encoder_layers_6_self_attn_q_proj_weight, x = hidden_states_37)[name = tensor<string, []>("linear_36")];629 tensor<int32, [4]> concat_33x = const()[name = tensor<string, []>("concat_33x"), val = tensor<int32, [4]>([1, -1, 16, 64])];630 tensor<fp32, [1, ?, 16, 64]> var_495 = reshape(shape = concat_33x, x = var_493)[name = tensor<string, []>("op_495")];631 tensor<int32, [4]> query_13_perm_0 = const()[name = tensor<string, []>("query_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];632 tensor<fp32, [1, ?, 1024]> key_states_13 = linear(bias = encoder_layers_6_self_attn_k_proj_bias, weight = encoder_layers_6_self_attn_k_proj_weight, x = hidden_states_37)[name = tensor<string, []>("linear_37")];633 tensor<fp32, [1, ?, 1024]> value_states_13 = linear(bias = encoder_layers_6_self_attn_v_proj_bias, weight = encoder_layers_6_self_attn_v_proj_weight, x = hidden_states_37)[name = tensor<string, []>("linear_38")];634 tensor<int32, [4]> concat_34x = const()[name = tensor<string, []>("concat_34x"), val = tensor<int32, [4]>([1, -1, 16, 64])];635 tensor<fp32, [1, ?, 16, 64]> var_504 = reshape(shape = concat_34x, x = key_states_13)[name = tensor<string, []>("op_504")];636 tensor<int32, [4]> key_13_perm_0 = const()[name = tensor<string, []>("key_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];637 tensor<int32, [4]> concat_35x = const()[name = tensor<string, []>("concat_35x"), val = tensor<int32, [4]>([1, -1, 16, 64])];638 tensor<fp32, [1, ?, 16, 64]> var_507 = reshape(shape = concat_35x, x = value_states_13)[name = tensor<string, []>("op_507")];639 tensor<int32, [4]> value_13_perm_0 = const()[name = tensor<string, []>("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];640 tensor<fp32, [1, 16, ?, 64]> key_13 = transpose(perm = key_13_perm_0, x = var_504)[name = tensor<string, []>("transpose_22")];641 tensor<int32, [4]> var_509_shape = shape(x = key_13)[name = tensor<string, []>("op_509_shape")];642 tensor<int32, []> gather_26_indices_0 = const()[name = tensor<string, []>("gather_26_indices_0"), val = tensor<int32, []>(2)];643 tensor<int32, []> gather_26_axis_0 = const()[name = tensor<string, []>("gather_26_axis_0"), val = tensor<int32, []>(0)];644 tensor<int32, []> gather_26_batch_dims_0 = const()[name = tensor<string, []>("gather_26_batch_dims_0"), val = tensor<int32, []>(0)];645 tensor<int32, []> gather_26 = gather(axis = gather_26_axis_0, batch_dims = gather_26_batch_dims_0, indices = gather_26_indices_0, x = var_509_shape)[name = tensor<string, []>("gather_26")];646 tensor<int32, []> concat_36_values0_0 = const()[name = tensor<string, []>("concat_36_values0_0"), val = tensor<int32, []>(0)];647 tensor<int32, []> concat_36_values1_0 = const()[name = tensor<string, []>("concat_36_values1_0"), val = tensor<int32, []>(0)];648 tensor<int32, []> concat_36_values2_0 = const()[name = tensor<string, []>("concat_36_values2_0"), val = tensor<int32, []>(0)];649 tensor<int32, []> concat_36_axis_0 = const()[name = tensor<string, []>("concat_36_axis_0"), val = tensor<int32, []>(0)];650 tensor<bool, []> concat_36_interleave_0 = const()[name = tensor<string, []>("concat_36_interleave_0"), val = tensor<bool, []>(false)];651 tensor<int32, [4]> concat_36 = concat(axis = concat_36_axis_0, interleave = concat_36_interleave_0, values = (concat_36_values0_0, concat_36_values1_0, concat_36_values2_0, gather_26))[name = tensor<string, []>("concat_36")];652 tensor<int32, [4]> attention_mask_17_begin_0 = const()[name = tensor<string, []>("attention_mask_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];653 tensor<bool, [4]> attention_mask_17_end_mask_0 = const()[name = tensor<string, []>("attention_mask_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];654 tensor<fp32, [?, ?, ?, ?]> attention_mask_17 = slice_by_index(begin = attention_mask_17_begin_0, end = concat_36, end_mask = attention_mask_17_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_17")];655 tensor<fp32, [1, 16, ?, 64]> query_13 = transpose(perm = query_13_perm_0, x = var_495)[name = tensor<string, []>("transpose_23")];656 tensor<fp32, [1, 16, ?, 64]> mul_6 = mul(x = query_13, y = var_7)[name = tensor<string, []>("mul_6")];657 tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)];658 tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)];659 tensor<fp32, [1, 16, ?, ?]> matmul_6 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = mul_6, y = key_13)[name = tensor<string, []>("matmul_6")];660 tensor<fp32, [?, 16, ?, ?]> add_6 = add(x = matmul_6, y = attention_mask_17)[name = tensor<string, []>("add_6")];661 tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)];662 tensor<fp32, [?, 16, ?, ?]> softmax_6 = softmax(axis = softmax_6_axis_0, x = add_6)[name = tensor<string, []>("softmax_6")];663 tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];664 tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];665 tensor<fp32, [1, 16, ?, 64]> value_13 = transpose(perm = value_13_perm_0, x = var_507)[name = tensor<string, []>("transpose_21")];666 tensor<fp32, [?, 16, ?, 64]> attn_output_25 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_6, y = value_13)[name = tensor<string, []>("attn_output_25")];667 tensor<int32, [4]> var_515_perm_0 = const()[name = tensor<string, []>("op_515_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];668 tensor<int32, []> concat_37_axis_0 = const()[name = tensor<string, []>("concat_37_axis_0"), val = tensor<int32, []>(0)];669 tensor<bool, []> concat_37_interleave_0 = const()[name = tensor<string, []>("concat_37_interleave_0"), val = tensor<bool, []>(false)];670 tensor<int32, [3]> concat_37 = concat(axis = concat_37_axis_0, interleave = concat_37_interleave_0, values = (gather_24, gather_25, var_23))[name = tensor<string, []>("concat_37")];671 tensor<fp32, [?, ?, 16, 64]> var_515 = transpose(perm = var_515_perm_0, x = attn_output_25)[name = tensor<string, []>("transpose_20")];672 tensor<fp32, [1, ?, ?]> var_518 = reshape(shape = concat_37, x = var_515)[name = tensor<string, []>("op_518")];673 tensor<fp32, [1, ?, 1024]> input_117 = linear(bias = encoder_layers_6_self_attn_out_proj_bias, weight = encoder_layers_6_self_attn_out_proj_weight, x = var_518)[name = tensor<string, []>("linear_39")];674 tensor<fp32, [1, ?, 1024]> input_119 = add(x = input_113, y = input_117)[name = tensor<string, []>("input_119")];675 tensor<int32, [1]> input_121_axes_0 = const()[name = tensor<string, []>("input_121_axes_0"), val = tensor<int32, [1]>([-1])];676 tensor<fp32, [1, ?, 1024]> input_121 = layer_norm(axes = input_121_axes_0, beta = encoder_layers_6_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_6_final_layer_norm_weight, x = input_119)[name = tensor<string, []>("input_121")];677 tensor<fp32, [1, ?, 4096]> input_123 = linear(bias = encoder_layers_6_fc1_bias, weight = encoder_layers_6_fc1_weight, x = input_121)[name = tensor<string, []>("linear_40")];678 tensor<fp32, [1, ?, 4096]> input_125 = relu(x = input_123)[name = tensor<string, []>("input_125")];679 tensor<fp32, [1, ?, 1024]> input_129 = linear(bias = encoder_layers_6_fc2_bias, weight = encoder_layers_6_fc2_weight, x = input_125)[name = tensor<string, []>("linear_41")];680 tensor<fp32, [1, ?, 1024]> input_131 = add(x = input_119, y = input_129)[name = tensor<string, []>("input_131")];681 tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])];682 tensor<fp32, [1, ?, 1024]> hidden_states_43 = layer_norm(axes = hidden_states_43_axes_0, beta = encoder_layers_7_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_7_self_attn_layer_norm_weight, x = input_131)[name = tensor<string, []>("hidden_states_43")];683 tensor<int32, [3]> var_552_shape = shape(x = hidden_states_43)[name = tensor<string, []>("op_552_shape")];684 tensor<int32, []> gather_27 = const()[name = tensor<string, []>("gather_27"), val = tensor<int32, []>(1)];685 tensor<int32, []> gather_28_indices_0 = const()[name = tensor<string, []>("gather_28_indices_0"), val = tensor<int32, []>(1)];686 tensor<int32, []> gather_28_axis_0 = const()[name = tensor<string, []>("gather_28_axis_0"), val = tensor<int32, []>(0)];687 tensor<int32, []> gather_28_batch_dims_0 = const()[name = tensor<string, []>("gather_28_batch_dims_0"), val = tensor<int32, []>(0)];688 tensor<int32, []> gather_28 = gather(axis = gather_28_axis_0, batch_dims = gather_28_batch_dims_0, indices = gather_28_indices_0, x = var_552_shape)[name = tensor<string, []>("gather_28")];689 tensor<fp32, [1, ?, 1024]> var_556 = linear(bias = encoder_layers_7_self_attn_q_proj_bias, weight = encoder_layers_7_self_attn_q_proj_weight, x = hidden_states_43)[name = tensor<string, []>("linear_42")];690 tensor<int32, [4]> concat_38x = const()[name = tensor<string, []>("concat_38x"), val = tensor<int32, [4]>([1, -1, 16, 64])];691 tensor<fp32, [1, ?, 16, 64]> var_558 = reshape(shape = concat_38x, x = var_556)[name = tensor<string, []>("op_558")];692 tensor<int32, [4]> query_15_perm_0 = const()[name = tensor<string, []>("query_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];693 tensor<fp32, [1, ?, 1024]> key_states_15 = linear(bias = encoder_layers_7_self_attn_k_proj_bias, weight = encoder_layers_7_self_attn_k_proj_weight, x = hidden_states_43)[name = tensor<string, []>("linear_43")];694 tensor<fp32, [1, ?, 1024]> value_states_15 = linear(bias = encoder_layers_7_self_attn_v_proj_bias, weight = encoder_layers_7_self_attn_v_proj_weight, x = hidden_states_43)[name = tensor<string, []>("linear_44")];695 tensor<int32, [4]> concat_39x = const()[name = tensor<string, []>("concat_39x"), val = tensor<int32, [4]>([1, -1, 16, 64])];696 tensor<fp32, [1, ?, 16, 64]> var_567 = reshape(shape = concat_39x, x = key_states_15)[name = tensor<string, []>("op_567")];697 tensor<int32, [4]> key_15_perm_0 = const()[name = tensor<string, []>("key_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];698 tensor<int32, [4]> concat_40x = const()[name = tensor<string, []>("concat_40x"), val = tensor<int32, [4]>([1, -1, 16, 64])];699 tensor<fp32, [1, ?, 16, 64]> var_570 = reshape(shape = concat_40x, x = value_states_15)[name = tensor<string, []>("op_570")];700 tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];701 tensor<fp32, [1, 16, ?, 64]> key_15 = transpose(perm = key_15_perm_0, x = var_567)[name = tensor<string, []>("transpose_18")];702 tensor<int32, [4]> var_572_shape = shape(x = key_15)[name = tensor<string, []>("op_572_shape")];703 tensor<int32, []> gather_29_indices_0 = const()[name = tensor<string, []>("gather_29_indices_0"), val = tensor<int32, []>(2)];704 tensor<int32, []> gather_29_axis_0 = const()[name = tensor<string, []>("gather_29_axis_0"), val = tensor<int32, []>(0)];705 tensor<int32, []> gather_29_batch_dims_0 = const()[name = tensor<string, []>("gather_29_batch_dims_0"), val = tensor<int32, []>(0)];706 tensor<int32, []> gather_29 = gather(axis = gather_29_axis_0, batch_dims = gather_29_batch_dims_0, indices = gather_29_indices_0, x = var_572_shape)[name = tensor<string, []>("gather_29")];707 tensor<int32, []> concat_41_values0_0 = const()[name = tensor<string, []>("concat_41_values0_0"), val = tensor<int32, []>(0)];708 tensor<int32, []> concat_41_values1_0 = const()[name = tensor<string, []>("concat_41_values1_0"), val = tensor<int32, []>(0)];709 tensor<int32, []> concat_41_values2_0 = const()[name = tensor<string, []>("concat_41_values2_0"), val = tensor<int32, []>(0)];710 tensor<int32, []> concat_41_axis_0 = const()[name = tensor<string, []>("concat_41_axis_0"), val = tensor<int32, []>(0)];711 tensor<bool, []> concat_41_interleave_0 = const()[name = tensor<string, []>("concat_41_interleave_0"), val = tensor<bool, []>(false)];712 tensor<int32, [4]> concat_41 = concat(axis = concat_41_axis_0, interleave = concat_41_interleave_0, values = (concat_41_values0_0, concat_41_values1_0, concat_41_values2_0, gather_29))[name = tensor<string, []>("concat_41")];713 tensor<int32, [4]> attention_mask_19_begin_0 = const()[name = tensor<string, []>("attention_mask_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];714 tensor<bool, [4]> attention_mask_19_end_mask_0 = const()[name = tensor<string, []>("attention_mask_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];715 tensor<fp32, [?, ?, ?, ?]> attention_mask_19 = slice_by_index(begin = attention_mask_19_begin_0, end = concat_41, end_mask = attention_mask_19_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_19")];716 tensor<fp32, [1, 16, ?, 64]> query_15 = transpose(perm = query_15_perm_0, x = var_558)[name = tensor<string, []>("transpose_19")];717 tensor<fp32, [1, 16, ?, 64]> mul_7 = mul(x = query_15, y = var_7)[name = tensor<string, []>("mul_7")];718 tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)];719 tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)];720 tensor<fp32, [1, 16, ?, ?]> matmul_7 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = mul_7, y = key_15)[name = tensor<string, []>("matmul_7")];721 tensor<fp32, [?, 16, ?, ?]> add_7 = add(x = matmul_7, y = attention_mask_19)[name = tensor<string, []>("add_7")];722 tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)];723 tensor<fp32, [?, 16, ?, ?]> softmax_7 = softmax(axis = softmax_7_axis_0, x = add_7)[name = tensor<string, []>("softmax_7")];724 tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];725 tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];726 tensor<fp32, [1, 16, ?, 64]> value_15 = transpose(perm = value_15_perm_0, x = var_570)[name = tensor<string, []>("transpose_17")];727 tensor<fp32, [?, 16, ?, 64]> attn_output_29 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7, y = value_15)[name = tensor<string, []>("attn_output_29")];728 tensor<int32, [4]> var_578_perm_0 = const()[name = tensor<string, []>("op_578_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];729 tensor<int32, []> concat_42_axis_0 = const()[name = tensor<string, []>("concat_42_axis_0"), val = tensor<int32, []>(0)];730 tensor<bool, []> concat_42_interleave_0 = const()[name = tensor<string, []>("concat_42_interleave_0"), val = tensor<bool, []>(false)];731 tensor<int32, [3]> concat_42 = concat(axis = concat_42_axis_0, interleave = concat_42_interleave_0, values = (gather_27, gather_28, var_23))[name = tensor<string, []>("concat_42")];732 tensor<fp32, [?, ?, 16, 64]> var_578 = transpose(perm = var_578_perm_0, x = attn_output_29)[name = tensor<string, []>("transpose_16")];733 tensor<fp32, [1, ?, ?]> var_581 = reshape(shape = concat_42, x = var_578)[name = tensor<string, []>("op_581")];734 tensor<fp32, [1, ?, 1024]> input_135 = linear(bias = encoder_layers_7_self_attn_out_proj_bias, weight = encoder_layers_7_self_attn_out_proj_weight, x = var_581)[name = tensor<string, []>("linear_45")];735 tensor<fp32, [1, ?, 1024]> input_137 = add(x = input_131, y = input_135)[name = tensor<string, []>("input_137")];736 tensor<int32, [1]> input_139_axes_0 = const()[name = tensor<string, []>("input_139_axes_0"), val = tensor<int32, [1]>([-1])];737 tensor<fp32, [1, ?, 1024]> input_139 = layer_norm(axes = input_139_axes_0, beta = encoder_layers_7_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_7_final_layer_norm_weight, x = input_137)[name = tensor<string, []>("input_139")];738 tensor<fp32, [1, ?, 4096]> input_141 = linear(bias = encoder_layers_7_fc1_bias, weight = encoder_layers_7_fc1_weight, x = input_139)[name = tensor<string, []>("linear_46")];739 tensor<fp32, [1, ?, 4096]> input_143 = relu(x = input_141)[name = tensor<string, []>("input_143")];740 tensor<fp32, [1, ?, 1024]> input_147 = linear(bias = encoder_layers_7_fc2_bias, weight = encoder_layers_7_fc2_weight, x = input_143)[name = tensor<string, []>("linear_47")];741 tensor<fp32, [1, ?, 1024]> input_149 = add(x = input_137, y = input_147)[name = tensor<string, []>("input_149")];742 tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])];743 tensor<fp32, [1, ?, 1024]> hidden_states_49 = layer_norm(axes = hidden_states_49_axes_0, beta = encoder_layers_8_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_8_self_attn_layer_norm_weight, x = input_149)[name = tensor<string, []>("hidden_states_49")];744 tensor<int32, [3]> var_615_shape = shape(x = hidden_states_49)[name = tensor<string, []>("op_615_shape")];745 tensor<int32, []> gather_30 = const()[name = tensor<string, []>("gather_30"), val = tensor<int32, []>(1)];746 tensor<int32, []> gather_31_indices_0 = const()[name = tensor<string, []>("gather_31_indices_0"), val = tensor<int32, []>(1)];747 tensor<int32, []> gather_31_axis_0 = const()[name = tensor<string, []>("gather_31_axis_0"), val = tensor<int32, []>(0)];748 tensor<int32, []> gather_31_batch_dims_0 = const()[name = tensor<string, []>("gather_31_batch_dims_0"), val = tensor<int32, []>(0)];749 tensor<int32, []> gather_31 = gather(axis = gather_31_axis_0, batch_dims = gather_31_batch_dims_0, indices = gather_31_indices_0, x = var_615_shape)[name = tensor<string, []>("gather_31")];750 tensor<fp32, [1, ?, 1024]> var_619 = linear(bias = encoder_layers_8_self_attn_q_proj_bias, weight = encoder_layers_8_self_attn_q_proj_weight, x = hidden_states_49)[name = tensor<string, []>("linear_48")];751 tensor<int32, [4]> concat_43x = const()[name = tensor<string, []>("concat_43x"), val = tensor<int32, [4]>([1, -1, 16, 64])];752 tensor<fp32, [1, ?, 16, 64]> var_621 = reshape(shape = concat_43x, x = var_619)[name = tensor<string, []>("op_621")];753 tensor<int32, [4]> query_17_perm_0 = const()[name = tensor<string, []>("query_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];754 tensor<fp32, [1, ?, 1024]> key_states_17 = linear(bias = encoder_layers_8_self_attn_k_proj_bias, weight = encoder_layers_8_self_attn_k_proj_weight, x = hidden_states_49)[name = tensor<string, []>("linear_49")];755 tensor<fp32, [1, ?, 1024]> value_states_17 = linear(bias = encoder_layers_8_self_attn_v_proj_bias, weight = encoder_layers_8_self_attn_v_proj_weight, x = hidden_states_49)[name = tensor<string, []>("linear_50")];756 tensor<int32, [4]> concat_44x = const()[name = tensor<string, []>("concat_44x"), val = tensor<int32, [4]>([1, -1, 16, 64])];757 tensor<fp32, [1, ?, 16, 64]> var_630 = reshape(shape = concat_44x, x = key_states_17)[name = tensor<string, []>("op_630")];758 tensor<int32, [4]> key_17_perm_0 = const()[name = tensor<string, []>("key_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];759 tensor<int32, [4]> concat_45x = const()[name = tensor<string, []>("concat_45x"), val = tensor<int32, [4]>([1, -1, 16, 64])];760 tensor<fp32, [1, ?, 16, 64]> var_633 = reshape(shape = concat_45x, x = value_states_17)[name = tensor<string, []>("op_633")];761 tensor<int32, [4]> value_17_perm_0 = const()[name = tensor<string, []>("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];762 tensor<fp32, [1, 16, ?, 64]> key_17 = transpose(perm = key_17_perm_0, x = var_630)[name = tensor<string, []>("transpose_14")];763 tensor<int32, [4]> var_635_shape = shape(x = key_17)[name = tensor<string, []>("op_635_shape")];764 tensor<int32, []> gather_32_indices_0 = const()[name = tensor<string, []>("gather_32_indices_0"), val = tensor<int32, []>(2)];765 tensor<int32, []> gather_32_axis_0 = const()[name = tensor<string, []>("gather_32_axis_0"), val = tensor<int32, []>(0)];766 tensor<int32, []> gather_32_batch_dims_0 = const()[name = tensor<string, []>("gather_32_batch_dims_0"), val = tensor<int32, []>(0)];767 tensor<int32, []> gather_32 = gather(axis = gather_32_axis_0, batch_dims = gather_32_batch_dims_0, indices = gather_32_indices_0, x = var_635_shape)[name = tensor<string, []>("gather_32")];768 tensor<int32, []> concat_46_values0_0 = const()[name = tensor<string, []>("concat_46_values0_0"), val = tensor<int32, []>(0)];769 tensor<int32, []> concat_46_values1_0 = const()[name = tensor<string, []>("concat_46_values1_0"), val = tensor<int32, []>(0)];770 tensor<int32, []> concat_46_values2_0 = const()[name = tensor<string, []>("concat_46_values2_0"), val = tensor<int32, []>(0)];771 tensor<int32, []> concat_46_axis_0 = const()[name = tensor<string, []>("concat_46_axis_0"), val = tensor<int32, []>(0)];772 tensor<bool, []> concat_46_interleave_0 = const()[name = tensor<string, []>("concat_46_interleave_0"), val = tensor<bool, []>(false)];773 tensor<int32, [4]> concat_46 = concat(axis = concat_46_axis_0, interleave = concat_46_interleave_0, values = (concat_46_values0_0, concat_46_values1_0, concat_46_values2_0, gather_32))[name = tensor<string, []>("concat_46")];774 tensor<int32, [4]> attention_mask_21_begin_0 = const()[name = tensor<string, []>("attention_mask_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];775 tensor<bool, [4]> attention_mask_21_end_mask_0 = const()[name = tensor<string, []>("attention_mask_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];776 tensor<fp32, [?, ?, ?, ?]> attention_mask_21 = slice_by_index(begin = attention_mask_21_begin_0, end = concat_46, end_mask = attention_mask_21_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_21")];777 tensor<fp32, [1, 16, ?, 64]> query_17 = transpose(perm = query_17_perm_0, x = var_621)[name = tensor<string, []>("transpose_15")];778 tensor<fp32, [1, 16, ?, 64]> mul_8 = mul(x = query_17, y = var_7)[name = tensor<string, []>("mul_8")];779 tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)];780 tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)];781 tensor<fp32, [1, 16, ?, ?]> matmul_8 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = mul_8, y = key_17)[name = tensor<string, []>("matmul_8")];782 tensor<fp32, [?, 16, ?, ?]> add_8 = add(x = matmul_8, y = attention_mask_21)[name = tensor<string, []>("add_8")];783 tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)];784 tensor<fp32, [?, 16, ?, ?]> softmax_8 = softmax(axis = softmax_8_axis_0, x = add_8)[name = tensor<string, []>("softmax_8")];785 tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];786 tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];787 tensor<fp32, [1, 16, ?, 64]> value_17 = transpose(perm = value_17_perm_0, x = var_633)[name = tensor<string, []>("transpose_13")];788 tensor<fp32, [?, 16, ?, 64]> attn_output_33 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8, y = value_17)[name = tensor<string, []>("attn_output_33")];789 tensor<int32, [4]> var_641_perm_0 = const()[name = tensor<string, []>("op_641_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];790 tensor<int32, []> concat_47_axis_0 = const()[name = tensor<string, []>("concat_47_axis_0"), val = tensor<int32, []>(0)];791 tensor<bool, []> concat_47_interleave_0 = const()[name = tensor<string, []>("concat_47_interleave_0"), val = tensor<bool, []>(false)];792 tensor<int32, [3]> concat_47 = concat(axis = concat_47_axis_0, interleave = concat_47_interleave_0, values = (gather_30, gather_31, var_23))[name = tensor<string, []>("concat_47")];793 tensor<fp32, [?, ?, 16, 64]> var_641 = transpose(perm = var_641_perm_0, x = attn_output_33)[name = tensor<string, []>("transpose_12")];794 tensor<fp32, [1, ?, ?]> var_644 = reshape(shape = concat_47, x = var_641)[name = tensor<string, []>("op_644")];795 tensor<fp32, [1, ?, 1024]> input_153 = linear(bias = encoder_layers_8_self_attn_out_proj_bias, weight = encoder_layers_8_self_attn_out_proj_weight, x = var_644)[name = tensor<string, []>("linear_51")];796 tensor<fp32, [1, ?, 1024]> input_155 = add(x = input_149, y = input_153)[name = tensor<string, []>("input_155")];797 tensor<int32, [1]> input_157_axes_0 = const()[name = tensor<string, []>("input_157_axes_0"), val = tensor<int32, [1]>([-1])];798 tensor<fp32, [1, ?, 1024]> input_157 = layer_norm(axes = input_157_axes_0, beta = encoder_layers_8_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_8_final_layer_norm_weight, x = input_155)[name = tensor<string, []>("input_157")];799 tensor<fp32, [1, ?, 4096]> input_159 = linear(bias = encoder_layers_8_fc1_bias, weight = encoder_layers_8_fc1_weight, x = input_157)[name = tensor<string, []>("linear_52")];800 tensor<fp32, [1, ?, 4096]> input_161 = relu(x = input_159)[name = tensor<string, []>("input_161")];801 tensor<fp32, [1, ?, 1024]> input_165 = linear(bias = encoder_layers_8_fc2_bias, weight = encoder_layers_8_fc2_weight, x = input_161)[name = tensor<string, []>("linear_53")];802 tensor<fp32, [1, ?, 1024]> input_167 = add(x = input_155, y = input_165)[name = tensor<string, []>("input_167")];803 tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])];804 tensor<fp32, [1, ?, 1024]> hidden_states_55 = layer_norm(axes = hidden_states_55_axes_0, beta = encoder_layers_9_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_9_self_attn_layer_norm_weight, x = input_167)[name = tensor<string, []>("hidden_states_55")];805 tensor<int32, [3]> var_678_shape = shape(x = hidden_states_55)[name = tensor<string, []>("op_678_shape")];806 tensor<int32, []> gather_33 = const()[name = tensor<string, []>("gather_33"), val = tensor<int32, []>(1)];807 tensor<int32, []> gather_34_indices_0 = const()[name = tensor<string, []>("gather_34_indices_0"), val = tensor<int32, []>(1)];808 tensor<int32, []> gather_34_axis_0 = const()[name = tensor<string, []>("gather_34_axis_0"), val = tensor<int32, []>(0)];809 tensor<int32, []> gather_34_batch_dims_0 = const()[name = tensor<string, []>("gather_34_batch_dims_0"), val = tensor<int32, []>(0)];810 tensor<int32, []> gather_34 = gather(axis = gather_34_axis_0, batch_dims = gather_34_batch_dims_0, indices = gather_34_indices_0, x = var_678_shape)[name = tensor<string, []>("gather_34")];811 tensor<fp32, [1, ?, 1024]> var_682 = linear(bias = encoder_layers_9_self_attn_q_proj_bias, weight = encoder_layers_9_self_attn_q_proj_weight, x = hidden_states_55)[name = tensor<string, []>("linear_54")];812 tensor<int32, [4]> concat_48x = const()[name = tensor<string, []>("concat_48x"), val = tensor<int32, [4]>([1, -1, 16, 64])];813 tensor<fp32, [1, ?, 16, 64]> var_684 = reshape(shape = concat_48x, x = var_682)[name = tensor<string, []>("op_684")];814 tensor<int32, [4]> query_19_perm_0 = const()[name = tensor<string, []>("query_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];815 tensor<fp32, [1, ?, 1024]> key_states_19 = linear(bias = encoder_layers_9_self_attn_k_proj_bias, weight = encoder_layers_9_self_attn_k_proj_weight, x = hidden_states_55)[name = tensor<string, []>("linear_55")];816 tensor<fp32, [1, ?, 1024]> value_states_19 = linear(bias = encoder_layers_9_self_attn_v_proj_bias, weight = encoder_layers_9_self_attn_v_proj_weight, x = hidden_states_55)[name = tensor<string, []>("linear_56")];817 tensor<int32, [4]> concat_49x = const()[name = tensor<string, []>("concat_49x"), val = tensor<int32, [4]>([1, -1, 16, 64])];818 tensor<fp32, [1, ?, 16, 64]> var_693 = reshape(shape = concat_49x, x = key_states_19)[name = tensor<string, []>("op_693")];819 tensor<int32, [4]> key_19_perm_0 = const()[name = tensor<string, []>("key_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];820 tensor<int32, [4]> concat_50x = const()[name = tensor<string, []>("concat_50x"), val = tensor<int32, [4]>([1, -1, 16, 64])];821 tensor<fp32, [1, ?, 16, 64]> var_696 = reshape(shape = concat_50x, x = value_states_19)[name = tensor<string, []>("op_696")];822 tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];823 tensor<fp32, [1, 16, ?, 64]> key_19 = transpose(perm = key_19_perm_0, x = var_693)[name = tensor<string, []>("transpose_10")];824 tensor<int32, [4]> var_698_shape = shape(x = key_19)[name = tensor<string, []>("op_698_shape")];825 tensor<int32, []> gather_35_indices_0 = const()[name = tensor<string, []>("gather_35_indices_0"), val = tensor<int32, []>(2)];826 tensor<int32, []> gather_35_axis_0 = const()[name = tensor<string, []>("gather_35_axis_0"), val = tensor<int32, []>(0)];827 tensor<int32, []> gather_35_batch_dims_0 = const()[name = tensor<string, []>("gather_35_batch_dims_0"), val = tensor<int32, []>(0)];828 tensor<int32, []> gather_35 = gather(axis = gather_35_axis_0, batch_dims = gather_35_batch_dims_0, indices = gather_35_indices_0, x = var_698_shape)[name = tensor<string, []>("gather_35")];829 tensor<int32, []> concat_51_values0_0 = const()[name = tensor<string, []>("concat_51_values0_0"), val = tensor<int32, []>(0)];830 tensor<int32, []> concat_51_values1_0 = const()[name = tensor<string, []>("concat_51_values1_0"), val = tensor<int32, []>(0)];831 tensor<int32, []> concat_51_values2_0 = const()[name = tensor<string, []>("concat_51_values2_0"), val = tensor<int32, []>(0)];832 tensor<int32, []> concat_51_axis_0 = const()[name = tensor<string, []>("concat_51_axis_0"), val = tensor<int32, []>(0)];833 tensor<bool, []> concat_51_interleave_0 = const()[name = tensor<string, []>("concat_51_interleave_0"), val = tensor<bool, []>(false)];834 tensor<int32, [4]> concat_51 = concat(axis = concat_51_axis_0, interleave = concat_51_interleave_0, values = (concat_51_values0_0, concat_51_values1_0, concat_51_values2_0, gather_35))[name = tensor<string, []>("concat_51")];835 tensor<int32, [4]> attention_mask_23_begin_0 = const()[name = tensor<string, []>("attention_mask_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];836 tensor<bool, [4]> attention_mask_23_end_mask_0 = const()[name = tensor<string, []>("attention_mask_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];837 tensor<fp32, [?, ?, ?, ?]> attention_mask_23 = slice_by_index(begin = attention_mask_23_begin_0, end = concat_51, end_mask = attention_mask_23_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_23")];838 tensor<fp32, [1, 16, ?, 64]> query_19 = transpose(perm = query_19_perm_0, x = var_684)[name = tensor<string, []>("transpose_11")];839 tensor<fp32, [1, 16, ?, 64]> mul_9 = mul(x = query_19, y = var_7)[name = tensor<string, []>("mul_9")];840 tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)];841 tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)];842 tensor<fp32, [1, 16, ?, ?]> matmul_9 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = mul_9, y = key_19)[name = tensor<string, []>("matmul_9")];843 tensor<fp32, [?, 16, ?, ?]> add_9 = add(x = matmul_9, y = attention_mask_23)[name = tensor<string, []>("add_9")];844 tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)];845 tensor<fp32, [?, 16, ?, ?]> softmax_9 = softmax(axis = softmax_9_axis_0, x = add_9)[name = tensor<string, []>("softmax_9")];846 tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];847 tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];848 tensor<fp32, [1, 16, ?, 64]> value_19 = transpose(perm = value_19_perm_0, x = var_696)[name = tensor<string, []>("transpose_9")];849 tensor<fp32, [?, 16, ?, 64]> attn_output_37 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_9, y = value_19)[name = tensor<string, []>("attn_output_37")];850 tensor<int32, [4]> var_704_perm_0 = const()[name = tensor<string, []>("op_704_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];851 tensor<int32, []> concat_52_axis_0 = const()[name = tensor<string, []>("concat_52_axis_0"), val = tensor<int32, []>(0)];852 tensor<bool, []> concat_52_interleave_0 = const()[name = tensor<string, []>("concat_52_interleave_0"), val = tensor<bool, []>(false)];853 tensor<int32, [3]> concat_52 = concat(axis = concat_52_axis_0, interleave = concat_52_interleave_0, values = (gather_33, gather_34, var_23))[name = tensor<string, []>("concat_52")];854 tensor<fp32, [?, ?, 16, 64]> var_704 = transpose(perm = var_704_perm_0, x = attn_output_37)[name = tensor<string, []>("transpose_8")];855 tensor<fp32, [1, ?, ?]> var_707 = reshape(shape = concat_52, x = var_704)[name = tensor<string, []>("op_707")];856 tensor<fp32, [1, ?, 1024]> input_171 = linear(bias = encoder_layers_9_self_attn_out_proj_bias, weight = encoder_layers_9_self_attn_out_proj_weight, x = var_707)[name = tensor<string, []>("linear_57")];857 tensor<fp32, [1, ?, 1024]> input_173 = add(x = input_167, y = input_171)[name = tensor<string, []>("input_173")];858 tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([-1])];859 tensor<fp32, [1, ?, 1024]> input_175 = layer_norm(axes = input_175_axes_0, beta = encoder_layers_9_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_9_final_layer_norm_weight, x = input_173)[name = tensor<string, []>("input_175")];860 tensor<fp32, [1, ?, 4096]> input_177 = linear(bias = encoder_layers_9_fc1_bias, weight = encoder_layers_9_fc1_weight, x = input_175)[name = tensor<string, []>("linear_58")];861 tensor<fp32, [1, ?, 4096]> input_179 = relu(x = input_177)[name = tensor<string, []>("input_179")];862 tensor<fp32, [1, ?, 1024]> input_183 = linear(bias = encoder_layers_9_fc2_bias, weight = encoder_layers_9_fc2_weight, x = input_179)[name = tensor<string, []>("linear_59")];863 tensor<fp32, [1, ?, 1024]> input_185 = add(x = input_173, y = input_183)[name = tensor<string, []>("input_185")];864 tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])];865 tensor<fp32, [1, ?, 1024]> hidden_states_61 = layer_norm(axes = hidden_states_61_axes_0, beta = encoder_layers_10_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_10_self_attn_layer_norm_weight, x = input_185)[name = tensor<string, []>("hidden_states_61")];866 tensor<int32, [3]> var_741_shape = shape(x = hidden_states_61)[name = tensor<string, []>("op_741_shape")];867 tensor<int32, []> gather_36 = const()[name = tensor<string, []>("gather_36"), val = tensor<int32, []>(1)];868 tensor<int32, []> gather_37_indices_0 = const()[name = tensor<string, []>("gather_37_indices_0"), val = tensor<int32, []>(1)];869 tensor<int32, []> gather_37_axis_0 = const()[name = tensor<string, []>("gather_37_axis_0"), val = tensor<int32, []>(0)];870 tensor<int32, []> gather_37_batch_dims_0 = const()[name = tensor<string, []>("gather_37_batch_dims_0"), val = tensor<int32, []>(0)];871 tensor<int32, []> gather_37 = gather(axis = gather_37_axis_0, batch_dims = gather_37_batch_dims_0, indices = gather_37_indices_0, x = var_741_shape)[name = tensor<string, []>("gather_37")];872 tensor<fp32, [1, ?, 1024]> var_745 = linear(bias = encoder_layers_10_self_attn_q_proj_bias, weight = encoder_layers_10_self_attn_q_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_60")];873 tensor<int32, [4]> concat_53x = const()[name = tensor<string, []>("concat_53x"), val = tensor<int32, [4]>([1, -1, 16, 64])];874 tensor<fp32, [1, ?, 16, 64]> var_747 = reshape(shape = concat_53x, x = var_745)[name = tensor<string, []>("op_747")];875 tensor<int32, [4]> query_21_perm_0 = const()[name = tensor<string, []>("query_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];876 tensor<fp32, [1, ?, 1024]> key_states_21 = linear(bias = encoder_layers_10_self_attn_k_proj_bias, weight = encoder_layers_10_self_attn_k_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_61")];877 tensor<fp32, [1, ?, 1024]> value_states_21 = linear(bias = encoder_layers_10_self_attn_v_proj_bias, weight = encoder_layers_10_self_attn_v_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_62")];878 tensor<int32, [4]> concat_54x = const()[name = tensor<string, []>("concat_54x"), val = tensor<int32, [4]>([1, -1, 16, 64])];879 tensor<fp32, [1, ?, 16, 64]> var_756 = reshape(shape = concat_54x, x = key_states_21)[name = tensor<string, []>("op_756")];880 tensor<int32, [4]> key_21_perm_0 = const()[name = tensor<string, []>("key_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];881 tensor<int32, [4]> concat_55x = const()[name = tensor<string, []>("concat_55x"), val = tensor<int32, [4]>([1, -1, 16, 64])];882 tensor<fp32, [1, ?, 16, 64]> var_759 = reshape(shape = concat_55x, x = value_states_21)[name = tensor<string, []>("op_759")];883 tensor<int32, [4]> value_21_perm_0 = const()[name = tensor<string, []>("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];884 tensor<fp32, [1, 16, ?, 64]> key_21 = transpose(perm = key_21_perm_0, x = var_756)[name = tensor<string, []>("transpose_6")];885 tensor<int32, [4]> var_761_shape = shape(x = key_21)[name = tensor<string, []>("op_761_shape")];886 tensor<int32, []> gather_38_indices_0 = const()[name = tensor<string, []>("gather_38_indices_0"), val = tensor<int32, []>(2)];887 tensor<int32, []> gather_38_axis_0 = const()[name = tensor<string, []>("gather_38_axis_0"), val = tensor<int32, []>(0)];888 tensor<int32, []> gather_38_batch_dims_0 = const()[name = tensor<string, []>("gather_38_batch_dims_0"), val = tensor<int32, []>(0)];889 tensor<int32, []> gather_38 = gather(axis = gather_38_axis_0, batch_dims = gather_38_batch_dims_0, indices = gather_38_indices_0, x = var_761_shape)[name = tensor<string, []>("gather_38")];890 tensor<int32, []> concat_56_values0_0 = const()[name = tensor<string, []>("concat_56_values0_0"), val = tensor<int32, []>(0)];891 tensor<int32, []> concat_56_values1_0 = const()[name = tensor<string, []>("concat_56_values1_0"), val = tensor<int32, []>(0)];892 tensor<int32, []> concat_56_values2_0 = const()[name = tensor<string, []>("concat_56_values2_0"), val = tensor<int32, []>(0)];893 tensor<int32, []> concat_56_axis_0 = const()[name = tensor<string, []>("concat_56_axis_0"), val = tensor<int32, []>(0)];894 tensor<bool, []> concat_56_interleave_0 = const()[name = tensor<string, []>("concat_56_interleave_0"), val = tensor<bool, []>(false)];895 tensor<int32, [4]> concat_56 = concat(axis = concat_56_axis_0, interleave = concat_56_interleave_0, values = (concat_56_values0_0, concat_56_values1_0, concat_56_values2_0, gather_38))[name = tensor<string, []>("concat_56")];896 tensor<int32, [4]> attention_mask_25_begin_0 = const()[name = tensor<string, []>("attention_mask_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];897 tensor<bool, [4]> attention_mask_25_end_mask_0 = const()[name = tensor<string, []>("attention_mask_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];898 tensor<fp32, [?, ?, ?, ?]> attention_mask_25 = slice_by_index(begin = attention_mask_25_begin_0, end = concat_56, end_mask = attention_mask_25_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask_25")];899 tensor<fp32, [1, 16, ?, 64]> query_21 = transpose(perm = query_21_perm_0, x = var_747)[name = tensor<string, []>("transpose_7")];900 tensor<fp32, [1, 16, ?, 64]> mul_10 = mul(x = query_21, y = var_7)[name = tensor<string, []>("mul_10")];901 tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)];902 tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)];903 tensor<fp32, [1, 16, ?, ?]> matmul_10 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = mul_10, y = key_21)[name = tensor<string, []>("matmul_10")];904 tensor<fp32, [?, 16, ?, ?]> add_10 = add(x = matmul_10, y = attention_mask_25)[name = tensor<string, []>("add_10")];905 tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)];906 tensor<fp32, [?, 16, ?, ?]> softmax_10 = softmax(axis = softmax_10_axis_0, x = add_10)[name = tensor<string, []>("softmax_10")];907 tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];908 tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];909 tensor<fp32, [1, 16, ?, 64]> value_21 = transpose(perm = value_21_perm_0, x = var_759)[name = tensor<string, []>("transpose_5")];910 tensor<fp32, [?, 16, ?, 64]> attn_output_41 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10, y = value_21)[name = tensor<string, []>("attn_output_41")];911 tensor<int32, [4]> var_767_perm_0 = const()[name = tensor<string, []>("op_767_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];912 tensor<int32, []> concat_57_axis_0 = const()[name = tensor<string, []>("concat_57_axis_0"), val = tensor<int32, []>(0)];913 tensor<bool, []> concat_57_interleave_0 = const()[name = tensor<string, []>("concat_57_interleave_0"), val = tensor<bool, []>(false)];914 tensor<int32, [3]> concat_57 = concat(axis = concat_57_axis_0, interleave = concat_57_interleave_0, values = (gather_36, gather_37, var_23))[name = tensor<string, []>("concat_57")];915 tensor<fp32, [?, ?, 16, 64]> var_767 = transpose(perm = var_767_perm_0, x = attn_output_41)[name = tensor<string, []>("transpose_4")];916 tensor<fp32, [1, ?, ?]> var_770 = reshape(shape = concat_57, x = var_767)[name = tensor<string, []>("op_770")];917 tensor<fp32, [1, ?, 1024]> input_189 = linear(bias = encoder_layers_10_self_attn_out_proj_bias, weight = encoder_layers_10_self_attn_out_proj_weight, x = var_770)[name = tensor<string, []>("linear_63")];918 tensor<fp32, [1, ?, 1024]> input_191 = add(x = input_185, y = input_189)[name = tensor<string, []>("input_191")];919 tensor<int32, [1]> input_193_axes_0 = const()[name = tensor<string, []>("input_193_axes_0"), val = tensor<int32, [1]>([-1])];920 tensor<fp32, [1, ?, 1024]> input_193 = layer_norm(axes = input_193_axes_0, beta = encoder_layers_10_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_10_final_layer_norm_weight, x = input_191)[name = tensor<string, []>("input_193")];921 tensor<fp32, [1, ?, 4096]> input_195 = linear(bias = encoder_layers_10_fc1_bias, weight = encoder_layers_10_fc1_weight, x = input_193)[name = tensor<string, []>("linear_64")];922 tensor<fp32, [1, ?, 4096]> input_197 = relu(x = input_195)[name = tensor<string, []>("input_197")];923 tensor<fp32, [1, ?, 1024]> input_201 = linear(bias = encoder_layers_10_fc2_bias, weight = encoder_layers_10_fc2_weight, x = input_197)[name = tensor<string, []>("linear_65")];924 tensor<fp32, [1, ?, 1024]> input_203 = add(x = input_191, y = input_201)[name = tensor<string, []>("input_203")];925 tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])];926 tensor<fp32, [1, ?, 1024]> hidden_states_67 = layer_norm(axes = hidden_states_67_axes_0, beta = encoder_layers_11_self_attn_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_11_self_attn_layer_norm_weight, x = input_203)[name = tensor<string, []>("hidden_states_67")];927 tensor<int32, [3]> var_804_shape = shape(x = hidden_states_67)[name = tensor<string, []>("op_804_shape")];928 tensor<int32, []> gather_39 = const()[name = tensor<string, []>("gather_39"), val = tensor<int32, []>(1)];929 tensor<int32, []> gather_40_indices_0 = const()[name = tensor<string, []>("gather_40_indices_0"), val = tensor<int32, []>(1)];930 tensor<int32, []> gather_40_axis_0 = const()[name = tensor<string, []>("gather_40_axis_0"), val = tensor<int32, []>(0)];931 tensor<int32, []> gather_40_batch_dims_0 = const()[name = tensor<string, []>("gather_40_batch_dims_0"), val = tensor<int32, []>(0)];932 tensor<int32, []> gather_40 = gather(axis = gather_40_axis_0, batch_dims = gather_40_batch_dims_0, indices = gather_40_indices_0, x = var_804_shape)[name = tensor<string, []>("gather_40")];933 tensor<fp32, [1, ?, 1024]> var_808 = linear(bias = encoder_layers_11_self_attn_q_proj_bias, weight = encoder_layers_11_self_attn_q_proj_weight, x = hidden_states_67)[name = tensor<string, []>("linear_66")];934 tensor<int32, [4]> concat_58x = const()[name = tensor<string, []>("concat_58x"), val = tensor<int32, [4]>([1, -1, 16, 64])];935 tensor<fp32, [1, ?, 16, 64]> var_810 = reshape(shape = concat_58x, x = var_808)[name = tensor<string, []>("op_810")];936 tensor<int32, [4]> query_perm_0 = const()[name = tensor<string, []>("query_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];937 tensor<fp32, [1, ?, 1024]> key_states = linear(bias = encoder_layers_11_self_attn_k_proj_bias, weight = encoder_layers_11_self_attn_k_proj_weight, x = hidden_states_67)[name = tensor<string, []>("linear_67")];938 tensor<fp32, [1, ?, 1024]> value_states = linear(bias = encoder_layers_11_self_attn_v_proj_bias, weight = encoder_layers_11_self_attn_v_proj_weight, x = hidden_states_67)[name = tensor<string, []>("linear_68")];939 tensor<int32, [4]> concat_59x = const()[name = tensor<string, []>("concat_59x"), val = tensor<int32, [4]>([1, -1, 16, 64])];940 tensor<fp32, [1, ?, 16, 64]> var_819 = reshape(shape = concat_59x, x = key_states)[name = tensor<string, []>("op_819")];941 tensor<int32, [4]> key_perm_0 = const()[name = tensor<string, []>("key_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];942 tensor<int32, [4]> concat_60x = const()[name = tensor<string, []>("concat_60x"), val = tensor<int32, [4]>([1, -1, 16, 64])];943 tensor<fp32, [1, ?, 16, 64]> var_822 = reshape(shape = concat_60x, x = value_states)[name = tensor<string, []>("op_822")];944 tensor<int32, [4]> value_perm_0 = const()[name = tensor<string, []>("value_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];945 tensor<fp32, [1, 16, ?, 64]> key = transpose(perm = key_perm_0, x = var_819)[name = tensor<string, []>("transpose_2")];946 tensor<int32, [4]> var_824_shape = shape(x = key)[name = tensor<string, []>("op_824_shape")];947 tensor<int32, []> gather_41_indices_0 = const()[name = tensor<string, []>("gather_41_indices_0"), val = tensor<int32, []>(2)];948 tensor<int32, []> gather_41_axis_0 = const()[name = tensor<string, []>("gather_41_axis_0"), val = tensor<int32, []>(0)];949 tensor<int32, []> gather_41_batch_dims_0 = const()[name = tensor<string, []>("gather_41_batch_dims_0"), val = tensor<int32, []>(0)];950 tensor<int32, []> gather_41 = gather(axis = gather_41_axis_0, batch_dims = gather_41_batch_dims_0, indices = gather_41_indices_0, x = var_824_shape)[name = tensor<string, []>("gather_41")];951 tensor<int32, []> concat_61_values0_0 = const()[name = tensor<string, []>("concat_61_values0_0"), val = tensor<int32, []>(0)];952 tensor<int32, []> concat_61_values1_0 = const()[name = tensor<string, []>("concat_61_values1_0"), val = tensor<int32, []>(0)];953 tensor<int32, []> concat_61_values2_0 = const()[name = tensor<string, []>("concat_61_values2_0"), val = tensor<int32, []>(0)];954 tensor<int32, []> concat_61_axis_0 = const()[name = tensor<string, []>("concat_61_axis_0"), val = tensor<int32, []>(0)];955 tensor<bool, []> concat_61_interleave_0 = const()[name = tensor<string, []>("concat_61_interleave_0"), val = tensor<bool, []>(false)];956 tensor<int32, [4]> concat_61 = concat(axis = concat_61_axis_0, interleave = concat_61_interleave_0, values = (concat_61_values0_0, concat_61_values1_0, concat_61_values2_0, gather_41))[name = tensor<string, []>("concat_61")];957 tensor<int32, [4]> attention_mask_begin_0 = const()[name = tensor<string, []>("attention_mask_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];958 tensor<bool, [4]> attention_mask_end_mask_0 = const()[name = tensor<string, []>("attention_mask_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];959 tensor<fp32, [?, ?, ?, ?]> attention_mask_1 = slice_by_index(begin = attention_mask_begin_0, end = concat_61, end_mask = attention_mask_end_mask_0, x = attention_mask_3)[name = tensor<string, []>("attention_mask")];960 tensor<fp32, [1, 16, ?, 64]> query = transpose(perm = query_perm_0, x = var_810)[name = tensor<string, []>("transpose_3")];961 tensor<fp32, [1, 16, ?, 64]> mul_11 = mul(x = query, y = var_7)[name = tensor<string, []>("mul_11")];962 tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)];963 tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)];964 tensor<fp32, [1, 16, ?, ?]> matmul_11 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = mul_11, y = key)[name = tensor<string, []>("matmul_11")];965 tensor<fp32, [?, 16, ?, ?]> add_11 = add(x = matmul_11, y = attention_mask_1)[name = tensor<string, []>("add_11")];966 tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)];967 tensor<fp32, [?, 16, ?, ?]> softmax_11 = softmax(axis = softmax_11_axis_0, x = add_11)[name = tensor<string, []>("softmax_11")];968 tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];969 tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];970 tensor<fp32, [1, 16, ?, 64]> value = transpose(perm = value_perm_0, x = var_822)[name = tensor<string, []>("transpose_1")];971 tensor<fp32, [?, 16, ?, 64]> attn_output_45 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11, y = value)[name = tensor<string, []>("attn_output_45")];972 tensor<int32, [4]> var_830_perm_0 = const()[name = tensor<string, []>("op_830_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];973 tensor<int32, []> concat_62_axis_0 = const()[name = tensor<string, []>("concat_62_axis_0"), val = tensor<int32, []>(0)];974 tensor<bool, []> concat_62_interleave_0 = const()[name = tensor<string, []>("concat_62_interleave_0"), val = tensor<bool, []>(false)];975 tensor<int32, [3]> concat_62 = concat(axis = concat_62_axis_0, interleave = concat_62_interleave_0, values = (gather_39, gather_40, var_23))[name = tensor<string, []>("concat_62")];976 tensor<fp32, [?, ?, 16, 64]> var_830 = transpose(perm = var_830_perm_0, x = attn_output_45)[name = tensor<string, []>("transpose_0")];977 tensor<fp32, [1, ?, ?]> var_833 = reshape(shape = concat_62, x = var_830)[name = tensor<string, []>("op_833")];978 tensor<fp32, [1, ?, 1024]> input_207 = linear(bias = encoder_layers_11_self_attn_out_proj_bias, weight = encoder_layers_11_self_attn_out_proj_weight, x = var_833)[name = tensor<string, []>("linear_69")];979 tensor<fp32, [1, ?, 1024]> input_209 = add(x = input_203, y = input_207)[name = tensor<string, []>("input_209")];980 tensor<int32, [1]> input_211_axes_0 = const()[name = tensor<string, []>("input_211_axes_0"), val = tensor<int32, [1]>([-1])];981 tensor<fp32, [1, ?, 1024]> input_211 = layer_norm(axes = input_211_axes_0, beta = encoder_layers_11_final_layer_norm_bias, epsilon = var_5, gamma = encoder_layers_11_final_layer_norm_weight, x = input_209)[name = tensor<string, []>("input_211")];982 tensor<fp32, [1, ?, 4096]> input_213 = linear(bias = encoder_layers_11_fc1_bias, weight = encoder_layers_11_fc1_weight, x = input_211)[name = tensor<string, []>("linear_70")];983 tensor<fp32, [1, ?, 4096]> input_215 = relu(x = input_213)[name = tensor<string, []>("input_215")];984 tensor<fp32, [1, ?, 1024]> input_219 = linear(bias = encoder_layers_11_fc2_bias, weight = encoder_layers_11_fc2_weight, x = input_215)[name = tensor<string, []>("linear_71")];985 tensor<fp32, [1, ?, 1024]> input = add(x = input_209, y = input_219)[name = tensor<string, []>("input")];986 tensor<int32, [1]> var_857_axes_0 = const()[name = tensor<string, []>("op_857_axes_0"), val = tensor<int32, [1]>([-1])];987 tensor<fp32, [1, ?, 1024]> output = layer_norm(axes = var_857_axes_0, beta = encoder_layer_norm_bias, epsilon = var_5, gamma = encoder_layer_norm_weight, x = input)[name = tensor<string, []>("op_857")];988 } -> (output);989}