diff --git "a/NLLB_Encoder_1024.mlmodelc/model.mil" "b/NLLB_Encoder_1024.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/NLLB_Encoder_1024.mlmodelc/model.mil" @@ -0,0 +1,989 @@ +program(1.0) +[buildInfo = dict, tensor>({{"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"}})] +{ + func main(tensor attention_mask, tensor input_ids) [FlexibleShapeInformation = tuple, dict, tensor>>, tuple, dict, list, ?>>>>((("DefaultShapes", {{"attention_mask", [1, 1]}, {"input_ids", [1, 1]}}), ("RangeDims", {{"attention_mask", [[1, 1], [1, 1024]]}, {"input_ids", [[1, 1], [1, 1024]]}})))] { + tensor encoder_embed_tokens_weight = const()[name = tensor("encoder_embed_tokens_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor encoder_embed_positions_weights = const()[name = tensor("encoder_embed_positions_weights"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1049419904)))]; + tensor encoder_layers_0_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_0_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1053622464)))]; + tensor encoder_layers_0_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_0_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1053626624)))]; + tensor encoder_layers_0_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_0_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1053630784)))]; + tensor encoder_layers_0_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_0_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1053634944)))]; + tensor encoder_layers_0_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_0_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1057829312)))]; + tensor encoder_layers_0_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_0_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1057833472)))]; + tensor encoder_layers_0_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_0_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1062027840)))]; + tensor encoder_layers_0_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_0_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1062032000)))]; + tensor encoder_layers_0_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_0_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1066226368)))]; + tensor encoder_layers_0_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_0_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1066230528)))]; + tensor encoder_layers_0_final_layer_norm_bias = const()[name = tensor("encoder_layers_0_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070424896)))]; + tensor encoder_layers_0_final_layer_norm_weight = const()[name = tensor("encoder_layers_0_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070429056)))]; + tensor encoder_layers_0_fc1_bias = const()[name = tensor("encoder_layers_0_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070433216)))]; + tensor encoder_layers_0_fc1_weight = const()[name = tensor("encoder_layers_0_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070449664)))]; + tensor encoder_layers_0_fc2_bias = const()[name = tensor("encoder_layers_0_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087226944)))]; + tensor encoder_layers_0_fc2_weight = const()[name = tensor("encoder_layers_0_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087231104)))]; + tensor encoder_layers_1_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_1_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104008384)))]; + tensor encoder_layers_1_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_1_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104012544)))]; + tensor encoder_layers_1_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_1_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104016704)))]; + tensor encoder_layers_1_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_1_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1104020864)))]; + tensor encoder_layers_1_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_1_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1108215232)))]; + tensor encoder_layers_1_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_1_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1108219392)))]; + tensor encoder_layers_1_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_1_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1112413760)))]; + tensor encoder_layers_1_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_1_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1112417920)))]; + tensor encoder_layers_1_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_1_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116612288)))]; + tensor encoder_layers_1_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_1_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116616448)))]; + tensor encoder_layers_1_final_layer_norm_bias = const()[name = tensor("encoder_layers_1_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1120810816)))]; + tensor encoder_layers_1_final_layer_norm_weight = const()[name = tensor("encoder_layers_1_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1120814976)))]; + tensor encoder_layers_1_fc1_bias = const()[name = tensor("encoder_layers_1_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1120819136)))]; + tensor encoder_layers_1_fc1_weight = const()[name = tensor("encoder_layers_1_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1120835584)))]; + tensor encoder_layers_1_fc2_bias = const()[name = tensor("encoder_layers_1_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1137612864)))]; + tensor encoder_layers_1_fc2_weight = const()[name = tensor("encoder_layers_1_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1137617024)))]; + tensor encoder_layers_2_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_2_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1154394304)))]; + tensor encoder_layers_2_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_2_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1154398464)))]; + tensor encoder_layers_2_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_2_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1154402624)))]; + tensor encoder_layers_2_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_2_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1154406784)))]; + tensor encoder_layers_2_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_2_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158601152)))]; + tensor encoder_layers_2_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_2_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158605312)))]; + tensor encoder_layers_2_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_2_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162799680)))]; + tensor encoder_layers_2_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_2_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162803840)))]; + tensor encoder_layers_2_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_2_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1166998208)))]; + tensor encoder_layers_2_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_2_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1167002368)))]; + tensor encoder_layers_2_final_layer_norm_bias = const()[name = tensor("encoder_layers_2_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1171196736)))]; + tensor encoder_layers_2_final_layer_norm_weight = const()[name = tensor("encoder_layers_2_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1171200896)))]; + tensor encoder_layers_2_fc1_bias = const()[name = tensor("encoder_layers_2_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1171205056)))]; + tensor encoder_layers_2_fc1_weight = const()[name = tensor("encoder_layers_2_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1171221504)))]; + tensor encoder_layers_2_fc2_bias = const()[name = tensor("encoder_layers_2_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1187998784)))]; + tensor encoder_layers_2_fc2_weight = const()[name = tensor("encoder_layers_2_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1188002944)))]; + tensor encoder_layers_3_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_3_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204780224)))]; + tensor encoder_layers_3_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_3_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204784384)))]; + tensor encoder_layers_3_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_3_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204788544)))]; + tensor encoder_layers_3_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_3_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204792704)))]; + tensor encoder_layers_3_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_3_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208987072)))]; + tensor encoder_layers_3_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_3_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208991232)))]; + tensor encoder_layers_3_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_3_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1213185600)))]; + tensor encoder_layers_3_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_3_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1213189760)))]; + tensor encoder_layers_3_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_3_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1217384128)))]; + tensor encoder_layers_3_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_3_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1217388288)))]; + tensor encoder_layers_3_final_layer_norm_bias = const()[name = tensor("encoder_layers_3_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221582656)))]; + tensor encoder_layers_3_final_layer_norm_weight = const()[name = tensor("encoder_layers_3_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221586816)))]; + tensor encoder_layers_3_fc1_bias = const()[name = tensor("encoder_layers_3_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221590976)))]; + tensor encoder_layers_3_fc1_weight = const()[name = tensor("encoder_layers_3_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221607424)))]; + tensor encoder_layers_3_fc2_bias = const()[name = tensor("encoder_layers_3_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1238384704)))]; + tensor encoder_layers_3_fc2_weight = const()[name = tensor("encoder_layers_3_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1238388864)))]; + tensor encoder_layers_4_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_4_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1255166144)))]; + tensor encoder_layers_4_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_4_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1255170304)))]; + tensor encoder_layers_4_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_4_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1255174464)))]; + tensor encoder_layers_4_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_4_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1255178624)))]; + tensor encoder_layers_4_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_4_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1259372992)))]; + tensor encoder_layers_4_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_4_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1259377152)))]; + tensor encoder_layers_4_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_4_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1263571520)))]; + tensor encoder_layers_4_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_4_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1263575680)))]; + tensor encoder_layers_4_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_4_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1267770048)))]; + tensor encoder_layers_4_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_4_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1267774208)))]; + tensor encoder_layers_4_final_layer_norm_bias = const()[name = tensor("encoder_layers_4_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1271968576)))]; + tensor encoder_layers_4_final_layer_norm_weight = const()[name = tensor("encoder_layers_4_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1271972736)))]; + tensor encoder_layers_4_fc1_bias = const()[name = tensor("encoder_layers_4_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1271976896)))]; + tensor encoder_layers_4_fc1_weight = const()[name = tensor("encoder_layers_4_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1271993344)))]; + tensor encoder_layers_4_fc2_bias = const()[name = tensor("encoder_layers_4_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288770624)))]; + tensor encoder_layers_4_fc2_weight = const()[name = tensor("encoder_layers_4_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1288774784)))]; + tensor encoder_layers_5_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_5_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1305552064)))]; + tensor encoder_layers_5_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_5_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1305556224)))]; + tensor encoder_layers_5_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_5_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1305560384)))]; + tensor encoder_layers_5_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_5_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1305564544)))]; + tensor encoder_layers_5_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_5_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309758912)))]; + tensor encoder_layers_5_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_5_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1309763072)))]; + tensor encoder_layers_5_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_5_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1313957440)))]; + tensor encoder_layers_5_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_5_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1313961600)))]; + tensor encoder_layers_5_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_5_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1318155968)))]; + tensor encoder_layers_5_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_5_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1318160128)))]; + tensor encoder_layers_5_final_layer_norm_bias = const()[name = tensor("encoder_layers_5_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322354496)))]; + tensor encoder_layers_5_final_layer_norm_weight = const()[name = tensor("encoder_layers_5_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322358656)))]; + tensor encoder_layers_5_fc1_bias = const()[name = tensor("encoder_layers_5_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322362816)))]; + tensor encoder_layers_5_fc1_weight = const()[name = tensor("encoder_layers_5_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1322379264)))]; + tensor encoder_layers_5_fc2_bias = const()[name = tensor("encoder_layers_5_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339156544)))]; + tensor encoder_layers_5_fc2_weight = const()[name = tensor("encoder_layers_5_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1339160704)))]; + tensor encoder_layers_6_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_6_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1355937984)))]; + tensor encoder_layers_6_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_6_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1355942144)))]; + tensor encoder_layers_6_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_6_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1355946304)))]; + tensor encoder_layers_6_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_6_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1355950464)))]; + tensor encoder_layers_6_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_6_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1360144832)))]; + tensor encoder_layers_6_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_6_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1360148992)))]; + tensor encoder_layers_6_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_6_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1364343360)))]; + tensor encoder_layers_6_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_6_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1364347520)))]; + tensor encoder_layers_6_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_6_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1368541888)))]; + tensor encoder_layers_6_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_6_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1368546048)))]; + tensor encoder_layers_6_final_layer_norm_bias = const()[name = tensor("encoder_layers_6_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1372740416)))]; + tensor encoder_layers_6_final_layer_norm_weight = const()[name = tensor("encoder_layers_6_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1372744576)))]; + tensor encoder_layers_6_fc1_bias = const()[name = tensor("encoder_layers_6_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1372748736)))]; + tensor encoder_layers_6_fc1_weight = const()[name = tensor("encoder_layers_6_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1372765184)))]; + tensor encoder_layers_6_fc2_bias = const()[name = tensor("encoder_layers_6_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1389542464)))]; + tensor encoder_layers_6_fc2_weight = const()[name = tensor("encoder_layers_6_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1389546624)))]; + tensor encoder_layers_7_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_7_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406323904)))]; + tensor encoder_layers_7_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_7_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406328064)))]; + tensor encoder_layers_7_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_7_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406332224)))]; + tensor encoder_layers_7_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_7_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1406336384)))]; + tensor encoder_layers_7_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_7_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1410530752)))]; + tensor encoder_layers_7_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_7_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1410534912)))]; + tensor encoder_layers_7_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_7_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1414729280)))]; + tensor encoder_layers_7_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_7_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1414733440)))]; + tensor encoder_layers_7_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_7_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1418927808)))]; + tensor encoder_layers_7_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_7_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1418931968)))]; + tensor encoder_layers_7_final_layer_norm_bias = const()[name = tensor("encoder_layers_7_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1423126336)))]; + tensor encoder_layers_7_final_layer_norm_weight = const()[name = tensor("encoder_layers_7_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1423130496)))]; + tensor encoder_layers_7_fc1_bias = const()[name = tensor("encoder_layers_7_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1423134656)))]; + tensor encoder_layers_7_fc1_weight = const()[name = tensor("encoder_layers_7_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1423151104)))]; + tensor encoder_layers_7_fc2_bias = const()[name = tensor("encoder_layers_7_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1439928384)))]; + tensor encoder_layers_7_fc2_weight = const()[name = tensor("encoder_layers_7_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1439932544)))]; + tensor encoder_layers_8_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_8_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1456709824)))]; + tensor encoder_layers_8_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_8_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1456713984)))]; + tensor encoder_layers_8_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_8_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1456718144)))]; + tensor encoder_layers_8_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_8_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1456722304)))]; + tensor encoder_layers_8_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_8_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1460916672)))]; + tensor encoder_layers_8_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_8_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1460920832)))]; + tensor encoder_layers_8_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_8_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1465115200)))]; + tensor encoder_layers_8_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_8_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1465119360)))]; + tensor encoder_layers_8_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_8_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1469313728)))]; + tensor encoder_layers_8_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_8_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1469317888)))]; + tensor encoder_layers_8_final_layer_norm_bias = const()[name = tensor("encoder_layers_8_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1473512256)))]; + tensor encoder_layers_8_final_layer_norm_weight = const()[name = tensor("encoder_layers_8_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1473516416)))]; + tensor encoder_layers_8_fc1_bias = const()[name = tensor("encoder_layers_8_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1473520576)))]; + tensor encoder_layers_8_fc1_weight = const()[name = tensor("encoder_layers_8_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1473537024)))]; + tensor encoder_layers_8_fc2_bias = const()[name = tensor("encoder_layers_8_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1490314304)))]; + tensor encoder_layers_8_fc2_weight = const()[name = tensor("encoder_layers_8_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1490318464)))]; + tensor encoder_layers_9_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_9_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1507095744)))]; + tensor encoder_layers_9_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_9_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1507099904)))]; + tensor encoder_layers_9_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_9_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1507104064)))]; + tensor encoder_layers_9_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_9_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1507108224)))]; + tensor encoder_layers_9_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_9_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511302592)))]; + tensor encoder_layers_9_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_9_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1511306752)))]; + tensor encoder_layers_9_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_9_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1515501120)))]; + tensor encoder_layers_9_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_9_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1515505280)))]; + tensor encoder_layers_9_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_9_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1519699648)))]; + tensor encoder_layers_9_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_9_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1519703808)))]; + tensor encoder_layers_9_final_layer_norm_bias = const()[name = tensor("encoder_layers_9_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1523898176)))]; + tensor encoder_layers_9_final_layer_norm_weight = const()[name = tensor("encoder_layers_9_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1523902336)))]; + tensor encoder_layers_9_fc1_bias = const()[name = tensor("encoder_layers_9_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1523906496)))]; + tensor encoder_layers_9_fc1_weight = const()[name = tensor("encoder_layers_9_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1523922944)))]; + tensor encoder_layers_9_fc2_bias = const()[name = tensor("encoder_layers_9_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1540700224)))]; + tensor encoder_layers_9_fc2_weight = const()[name = tensor("encoder_layers_9_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1540704384)))]; + tensor encoder_layers_10_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_10_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1557481664)))]; + tensor encoder_layers_10_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_10_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1557485824)))]; + tensor encoder_layers_10_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_10_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1557489984)))]; + tensor encoder_layers_10_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_10_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1557494144)))]; + tensor encoder_layers_10_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_10_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1561688512)))]; + tensor encoder_layers_10_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_10_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1561692672)))]; + tensor encoder_layers_10_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_10_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1565887040)))]; + tensor encoder_layers_10_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_10_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1565891200)))]; + tensor encoder_layers_10_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_10_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1570085568)))]; + tensor encoder_layers_10_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_10_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1570089728)))]; + tensor encoder_layers_10_final_layer_norm_bias = const()[name = tensor("encoder_layers_10_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574284096)))]; + tensor encoder_layers_10_final_layer_norm_weight = const()[name = tensor("encoder_layers_10_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574288256)))]; + tensor encoder_layers_10_fc1_bias = const()[name = tensor("encoder_layers_10_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574292416)))]; + tensor encoder_layers_10_fc1_weight = const()[name = tensor("encoder_layers_10_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1574308864)))]; + tensor encoder_layers_10_fc2_bias = const()[name = tensor("encoder_layers_10_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1591086144)))]; + tensor encoder_layers_10_fc2_weight = const()[name = tensor("encoder_layers_10_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1591090304)))]; + tensor encoder_layers_11_self_attn_layer_norm_bias = const()[name = tensor("encoder_layers_11_self_attn_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1607867584)))]; + tensor encoder_layers_11_self_attn_layer_norm_weight = const()[name = tensor("encoder_layers_11_self_attn_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1607871744)))]; + tensor encoder_layers_11_self_attn_q_proj_bias = const()[name = tensor("encoder_layers_11_self_attn_q_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1607875904)))]; + tensor encoder_layers_11_self_attn_q_proj_weight = const()[name = tensor("encoder_layers_11_self_attn_q_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1607880064)))]; + tensor encoder_layers_11_self_attn_k_proj_bias = const()[name = tensor("encoder_layers_11_self_attn_k_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1612074432)))]; + tensor encoder_layers_11_self_attn_k_proj_weight = const()[name = tensor("encoder_layers_11_self_attn_k_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1612078592)))]; + tensor encoder_layers_11_self_attn_v_proj_bias = const()[name = tensor("encoder_layers_11_self_attn_v_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616272960)))]; + tensor encoder_layers_11_self_attn_v_proj_weight = const()[name = tensor("encoder_layers_11_self_attn_v_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1616277120)))]; + tensor encoder_layers_11_self_attn_out_proj_bias = const()[name = tensor("encoder_layers_11_self_attn_out_proj_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1620471488)))]; + tensor encoder_layers_11_self_attn_out_proj_weight = const()[name = tensor("encoder_layers_11_self_attn_out_proj_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1620475648)))]; + tensor encoder_layers_11_final_layer_norm_bias = const()[name = tensor("encoder_layers_11_final_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1624670016)))]; + tensor encoder_layers_11_final_layer_norm_weight = const()[name = tensor("encoder_layers_11_final_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1624674176)))]; + tensor encoder_layers_11_fc1_bias = const()[name = tensor("encoder_layers_11_fc1_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1624678336)))]; + tensor encoder_layers_11_fc1_weight = const()[name = tensor("encoder_layers_11_fc1_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1624694784)))]; + tensor encoder_layers_11_fc2_bias = const()[name = tensor("encoder_layers_11_fc2_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1641472064)))]; + tensor encoder_layers_11_fc2_weight = const()[name = tensor("encoder_layers_11_fc2_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1641476224)))]; + tensor encoder_layer_norm_bias = const()[name = tensor("encoder_layer_norm_bias"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1658253504)))]; + tensor encoder_layer_norm_weight = const()[name = tensor("encoder_layer_norm_weight"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1658257664)))]; + tensor var_5 = const()[name = tensor("op_5"), val = tensor(0x1.4f8b58p-17)]; + tensor var_7 = const()[name = tensor("op_7"), val = tensor(0x1p-3)]; + tensor var_10 = const()[name = tensor("op_10"), val = tensor(-0x1.fffffep+127)]; + tensor var_17 = const()[name = tensor("op_17"), val = tensor(0)]; + tensor var_23 = const()[name = tensor("op_23"), val = tensor(-1)]; + tensor var_24 = const()[name = tensor("op_24"), val = tensor(1)]; + tensor var_56_axis_0 = const()[name = tensor("op_56_axis_0"), val = tensor(0)]; + tensor var_56_batch_dims_0 = const()[name = tensor("op_56_batch_dims_0"), val = tensor(0)]; + tensor 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("op_56")]; + tensor var_57 = const()[name = tensor("op_57"), val = tensor(0x1p+5)]; + tensor inputs_embeds = mul(x = var_56, y = var_57)[name = tensor("inputs_embeds")]; + tensor var_62 = not_equal(x = input_ids, y = var_24)[name = tensor("op_62")]; + tensor mask_dtype_0 = const()[name = tensor("mask_dtype_0"), val = tensor("int32")]; + tensor var_64_exclusive_0 = const()[name = tensor("op_64_exclusive_0"), val = tensor(false)]; + tensor var_64_reverse_0 = const()[name = tensor("op_64_reverse_0"), val = tensor(false)]; + tensor mask = cast(dtype = mask_dtype_0, x = var_62)[name = tensor("cast_55")]; + tensor var_64 = cumsum(axis = var_24, exclusive = var_64_exclusive_0, reverse = var_64_reverse_0, x = mask)[name = tensor("op_64")]; + tensor incremental_indices = mul(x = var_64, y = mask)[name = tensor("incremental_indices")]; + tensor var_70 = const()[name = tensor("op_70"), val = tensor(1)]; + tensor var_71 = add(x = incremental_indices, y = var_70)[name = tensor("op_71")]; + tensor var_73 = const()[name = tensor("op_73"), val = tensor([-1])]; + tensor var_74 = reshape(shape = var_73, x = var_71)[name = tensor("op_74")]; + tensor var_75_batch_dims_0 = const()[name = tensor("op_75_batch_dims_0"), val = tensor(0)]; + tensor var_75 = gather(axis = var_17, batch_dims = var_75_batch_dims_0, indices = var_74, x = encoder_embed_positions_weights)[name = tensor("op_75")]; + tensor concat_1x = const()[name = tensor("concat_1x"), val = tensor([1, -1, 1024])]; + tensor var_78 = reshape(shape = concat_1x, x = var_75)[name = tensor("op_78")]; + tensor input_3 = add(x = inputs_embeds, y = var_78)[name = tensor("input_3")]; + tensor var_83_shape = shape(x = attention_mask)[name = tensor("op_83_shape")]; + tensor gather_3_indices_0 = const()[name = tensor("gather_3_indices_0"), val = tensor(1)]; + tensor gather_3_axis_0 = const()[name = tensor("gather_3_axis_0"), val = tensor(0)]; + tensor gather_3_batch_dims_0 = const()[name = tensor("gather_3_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_3")]; + tensor gather_4 = const()[name = tensor("gather_4"), val = tensor(1)]; + tensor var_87_axes_0 = const()[name = tensor("op_87_axes_0"), val = tensor([1])]; + tensor var_87 = expand_dims(axes = var_87_axes_0, x = attention_mask)[name = tensor("op_87")]; + tensor var_88_axes_0 = const()[name = tensor("op_88_axes_0"), val = tensor([2])]; + tensor var_88 = expand_dims(axes = var_88_axes_0, x = var_87)[name = tensor("op_88")]; + tensor concat_2_axis_0 = const()[name = tensor("concat_2_axis_0"), val = tensor(0)]; + tensor concat_2_interleave_0 = const()[name = tensor("concat_2_interleave_0"), val = tensor(false)]; + tensor concat_2 = concat(axis = concat_2_axis_0, interleave = concat_2_interleave_0, values = (gather_4, var_24, gather_3, gather_3))[name = tensor("concat_2")]; + tensor shape_0 = shape(x = var_88)[name = tensor("shape_0")]; + tensor equal_0_y_0 = const()[name = tensor("equal_0_y_0"), val = tensor(-1)]; + tensor equal_0 = equal(x = concat_2, y = equal_0_y_0)[name = tensor("equal_0")]; + tensor select_0 = select(a = shape_0, b = concat_2, cond = equal_0)[name = tensor("select_0")]; + tensor real_div_0 = real_div(x = select_0, y = shape_0)[name = tensor("real_div_0")]; + tensor var_91 = tile(reps = real_div_0, x = var_88)[name = tensor("op_91")]; + tensor expanded_mask_dtype_0 = const()[name = tensor("expanded_mask_dtype_0"), val = tensor("fp32")]; + tensor const_1 = const()[name = tensor("const_1"), val = tensor(0x1p+0)]; + tensor expanded_mask = cast(dtype = expanded_mask_dtype_0, x = var_91)[name = tensor("cast_54")]; + tensor inverted_mask = sub(x = const_1, y = expanded_mask)[name = tensor("inverted_mask")]; + tensor var_96_dtype_0 = const()[name = tensor("op_96_dtype_0"), val = tensor("bool")]; + tensor var_96 = cast(dtype = var_96_dtype_0, x = inverted_mask)[name = tensor("cast_53")]; + tensor attention_mask_3 = select(a = var_10, b = inverted_mask, cond = var_96)[name = tensor("attention_mask_3")]; + tensor hidden_states_1_axes_0 = const()[name = tensor("hidden_states_1_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_1")]; + tensor var_111_shape = shape(x = hidden_states_1)[name = tensor("op_111_shape")]; + tensor gather_6 = const()[name = tensor("gather_6"), val = tensor(1)]; + tensor gather_7_indices_0 = const()[name = tensor("gather_7_indices_0"), val = tensor(1)]; + tensor gather_7_axis_0 = const()[name = tensor("gather_7_axis_0"), val = tensor(0)]; + tensor gather_7_batch_dims_0 = const()[name = tensor("gather_7_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_7")]; + tensor 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("linear_0")]; + tensor concat_3x = const()[name = tensor("concat_3x"), val = tensor([1, -1, 16, 64])]; + tensor var_117 = reshape(shape = concat_3x, x = var_115)[name = tensor("op_117")]; + tensor query_1_perm_0 = const()[name = tensor("query_1_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_1")]; + tensor 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("linear_2")]; + tensor concat_4x = const()[name = tensor("concat_4x"), val = tensor([1, -1, 16, 64])]; + tensor var_126 = reshape(shape = concat_4x, x = key_states_1)[name = tensor("op_126")]; + tensor key_1_perm_0 = const()[name = tensor("key_1_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_5x = const()[name = tensor("concat_5x"), val = tensor([1, -1, 16, 64])]; + tensor var_129 = reshape(shape = concat_5x, x = value_states_1)[name = tensor("op_129")]; + tensor value_1_perm_0 = const()[name = tensor("value_1_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_1 = transpose(perm = key_1_perm_0, x = var_126)[name = tensor("transpose_46")]; + tensor var_131_shape = shape(x = key_1)[name = tensor("op_131_shape")]; + tensor gather_8_indices_0 = const()[name = tensor("gather_8_indices_0"), val = tensor(2)]; + tensor gather_8_axis_0 = const()[name = tensor("gather_8_axis_0"), val = tensor(0)]; + tensor gather_8_batch_dims_0 = const()[name = tensor("gather_8_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_8")]; + tensor concat_6_values0_0 = const()[name = tensor("concat_6_values0_0"), val = tensor(0)]; + tensor concat_6_values1_0 = const()[name = tensor("concat_6_values1_0"), val = tensor(0)]; + tensor concat_6_values2_0 = const()[name = tensor("concat_6_values2_0"), val = tensor(0)]; + tensor concat_6_axis_0 = const()[name = tensor("concat_6_axis_0"), val = tensor(0)]; + tensor concat_6_interleave_0 = const()[name = tensor("concat_6_interleave_0"), val = tensor(false)]; + tensor 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("concat_6")]; + tensor attention_mask_5_begin_0 = const()[name = tensor("attention_mask_5_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_5_end_mask_0 = const()[name = tensor("attention_mask_5_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_5")]; + tensor query_1 = transpose(perm = query_1_perm_0, x = var_117)[name = tensor("transpose_47")]; + tensor mul_0 = mul(x = query_1, y = var_7)[name = tensor("mul_0")]; + tensor matmul_0_transpose_y_0 = const()[name = tensor("matmul_0_transpose_y_0"), val = tensor(true)]; + tensor matmul_0_transpose_x_0 = const()[name = tensor("matmul_0_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_0")]; + tensor add_0 = add(x = matmul_0, y = attention_mask_5)[name = tensor("add_0")]; + tensor softmax_0_axis_0 = const()[name = tensor("softmax_0_axis_0"), val = tensor(-1)]; + tensor softmax_0 = softmax(axis = softmax_0_axis_0, x = add_0)[name = tensor("softmax_0")]; + tensor attn_output_1_transpose_x_0 = const()[name = tensor("attn_output_1_transpose_x_0"), val = tensor(false)]; + tensor attn_output_1_transpose_y_0 = const()[name = tensor("attn_output_1_transpose_y_0"), val = tensor(false)]; + tensor value_1 = transpose(perm = value_1_perm_0, x = var_129)[name = tensor("transpose_45")]; + tensor 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("attn_output_1")]; + tensor var_137_perm_0 = const()[name = tensor("op_137_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_7_axis_0 = const()[name = tensor("concat_7_axis_0"), val = tensor(0)]; + tensor concat_7_interleave_0 = const()[name = tensor("concat_7_interleave_0"), val = tensor(false)]; + tensor concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (gather_6, gather_7, var_23))[name = tensor("concat_7")]; + tensor var_137 = transpose(perm = var_137_perm_0, x = attn_output_1)[name = tensor("transpose_44")]; + tensor var_140 = reshape(shape = concat_7, x = var_137)[name = tensor("op_140")]; + tensor 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("linear_3")]; + tensor input_11 = add(x = input_3, y = input_9)[name = tensor("input_11")]; + tensor input_13_axes_0 = const()[name = tensor("input_13_axes_0"), val = tensor([-1])]; + tensor 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("input_13")]; + tensor input_15 = linear(bias = encoder_layers_0_fc1_bias, weight = encoder_layers_0_fc1_weight, x = input_13)[name = tensor("linear_4")]; + tensor input_17 = relu(x = input_15)[name = tensor("input_17")]; + tensor input_21 = linear(bias = encoder_layers_0_fc2_bias, weight = encoder_layers_0_fc2_weight, x = input_17)[name = tensor("linear_5")]; + tensor input_23 = add(x = input_11, y = input_21)[name = tensor("input_23")]; + tensor hidden_states_7_axes_0 = const()[name = tensor("hidden_states_7_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_7")]; + tensor var_174_shape = shape(x = hidden_states_7)[name = tensor("op_174_shape")]; + tensor gather_9 = const()[name = tensor("gather_9"), val = tensor(1)]; + tensor gather_10_indices_0 = const()[name = tensor("gather_10_indices_0"), val = tensor(1)]; + tensor gather_10_axis_0 = const()[name = tensor("gather_10_axis_0"), val = tensor(0)]; + tensor gather_10_batch_dims_0 = const()[name = tensor("gather_10_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_10")]; + tensor 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("linear_6")]; + tensor concat_8x = const()[name = tensor("concat_8x"), val = tensor([1, -1, 16, 64])]; + tensor var_180 = reshape(shape = concat_8x, x = var_178)[name = tensor("op_180")]; + tensor query_3_perm_0 = const()[name = tensor("query_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_7")]; + tensor 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("linear_8")]; + tensor concat_9x = const()[name = tensor("concat_9x"), val = tensor([1, -1, 16, 64])]; + tensor var_189 = reshape(shape = concat_9x, x = key_states_3)[name = tensor("op_189")]; + tensor key_3_perm_0 = const()[name = tensor("key_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_10x = const()[name = tensor("concat_10x"), val = tensor([1, -1, 16, 64])]; + tensor var_192 = reshape(shape = concat_10x, x = value_states_3)[name = tensor("op_192")]; + tensor value_3_perm_0 = const()[name = tensor("value_3_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_3 = transpose(perm = key_3_perm_0, x = var_189)[name = tensor("transpose_42")]; + tensor var_194_shape = shape(x = key_3)[name = tensor("op_194_shape")]; + tensor gather_11_indices_0 = const()[name = tensor("gather_11_indices_0"), val = tensor(2)]; + tensor gather_11_axis_0 = const()[name = tensor("gather_11_axis_0"), val = tensor(0)]; + tensor gather_11_batch_dims_0 = const()[name = tensor("gather_11_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_11")]; + tensor concat_11_values0_0 = const()[name = tensor("concat_11_values0_0"), val = tensor(0)]; + tensor concat_11_values1_0 = const()[name = tensor("concat_11_values1_0"), val = tensor(0)]; + tensor concat_11_values2_0 = const()[name = tensor("concat_11_values2_0"), val = tensor(0)]; + tensor concat_11_axis_0 = const()[name = tensor("concat_11_axis_0"), val = tensor(0)]; + tensor concat_11_interleave_0 = const()[name = tensor("concat_11_interleave_0"), val = tensor(false)]; + tensor 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("concat_11")]; + tensor attention_mask_7_begin_0 = const()[name = tensor("attention_mask_7_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_7_end_mask_0 = const()[name = tensor("attention_mask_7_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_7")]; + tensor query_3 = transpose(perm = query_3_perm_0, x = var_180)[name = tensor("transpose_43")]; + tensor mul_1 = mul(x = query_3, y = var_7)[name = tensor("mul_1")]; + tensor matmul_1_transpose_y_0 = const()[name = tensor("matmul_1_transpose_y_0"), val = tensor(true)]; + tensor matmul_1_transpose_x_0 = const()[name = tensor("matmul_1_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_1")]; + tensor add_1 = add(x = matmul_1, y = attention_mask_7)[name = tensor("add_1")]; + tensor softmax_1_axis_0 = const()[name = tensor("softmax_1_axis_0"), val = tensor(-1)]; + tensor softmax_1 = softmax(axis = softmax_1_axis_0, x = add_1)[name = tensor("softmax_1")]; + tensor attn_output_5_transpose_x_0 = const()[name = tensor("attn_output_5_transpose_x_0"), val = tensor(false)]; + tensor attn_output_5_transpose_y_0 = const()[name = tensor("attn_output_5_transpose_y_0"), val = tensor(false)]; + tensor value_3 = transpose(perm = value_3_perm_0, x = var_192)[name = tensor("transpose_41")]; + tensor 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("attn_output_5")]; + tensor var_200_perm_0 = const()[name = tensor("op_200_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_12_axis_0 = const()[name = tensor("concat_12_axis_0"), val = tensor(0)]; + tensor concat_12_interleave_0 = const()[name = tensor("concat_12_interleave_0"), val = tensor(false)]; + tensor concat_12 = concat(axis = concat_12_axis_0, interleave = concat_12_interleave_0, values = (gather_9, gather_10, var_23))[name = tensor("concat_12")]; + tensor var_200 = transpose(perm = var_200_perm_0, x = attn_output_5)[name = tensor("transpose_40")]; + tensor var_203 = reshape(shape = concat_12, x = var_200)[name = tensor("op_203")]; + tensor 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("linear_9")]; + tensor input_29 = add(x = input_23, y = input_27)[name = tensor("input_29")]; + tensor input_31_axes_0 = const()[name = tensor("input_31_axes_0"), val = tensor([-1])]; + tensor 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("input_31")]; + tensor input_33 = linear(bias = encoder_layers_1_fc1_bias, weight = encoder_layers_1_fc1_weight, x = input_31)[name = tensor("linear_10")]; + tensor input_35 = relu(x = input_33)[name = tensor("input_35")]; + tensor input_39 = linear(bias = encoder_layers_1_fc2_bias, weight = encoder_layers_1_fc2_weight, x = input_35)[name = tensor("linear_11")]; + tensor input_41 = add(x = input_29, y = input_39)[name = tensor("input_41")]; + tensor hidden_states_13_axes_0 = const()[name = tensor("hidden_states_13_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_13")]; + tensor var_237_shape = shape(x = hidden_states_13)[name = tensor("op_237_shape")]; + tensor gather_12 = const()[name = tensor("gather_12"), val = tensor(1)]; + tensor gather_13_indices_0 = const()[name = tensor("gather_13_indices_0"), val = tensor(1)]; + tensor gather_13_axis_0 = const()[name = tensor("gather_13_axis_0"), val = tensor(0)]; + tensor gather_13_batch_dims_0 = const()[name = tensor("gather_13_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_13")]; + tensor 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("linear_12")]; + tensor concat_13x = const()[name = tensor("concat_13x"), val = tensor([1, -1, 16, 64])]; + tensor var_243 = reshape(shape = concat_13x, x = var_241)[name = tensor("op_243")]; + tensor query_5_perm_0 = const()[name = tensor("query_5_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_13")]; + tensor 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("linear_14")]; + tensor concat_14x = const()[name = tensor("concat_14x"), val = tensor([1, -1, 16, 64])]; + tensor var_252 = reshape(shape = concat_14x, x = key_states_5)[name = tensor("op_252")]; + tensor key_5_perm_0 = const()[name = tensor("key_5_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_15x = const()[name = tensor("concat_15x"), val = tensor([1, -1, 16, 64])]; + tensor var_255 = reshape(shape = concat_15x, x = value_states_5)[name = tensor("op_255")]; + tensor value_5_perm_0 = const()[name = tensor("value_5_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_5 = transpose(perm = key_5_perm_0, x = var_252)[name = tensor("transpose_38")]; + tensor var_257_shape = shape(x = key_5)[name = tensor("op_257_shape")]; + tensor gather_14_indices_0 = const()[name = tensor("gather_14_indices_0"), val = tensor(2)]; + tensor gather_14_axis_0 = const()[name = tensor("gather_14_axis_0"), val = tensor(0)]; + tensor gather_14_batch_dims_0 = const()[name = tensor("gather_14_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_14")]; + tensor concat_16_values0_0 = const()[name = tensor("concat_16_values0_0"), val = tensor(0)]; + tensor concat_16_values1_0 = const()[name = tensor("concat_16_values1_0"), val = tensor(0)]; + tensor concat_16_values2_0 = const()[name = tensor("concat_16_values2_0"), val = tensor(0)]; + tensor concat_16_axis_0 = const()[name = tensor("concat_16_axis_0"), val = tensor(0)]; + tensor concat_16_interleave_0 = const()[name = tensor("concat_16_interleave_0"), val = tensor(false)]; + tensor 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("concat_16")]; + tensor attention_mask_9_begin_0 = const()[name = tensor("attention_mask_9_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_9_end_mask_0 = const()[name = tensor("attention_mask_9_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_9")]; + tensor query_5 = transpose(perm = query_5_perm_0, x = var_243)[name = tensor("transpose_39")]; + tensor mul_2 = mul(x = query_5, y = var_7)[name = tensor("mul_2")]; + tensor matmul_2_transpose_y_0 = const()[name = tensor("matmul_2_transpose_y_0"), val = tensor(true)]; + tensor matmul_2_transpose_x_0 = const()[name = tensor("matmul_2_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_2")]; + tensor add_2 = add(x = matmul_2, y = attention_mask_9)[name = tensor("add_2")]; + tensor softmax_2_axis_0 = const()[name = tensor("softmax_2_axis_0"), val = tensor(-1)]; + tensor softmax_2 = softmax(axis = softmax_2_axis_0, x = add_2)[name = tensor("softmax_2")]; + tensor attn_output_9_transpose_x_0 = const()[name = tensor("attn_output_9_transpose_x_0"), val = tensor(false)]; + tensor attn_output_9_transpose_y_0 = const()[name = tensor("attn_output_9_transpose_y_0"), val = tensor(false)]; + tensor value_5 = transpose(perm = value_5_perm_0, x = var_255)[name = tensor("transpose_37")]; + tensor 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("attn_output_9")]; + tensor var_263_perm_0 = const()[name = tensor("op_263_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_17_axis_0 = const()[name = tensor("concat_17_axis_0"), val = tensor(0)]; + tensor concat_17_interleave_0 = const()[name = tensor("concat_17_interleave_0"), val = tensor(false)]; + tensor concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (gather_12, gather_13, var_23))[name = tensor("concat_17")]; + tensor var_263 = transpose(perm = var_263_perm_0, x = attn_output_9)[name = tensor("transpose_36")]; + tensor var_266 = reshape(shape = concat_17, x = var_263)[name = tensor("op_266")]; + tensor 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("linear_15")]; + tensor input_47 = add(x = input_41, y = input_45)[name = tensor("input_47")]; + tensor input_49_axes_0 = const()[name = tensor("input_49_axes_0"), val = tensor([-1])]; + tensor 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("input_49")]; + tensor input_51 = linear(bias = encoder_layers_2_fc1_bias, weight = encoder_layers_2_fc1_weight, x = input_49)[name = tensor("linear_16")]; + tensor input_53 = relu(x = input_51)[name = tensor("input_53")]; + tensor input_57 = linear(bias = encoder_layers_2_fc2_bias, weight = encoder_layers_2_fc2_weight, x = input_53)[name = tensor("linear_17")]; + tensor input_59 = add(x = input_47, y = input_57)[name = tensor("input_59")]; + tensor hidden_states_19_axes_0 = const()[name = tensor("hidden_states_19_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_19")]; + tensor var_300_shape = shape(x = hidden_states_19)[name = tensor("op_300_shape")]; + tensor gather_15 = const()[name = tensor("gather_15"), val = tensor(1)]; + tensor gather_16_indices_0 = const()[name = tensor("gather_16_indices_0"), val = tensor(1)]; + tensor gather_16_axis_0 = const()[name = tensor("gather_16_axis_0"), val = tensor(0)]; + tensor gather_16_batch_dims_0 = const()[name = tensor("gather_16_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_16")]; + tensor 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("linear_18")]; + tensor concat_18x = const()[name = tensor("concat_18x"), val = tensor([1, -1, 16, 64])]; + tensor var_306 = reshape(shape = concat_18x, x = var_304)[name = tensor("op_306")]; + tensor query_7_perm_0 = const()[name = tensor("query_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_19")]; + tensor 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("linear_20")]; + tensor concat_19x = const()[name = tensor("concat_19x"), val = tensor([1, -1, 16, 64])]; + tensor var_315 = reshape(shape = concat_19x, x = key_states_7)[name = tensor("op_315")]; + tensor key_7_perm_0 = const()[name = tensor("key_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_20x = const()[name = tensor("concat_20x"), val = tensor([1, -1, 16, 64])]; + tensor var_318 = reshape(shape = concat_20x, x = value_states_7)[name = tensor("op_318")]; + tensor value_7_perm_0 = const()[name = tensor("value_7_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_7 = transpose(perm = key_7_perm_0, x = var_315)[name = tensor("transpose_34")]; + tensor var_320_shape = shape(x = key_7)[name = tensor("op_320_shape")]; + tensor gather_17_indices_0 = const()[name = tensor("gather_17_indices_0"), val = tensor(2)]; + tensor gather_17_axis_0 = const()[name = tensor("gather_17_axis_0"), val = tensor(0)]; + tensor gather_17_batch_dims_0 = const()[name = tensor("gather_17_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_17")]; + tensor concat_21_values0_0 = const()[name = tensor("concat_21_values0_0"), val = tensor(0)]; + tensor concat_21_values1_0 = const()[name = tensor("concat_21_values1_0"), val = tensor(0)]; + tensor concat_21_values2_0 = const()[name = tensor("concat_21_values2_0"), val = tensor(0)]; + tensor concat_21_axis_0 = const()[name = tensor("concat_21_axis_0"), val = tensor(0)]; + tensor concat_21_interleave_0 = const()[name = tensor("concat_21_interleave_0"), val = tensor(false)]; + tensor 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("concat_21")]; + tensor attention_mask_11_begin_0 = const()[name = tensor("attention_mask_11_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_11_end_mask_0 = const()[name = tensor("attention_mask_11_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_11")]; + tensor query_7 = transpose(perm = query_7_perm_0, x = var_306)[name = tensor("transpose_35")]; + tensor mul_3 = mul(x = query_7, y = var_7)[name = tensor("mul_3")]; + tensor matmul_3_transpose_y_0 = const()[name = tensor("matmul_3_transpose_y_0"), val = tensor(true)]; + tensor matmul_3_transpose_x_0 = const()[name = tensor("matmul_3_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_3")]; + tensor add_3 = add(x = matmul_3, y = attention_mask_11)[name = tensor("add_3")]; + tensor softmax_3_axis_0 = const()[name = tensor("softmax_3_axis_0"), val = tensor(-1)]; + tensor softmax_3 = softmax(axis = softmax_3_axis_0, x = add_3)[name = tensor("softmax_3")]; + tensor attn_output_13_transpose_x_0 = const()[name = tensor("attn_output_13_transpose_x_0"), val = tensor(false)]; + tensor attn_output_13_transpose_y_0 = const()[name = tensor("attn_output_13_transpose_y_0"), val = tensor(false)]; + tensor value_7 = transpose(perm = value_7_perm_0, x = var_318)[name = tensor("transpose_33")]; + tensor 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("attn_output_13")]; + tensor var_326_perm_0 = const()[name = tensor("op_326_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_22_axis_0 = const()[name = tensor("concat_22_axis_0"), val = tensor(0)]; + tensor concat_22_interleave_0 = const()[name = tensor("concat_22_interleave_0"), val = tensor(false)]; + tensor concat_22 = concat(axis = concat_22_axis_0, interleave = concat_22_interleave_0, values = (gather_15, gather_16, var_23))[name = tensor("concat_22")]; + tensor var_326 = transpose(perm = var_326_perm_0, x = attn_output_13)[name = tensor("transpose_32")]; + tensor var_329 = reshape(shape = concat_22, x = var_326)[name = tensor("op_329")]; + tensor 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("linear_21")]; + tensor input_65 = add(x = input_59, y = input_63)[name = tensor("input_65")]; + tensor input_67_axes_0 = const()[name = tensor("input_67_axes_0"), val = tensor([-1])]; + tensor 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("input_67")]; + tensor input_69 = linear(bias = encoder_layers_3_fc1_bias, weight = encoder_layers_3_fc1_weight, x = input_67)[name = tensor("linear_22")]; + tensor input_71 = relu(x = input_69)[name = tensor("input_71")]; + tensor input_75 = linear(bias = encoder_layers_3_fc2_bias, weight = encoder_layers_3_fc2_weight, x = input_71)[name = tensor("linear_23")]; + tensor input_77 = add(x = input_65, y = input_75)[name = tensor("input_77")]; + tensor hidden_states_25_axes_0 = const()[name = tensor("hidden_states_25_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_25")]; + tensor var_363_shape = shape(x = hidden_states_25)[name = tensor("op_363_shape")]; + tensor gather_18 = const()[name = tensor("gather_18"), val = tensor(1)]; + tensor gather_19_indices_0 = const()[name = tensor("gather_19_indices_0"), val = tensor(1)]; + tensor gather_19_axis_0 = const()[name = tensor("gather_19_axis_0"), val = tensor(0)]; + tensor gather_19_batch_dims_0 = const()[name = tensor("gather_19_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_19")]; + tensor 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("linear_24")]; + tensor concat_23x = const()[name = tensor("concat_23x"), val = tensor([1, -1, 16, 64])]; + tensor var_369 = reshape(shape = concat_23x, x = var_367)[name = tensor("op_369")]; + tensor query_9_perm_0 = const()[name = tensor("query_9_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_25")]; + tensor 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("linear_26")]; + tensor concat_24x = const()[name = tensor("concat_24x"), val = tensor([1, -1, 16, 64])]; + tensor var_378 = reshape(shape = concat_24x, x = key_states_9)[name = tensor("op_378")]; + tensor key_9_perm_0 = const()[name = tensor("key_9_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_25x = const()[name = tensor("concat_25x"), val = tensor([1, -1, 16, 64])]; + tensor var_381 = reshape(shape = concat_25x, x = value_states_9)[name = tensor("op_381")]; + tensor value_9_perm_0 = const()[name = tensor("value_9_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_9 = transpose(perm = key_9_perm_0, x = var_378)[name = tensor("transpose_30")]; + tensor var_383_shape = shape(x = key_9)[name = tensor("op_383_shape")]; + tensor gather_20_indices_0 = const()[name = tensor("gather_20_indices_0"), val = tensor(2)]; + tensor gather_20_axis_0 = const()[name = tensor("gather_20_axis_0"), val = tensor(0)]; + tensor gather_20_batch_dims_0 = const()[name = tensor("gather_20_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_20")]; + tensor concat_26_values0_0 = const()[name = tensor("concat_26_values0_0"), val = tensor(0)]; + tensor concat_26_values1_0 = const()[name = tensor("concat_26_values1_0"), val = tensor(0)]; + tensor concat_26_values2_0 = const()[name = tensor("concat_26_values2_0"), val = tensor(0)]; + tensor concat_26_axis_0 = const()[name = tensor("concat_26_axis_0"), val = tensor(0)]; + tensor concat_26_interleave_0 = const()[name = tensor("concat_26_interleave_0"), val = tensor(false)]; + tensor 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("concat_26")]; + tensor attention_mask_13_begin_0 = const()[name = tensor("attention_mask_13_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_13_end_mask_0 = const()[name = tensor("attention_mask_13_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_13")]; + tensor query_9 = transpose(perm = query_9_perm_0, x = var_369)[name = tensor("transpose_31")]; + tensor mul_4 = mul(x = query_9, y = var_7)[name = tensor("mul_4")]; + tensor matmul_4_transpose_y_0 = const()[name = tensor("matmul_4_transpose_y_0"), val = tensor(true)]; + tensor matmul_4_transpose_x_0 = const()[name = tensor("matmul_4_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_4")]; + tensor add_4 = add(x = matmul_4, y = attention_mask_13)[name = tensor("add_4")]; + tensor softmax_4_axis_0 = const()[name = tensor("softmax_4_axis_0"), val = tensor(-1)]; + tensor softmax_4 = softmax(axis = softmax_4_axis_0, x = add_4)[name = tensor("softmax_4")]; + tensor attn_output_17_transpose_x_0 = const()[name = tensor("attn_output_17_transpose_x_0"), val = tensor(false)]; + tensor attn_output_17_transpose_y_0 = const()[name = tensor("attn_output_17_transpose_y_0"), val = tensor(false)]; + tensor value_9 = transpose(perm = value_9_perm_0, x = var_381)[name = tensor("transpose_29")]; + tensor 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("attn_output_17")]; + tensor var_389_perm_0 = const()[name = tensor("op_389_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_27_axis_0 = const()[name = tensor("concat_27_axis_0"), val = tensor(0)]; + tensor concat_27_interleave_0 = const()[name = tensor("concat_27_interleave_0"), val = tensor(false)]; + tensor concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (gather_18, gather_19, var_23))[name = tensor("concat_27")]; + tensor var_389 = transpose(perm = var_389_perm_0, x = attn_output_17)[name = tensor("transpose_28")]; + tensor var_392 = reshape(shape = concat_27, x = var_389)[name = tensor("op_392")]; + tensor 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("linear_27")]; + tensor input_83 = add(x = input_77, y = input_81)[name = tensor("input_83")]; + tensor input_85_axes_0 = const()[name = tensor("input_85_axes_0"), val = tensor([-1])]; + tensor 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("input_85")]; + tensor input_87 = linear(bias = encoder_layers_4_fc1_bias, weight = encoder_layers_4_fc1_weight, x = input_85)[name = tensor("linear_28")]; + tensor input_89 = relu(x = input_87)[name = tensor("input_89")]; + tensor input_93 = linear(bias = encoder_layers_4_fc2_bias, weight = encoder_layers_4_fc2_weight, x = input_89)[name = tensor("linear_29")]; + tensor input_95 = add(x = input_83, y = input_93)[name = tensor("input_95")]; + tensor hidden_states_31_axes_0 = const()[name = tensor("hidden_states_31_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_31")]; + tensor var_426_shape = shape(x = hidden_states_31)[name = tensor("op_426_shape")]; + tensor gather_21 = const()[name = tensor("gather_21"), val = tensor(1)]; + tensor gather_22_indices_0 = const()[name = tensor("gather_22_indices_0"), val = tensor(1)]; + tensor gather_22_axis_0 = const()[name = tensor("gather_22_axis_0"), val = tensor(0)]; + tensor gather_22_batch_dims_0 = const()[name = tensor("gather_22_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_22")]; + tensor 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("linear_30")]; + tensor concat_28x = const()[name = tensor("concat_28x"), val = tensor([1, -1, 16, 64])]; + tensor var_432 = reshape(shape = concat_28x, x = var_430)[name = tensor("op_432")]; + tensor query_11_perm_0 = const()[name = tensor("query_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_31")]; + tensor 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("linear_32")]; + tensor concat_29x = const()[name = tensor("concat_29x"), val = tensor([1, -1, 16, 64])]; + tensor var_441 = reshape(shape = concat_29x, x = key_states_11)[name = tensor("op_441")]; + tensor key_11_perm_0 = const()[name = tensor("key_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_30x = const()[name = tensor("concat_30x"), val = tensor([1, -1, 16, 64])]; + tensor var_444 = reshape(shape = concat_30x, x = value_states_11)[name = tensor("op_444")]; + tensor value_11_perm_0 = const()[name = tensor("value_11_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_11 = transpose(perm = key_11_perm_0, x = var_441)[name = tensor("transpose_26")]; + tensor var_446_shape = shape(x = key_11)[name = tensor("op_446_shape")]; + tensor gather_23_indices_0 = const()[name = tensor("gather_23_indices_0"), val = tensor(2)]; + tensor gather_23_axis_0 = const()[name = tensor("gather_23_axis_0"), val = tensor(0)]; + tensor gather_23_batch_dims_0 = const()[name = tensor("gather_23_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_23")]; + tensor concat_31_values0_0 = const()[name = tensor("concat_31_values0_0"), val = tensor(0)]; + tensor concat_31_values1_0 = const()[name = tensor("concat_31_values1_0"), val = tensor(0)]; + tensor concat_31_values2_0 = const()[name = tensor("concat_31_values2_0"), val = tensor(0)]; + tensor concat_31_axis_0 = const()[name = tensor("concat_31_axis_0"), val = tensor(0)]; + tensor concat_31_interleave_0 = const()[name = tensor("concat_31_interleave_0"), val = tensor(false)]; + tensor 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("concat_31")]; + tensor attention_mask_15_begin_0 = const()[name = tensor("attention_mask_15_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_15_end_mask_0 = const()[name = tensor("attention_mask_15_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_15")]; + tensor query_11 = transpose(perm = query_11_perm_0, x = var_432)[name = tensor("transpose_27")]; + tensor mul_5 = mul(x = query_11, y = var_7)[name = tensor("mul_5")]; + tensor matmul_5_transpose_y_0 = const()[name = tensor("matmul_5_transpose_y_0"), val = tensor(true)]; + tensor matmul_5_transpose_x_0 = const()[name = tensor("matmul_5_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_5")]; + tensor add_5 = add(x = matmul_5, y = attention_mask_15)[name = tensor("add_5")]; + tensor softmax_5_axis_0 = const()[name = tensor("softmax_5_axis_0"), val = tensor(-1)]; + tensor softmax_5 = softmax(axis = softmax_5_axis_0, x = add_5)[name = tensor("softmax_5")]; + tensor attn_output_21_transpose_x_0 = const()[name = tensor("attn_output_21_transpose_x_0"), val = tensor(false)]; + tensor attn_output_21_transpose_y_0 = const()[name = tensor("attn_output_21_transpose_y_0"), val = tensor(false)]; + tensor value_11 = transpose(perm = value_11_perm_0, x = var_444)[name = tensor("transpose_25")]; + tensor 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("attn_output_21")]; + tensor var_452_perm_0 = const()[name = tensor("op_452_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_32_axis_0 = const()[name = tensor("concat_32_axis_0"), val = tensor(0)]; + tensor concat_32_interleave_0 = const()[name = tensor("concat_32_interleave_0"), val = tensor(false)]; + tensor concat_32 = concat(axis = concat_32_axis_0, interleave = concat_32_interleave_0, values = (gather_21, gather_22, var_23))[name = tensor("concat_32")]; + tensor var_452 = transpose(perm = var_452_perm_0, x = attn_output_21)[name = tensor("transpose_24")]; + tensor var_455 = reshape(shape = concat_32, x = var_452)[name = tensor("op_455")]; + tensor 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("linear_33")]; + tensor input_101 = add(x = input_95, y = input_99)[name = tensor("input_101")]; + tensor input_103_axes_0 = const()[name = tensor("input_103_axes_0"), val = tensor([-1])]; + tensor 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("input_103")]; + tensor input_105 = linear(bias = encoder_layers_5_fc1_bias, weight = encoder_layers_5_fc1_weight, x = input_103)[name = tensor("linear_34")]; + tensor input_107 = relu(x = input_105)[name = tensor("input_107")]; + tensor input_111 = linear(bias = encoder_layers_5_fc2_bias, weight = encoder_layers_5_fc2_weight, x = input_107)[name = tensor("linear_35")]; + tensor input_113 = add(x = input_101, y = input_111)[name = tensor("input_113")]; + tensor hidden_states_37_axes_0 = const()[name = tensor("hidden_states_37_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_37")]; + tensor var_489_shape = shape(x = hidden_states_37)[name = tensor("op_489_shape")]; + tensor gather_24 = const()[name = tensor("gather_24"), val = tensor(1)]; + tensor gather_25_indices_0 = const()[name = tensor("gather_25_indices_0"), val = tensor(1)]; + tensor gather_25_axis_0 = const()[name = tensor("gather_25_axis_0"), val = tensor(0)]; + tensor gather_25_batch_dims_0 = const()[name = tensor("gather_25_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_25")]; + tensor 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("linear_36")]; + tensor concat_33x = const()[name = tensor("concat_33x"), val = tensor([1, -1, 16, 64])]; + tensor var_495 = reshape(shape = concat_33x, x = var_493)[name = tensor("op_495")]; + tensor query_13_perm_0 = const()[name = tensor("query_13_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_37")]; + tensor 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("linear_38")]; + tensor concat_34x = const()[name = tensor("concat_34x"), val = tensor([1, -1, 16, 64])]; + tensor var_504 = reshape(shape = concat_34x, x = key_states_13)[name = tensor("op_504")]; + tensor key_13_perm_0 = const()[name = tensor("key_13_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_35x = const()[name = tensor("concat_35x"), val = tensor([1, -1, 16, 64])]; + tensor var_507 = reshape(shape = concat_35x, x = value_states_13)[name = tensor("op_507")]; + tensor value_13_perm_0 = const()[name = tensor("value_13_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_13 = transpose(perm = key_13_perm_0, x = var_504)[name = tensor("transpose_22")]; + tensor var_509_shape = shape(x = key_13)[name = tensor("op_509_shape")]; + tensor gather_26_indices_0 = const()[name = tensor("gather_26_indices_0"), val = tensor(2)]; + tensor gather_26_axis_0 = const()[name = tensor("gather_26_axis_0"), val = tensor(0)]; + tensor gather_26_batch_dims_0 = const()[name = tensor("gather_26_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_26")]; + tensor concat_36_values0_0 = const()[name = tensor("concat_36_values0_0"), val = tensor(0)]; + tensor concat_36_values1_0 = const()[name = tensor("concat_36_values1_0"), val = tensor(0)]; + tensor concat_36_values2_0 = const()[name = tensor("concat_36_values2_0"), val = tensor(0)]; + tensor concat_36_axis_0 = const()[name = tensor("concat_36_axis_0"), val = tensor(0)]; + tensor concat_36_interleave_0 = const()[name = tensor("concat_36_interleave_0"), val = tensor(false)]; + tensor 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("concat_36")]; + tensor attention_mask_17_begin_0 = const()[name = tensor("attention_mask_17_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_17_end_mask_0 = const()[name = tensor("attention_mask_17_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_17")]; + tensor query_13 = transpose(perm = query_13_perm_0, x = var_495)[name = tensor("transpose_23")]; + tensor mul_6 = mul(x = query_13, y = var_7)[name = tensor("mul_6")]; + tensor matmul_6_transpose_y_0 = const()[name = tensor("matmul_6_transpose_y_0"), val = tensor(true)]; + tensor matmul_6_transpose_x_0 = const()[name = tensor("matmul_6_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_6")]; + tensor add_6 = add(x = matmul_6, y = attention_mask_17)[name = tensor("add_6")]; + tensor softmax_6_axis_0 = const()[name = tensor("softmax_6_axis_0"), val = tensor(-1)]; + tensor softmax_6 = softmax(axis = softmax_6_axis_0, x = add_6)[name = tensor("softmax_6")]; + tensor attn_output_25_transpose_x_0 = const()[name = tensor("attn_output_25_transpose_x_0"), val = tensor(false)]; + tensor attn_output_25_transpose_y_0 = const()[name = tensor("attn_output_25_transpose_y_0"), val = tensor(false)]; + tensor value_13 = transpose(perm = value_13_perm_0, x = var_507)[name = tensor("transpose_21")]; + tensor 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("attn_output_25")]; + tensor var_515_perm_0 = const()[name = tensor("op_515_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_37_axis_0 = const()[name = tensor("concat_37_axis_0"), val = tensor(0)]; + tensor concat_37_interleave_0 = const()[name = tensor("concat_37_interleave_0"), val = tensor(false)]; + tensor concat_37 = concat(axis = concat_37_axis_0, interleave = concat_37_interleave_0, values = (gather_24, gather_25, var_23))[name = tensor("concat_37")]; + tensor var_515 = transpose(perm = var_515_perm_0, x = attn_output_25)[name = tensor("transpose_20")]; + tensor var_518 = reshape(shape = concat_37, x = var_515)[name = tensor("op_518")]; + tensor 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("linear_39")]; + tensor input_119 = add(x = input_113, y = input_117)[name = tensor("input_119")]; + tensor input_121_axes_0 = const()[name = tensor("input_121_axes_0"), val = tensor([-1])]; + tensor 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("input_121")]; + tensor input_123 = linear(bias = encoder_layers_6_fc1_bias, weight = encoder_layers_6_fc1_weight, x = input_121)[name = tensor("linear_40")]; + tensor input_125 = relu(x = input_123)[name = tensor("input_125")]; + tensor input_129 = linear(bias = encoder_layers_6_fc2_bias, weight = encoder_layers_6_fc2_weight, x = input_125)[name = tensor("linear_41")]; + tensor input_131 = add(x = input_119, y = input_129)[name = tensor("input_131")]; + tensor hidden_states_43_axes_0 = const()[name = tensor("hidden_states_43_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_43")]; + tensor var_552_shape = shape(x = hidden_states_43)[name = tensor("op_552_shape")]; + tensor gather_27 = const()[name = tensor("gather_27"), val = tensor(1)]; + tensor gather_28_indices_0 = const()[name = tensor("gather_28_indices_0"), val = tensor(1)]; + tensor gather_28_axis_0 = const()[name = tensor("gather_28_axis_0"), val = tensor(0)]; + tensor gather_28_batch_dims_0 = const()[name = tensor("gather_28_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_28")]; + tensor 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("linear_42")]; + tensor concat_38x = const()[name = tensor("concat_38x"), val = tensor([1, -1, 16, 64])]; + tensor var_558 = reshape(shape = concat_38x, x = var_556)[name = tensor("op_558")]; + tensor query_15_perm_0 = const()[name = tensor("query_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_43")]; + tensor 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("linear_44")]; + tensor concat_39x = const()[name = tensor("concat_39x"), val = tensor([1, -1, 16, 64])]; + tensor var_567 = reshape(shape = concat_39x, x = key_states_15)[name = tensor("op_567")]; + tensor key_15_perm_0 = const()[name = tensor("key_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_40x = const()[name = tensor("concat_40x"), val = tensor([1, -1, 16, 64])]; + tensor var_570 = reshape(shape = concat_40x, x = value_states_15)[name = tensor("op_570")]; + tensor value_15_perm_0 = const()[name = tensor("value_15_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_15 = transpose(perm = key_15_perm_0, x = var_567)[name = tensor("transpose_18")]; + tensor var_572_shape = shape(x = key_15)[name = tensor("op_572_shape")]; + tensor gather_29_indices_0 = const()[name = tensor("gather_29_indices_0"), val = tensor(2)]; + tensor gather_29_axis_0 = const()[name = tensor("gather_29_axis_0"), val = tensor(0)]; + tensor gather_29_batch_dims_0 = const()[name = tensor("gather_29_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_29")]; + tensor concat_41_values0_0 = const()[name = tensor("concat_41_values0_0"), val = tensor(0)]; + tensor concat_41_values1_0 = const()[name = tensor("concat_41_values1_0"), val = tensor(0)]; + tensor concat_41_values2_0 = const()[name = tensor("concat_41_values2_0"), val = tensor(0)]; + tensor concat_41_axis_0 = const()[name = tensor("concat_41_axis_0"), val = tensor(0)]; + tensor concat_41_interleave_0 = const()[name = tensor("concat_41_interleave_0"), val = tensor(false)]; + tensor 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("concat_41")]; + tensor attention_mask_19_begin_0 = const()[name = tensor("attention_mask_19_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_19_end_mask_0 = const()[name = tensor("attention_mask_19_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_19")]; + tensor query_15 = transpose(perm = query_15_perm_0, x = var_558)[name = tensor("transpose_19")]; + tensor mul_7 = mul(x = query_15, y = var_7)[name = tensor("mul_7")]; + tensor matmul_7_transpose_y_0 = const()[name = tensor("matmul_7_transpose_y_0"), val = tensor(true)]; + tensor matmul_7_transpose_x_0 = const()[name = tensor("matmul_7_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_7")]; + tensor add_7 = add(x = matmul_7, y = attention_mask_19)[name = tensor("add_7")]; + tensor softmax_7_axis_0 = const()[name = tensor("softmax_7_axis_0"), val = tensor(-1)]; + tensor softmax_7 = softmax(axis = softmax_7_axis_0, x = add_7)[name = tensor("softmax_7")]; + tensor attn_output_29_transpose_x_0 = const()[name = tensor("attn_output_29_transpose_x_0"), val = tensor(false)]; + tensor attn_output_29_transpose_y_0 = const()[name = tensor("attn_output_29_transpose_y_0"), val = tensor(false)]; + tensor value_15 = transpose(perm = value_15_perm_0, x = var_570)[name = tensor("transpose_17")]; + tensor 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("attn_output_29")]; + tensor var_578_perm_0 = const()[name = tensor("op_578_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_42_axis_0 = const()[name = tensor("concat_42_axis_0"), val = tensor(0)]; + tensor concat_42_interleave_0 = const()[name = tensor("concat_42_interleave_0"), val = tensor(false)]; + tensor concat_42 = concat(axis = concat_42_axis_0, interleave = concat_42_interleave_0, values = (gather_27, gather_28, var_23))[name = tensor("concat_42")]; + tensor var_578 = transpose(perm = var_578_perm_0, x = attn_output_29)[name = tensor("transpose_16")]; + tensor var_581 = reshape(shape = concat_42, x = var_578)[name = tensor("op_581")]; + tensor 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("linear_45")]; + tensor input_137 = add(x = input_131, y = input_135)[name = tensor("input_137")]; + tensor input_139_axes_0 = const()[name = tensor("input_139_axes_0"), val = tensor([-1])]; + tensor 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("input_139")]; + tensor input_141 = linear(bias = encoder_layers_7_fc1_bias, weight = encoder_layers_7_fc1_weight, x = input_139)[name = tensor("linear_46")]; + tensor input_143 = relu(x = input_141)[name = tensor("input_143")]; + tensor input_147 = linear(bias = encoder_layers_7_fc2_bias, weight = encoder_layers_7_fc2_weight, x = input_143)[name = tensor("linear_47")]; + tensor input_149 = add(x = input_137, y = input_147)[name = tensor("input_149")]; + tensor hidden_states_49_axes_0 = const()[name = tensor("hidden_states_49_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_49")]; + tensor var_615_shape = shape(x = hidden_states_49)[name = tensor("op_615_shape")]; + tensor gather_30 = const()[name = tensor("gather_30"), val = tensor(1)]; + tensor gather_31_indices_0 = const()[name = tensor("gather_31_indices_0"), val = tensor(1)]; + tensor gather_31_axis_0 = const()[name = tensor("gather_31_axis_0"), val = tensor(0)]; + tensor gather_31_batch_dims_0 = const()[name = tensor("gather_31_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_31")]; + tensor 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("linear_48")]; + tensor concat_43x = const()[name = tensor("concat_43x"), val = tensor([1, -1, 16, 64])]; + tensor var_621 = reshape(shape = concat_43x, x = var_619)[name = tensor("op_621")]; + tensor query_17_perm_0 = const()[name = tensor("query_17_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_49")]; + tensor 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("linear_50")]; + tensor concat_44x = const()[name = tensor("concat_44x"), val = tensor([1, -1, 16, 64])]; + tensor var_630 = reshape(shape = concat_44x, x = key_states_17)[name = tensor("op_630")]; + tensor key_17_perm_0 = const()[name = tensor("key_17_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_45x = const()[name = tensor("concat_45x"), val = tensor([1, -1, 16, 64])]; + tensor var_633 = reshape(shape = concat_45x, x = value_states_17)[name = tensor("op_633")]; + tensor value_17_perm_0 = const()[name = tensor("value_17_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_17 = transpose(perm = key_17_perm_0, x = var_630)[name = tensor("transpose_14")]; + tensor var_635_shape = shape(x = key_17)[name = tensor("op_635_shape")]; + tensor gather_32_indices_0 = const()[name = tensor("gather_32_indices_0"), val = tensor(2)]; + tensor gather_32_axis_0 = const()[name = tensor("gather_32_axis_0"), val = tensor(0)]; + tensor gather_32_batch_dims_0 = const()[name = tensor("gather_32_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_32")]; + tensor concat_46_values0_0 = const()[name = tensor("concat_46_values0_0"), val = tensor(0)]; + tensor concat_46_values1_0 = const()[name = tensor("concat_46_values1_0"), val = tensor(0)]; + tensor concat_46_values2_0 = const()[name = tensor("concat_46_values2_0"), val = tensor(0)]; + tensor concat_46_axis_0 = const()[name = tensor("concat_46_axis_0"), val = tensor(0)]; + tensor concat_46_interleave_0 = const()[name = tensor("concat_46_interleave_0"), val = tensor(false)]; + tensor 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("concat_46")]; + tensor attention_mask_21_begin_0 = const()[name = tensor("attention_mask_21_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_21_end_mask_0 = const()[name = tensor("attention_mask_21_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_21")]; + tensor query_17 = transpose(perm = query_17_perm_0, x = var_621)[name = tensor("transpose_15")]; + tensor mul_8 = mul(x = query_17, y = var_7)[name = tensor("mul_8")]; + tensor matmul_8_transpose_y_0 = const()[name = tensor("matmul_8_transpose_y_0"), val = tensor(true)]; + tensor matmul_8_transpose_x_0 = const()[name = tensor("matmul_8_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_8")]; + tensor add_8 = add(x = matmul_8, y = attention_mask_21)[name = tensor("add_8")]; + tensor softmax_8_axis_0 = const()[name = tensor("softmax_8_axis_0"), val = tensor(-1)]; + tensor softmax_8 = softmax(axis = softmax_8_axis_0, x = add_8)[name = tensor("softmax_8")]; + tensor attn_output_33_transpose_x_0 = const()[name = tensor("attn_output_33_transpose_x_0"), val = tensor(false)]; + tensor attn_output_33_transpose_y_0 = const()[name = tensor("attn_output_33_transpose_y_0"), val = tensor(false)]; + tensor value_17 = transpose(perm = value_17_perm_0, x = var_633)[name = tensor("transpose_13")]; + tensor 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("attn_output_33")]; + tensor var_641_perm_0 = const()[name = tensor("op_641_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_47_axis_0 = const()[name = tensor("concat_47_axis_0"), val = tensor(0)]; + tensor concat_47_interleave_0 = const()[name = tensor("concat_47_interleave_0"), val = tensor(false)]; + tensor concat_47 = concat(axis = concat_47_axis_0, interleave = concat_47_interleave_0, values = (gather_30, gather_31, var_23))[name = tensor("concat_47")]; + tensor var_641 = transpose(perm = var_641_perm_0, x = attn_output_33)[name = tensor("transpose_12")]; + tensor var_644 = reshape(shape = concat_47, x = var_641)[name = tensor("op_644")]; + tensor 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("linear_51")]; + tensor input_155 = add(x = input_149, y = input_153)[name = tensor("input_155")]; + tensor input_157_axes_0 = const()[name = tensor("input_157_axes_0"), val = tensor([-1])]; + tensor 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("input_157")]; + tensor input_159 = linear(bias = encoder_layers_8_fc1_bias, weight = encoder_layers_8_fc1_weight, x = input_157)[name = tensor("linear_52")]; + tensor input_161 = relu(x = input_159)[name = tensor("input_161")]; + tensor input_165 = linear(bias = encoder_layers_8_fc2_bias, weight = encoder_layers_8_fc2_weight, x = input_161)[name = tensor("linear_53")]; + tensor input_167 = add(x = input_155, y = input_165)[name = tensor("input_167")]; + tensor hidden_states_55_axes_0 = const()[name = tensor("hidden_states_55_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_55")]; + tensor var_678_shape = shape(x = hidden_states_55)[name = tensor("op_678_shape")]; + tensor gather_33 = const()[name = tensor("gather_33"), val = tensor(1)]; + tensor gather_34_indices_0 = const()[name = tensor("gather_34_indices_0"), val = tensor(1)]; + tensor gather_34_axis_0 = const()[name = tensor("gather_34_axis_0"), val = tensor(0)]; + tensor gather_34_batch_dims_0 = const()[name = tensor("gather_34_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_34")]; + tensor 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("linear_54")]; + tensor concat_48x = const()[name = tensor("concat_48x"), val = tensor([1, -1, 16, 64])]; + tensor var_684 = reshape(shape = concat_48x, x = var_682)[name = tensor("op_684")]; + tensor query_19_perm_0 = const()[name = tensor("query_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_55")]; + tensor 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("linear_56")]; + tensor concat_49x = const()[name = tensor("concat_49x"), val = tensor([1, -1, 16, 64])]; + tensor var_693 = reshape(shape = concat_49x, x = key_states_19)[name = tensor("op_693")]; + tensor key_19_perm_0 = const()[name = tensor("key_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_50x = const()[name = tensor("concat_50x"), val = tensor([1, -1, 16, 64])]; + tensor var_696 = reshape(shape = concat_50x, x = value_states_19)[name = tensor("op_696")]; + tensor value_19_perm_0 = const()[name = tensor("value_19_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_19 = transpose(perm = key_19_perm_0, x = var_693)[name = tensor("transpose_10")]; + tensor var_698_shape = shape(x = key_19)[name = tensor("op_698_shape")]; + tensor gather_35_indices_0 = const()[name = tensor("gather_35_indices_0"), val = tensor(2)]; + tensor gather_35_axis_0 = const()[name = tensor("gather_35_axis_0"), val = tensor(0)]; + tensor gather_35_batch_dims_0 = const()[name = tensor("gather_35_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_35")]; + tensor concat_51_values0_0 = const()[name = tensor("concat_51_values0_0"), val = tensor(0)]; + tensor concat_51_values1_0 = const()[name = tensor("concat_51_values1_0"), val = tensor(0)]; + tensor concat_51_values2_0 = const()[name = tensor("concat_51_values2_0"), val = tensor(0)]; + tensor concat_51_axis_0 = const()[name = tensor("concat_51_axis_0"), val = tensor(0)]; + tensor concat_51_interleave_0 = const()[name = tensor("concat_51_interleave_0"), val = tensor(false)]; + tensor 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("concat_51")]; + tensor attention_mask_23_begin_0 = const()[name = tensor("attention_mask_23_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_23_end_mask_0 = const()[name = tensor("attention_mask_23_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_23")]; + tensor query_19 = transpose(perm = query_19_perm_0, x = var_684)[name = tensor("transpose_11")]; + tensor mul_9 = mul(x = query_19, y = var_7)[name = tensor("mul_9")]; + tensor matmul_9_transpose_y_0 = const()[name = tensor("matmul_9_transpose_y_0"), val = tensor(true)]; + tensor matmul_9_transpose_x_0 = const()[name = tensor("matmul_9_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_9")]; + tensor add_9 = add(x = matmul_9, y = attention_mask_23)[name = tensor("add_9")]; + tensor softmax_9_axis_0 = const()[name = tensor("softmax_9_axis_0"), val = tensor(-1)]; + tensor softmax_9 = softmax(axis = softmax_9_axis_0, x = add_9)[name = tensor("softmax_9")]; + tensor attn_output_37_transpose_x_0 = const()[name = tensor("attn_output_37_transpose_x_0"), val = tensor(false)]; + tensor attn_output_37_transpose_y_0 = const()[name = tensor("attn_output_37_transpose_y_0"), val = tensor(false)]; + tensor value_19 = transpose(perm = value_19_perm_0, x = var_696)[name = tensor("transpose_9")]; + tensor 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("attn_output_37")]; + tensor var_704_perm_0 = const()[name = tensor("op_704_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_52_axis_0 = const()[name = tensor("concat_52_axis_0"), val = tensor(0)]; + tensor concat_52_interleave_0 = const()[name = tensor("concat_52_interleave_0"), val = tensor(false)]; + tensor concat_52 = concat(axis = concat_52_axis_0, interleave = concat_52_interleave_0, values = (gather_33, gather_34, var_23))[name = tensor("concat_52")]; + tensor var_704 = transpose(perm = var_704_perm_0, x = attn_output_37)[name = tensor("transpose_8")]; + tensor var_707 = reshape(shape = concat_52, x = var_704)[name = tensor("op_707")]; + tensor 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("linear_57")]; + tensor input_173 = add(x = input_167, y = input_171)[name = tensor("input_173")]; + tensor input_175_axes_0 = const()[name = tensor("input_175_axes_0"), val = tensor([-1])]; + tensor 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("input_175")]; + tensor input_177 = linear(bias = encoder_layers_9_fc1_bias, weight = encoder_layers_9_fc1_weight, x = input_175)[name = tensor("linear_58")]; + tensor input_179 = relu(x = input_177)[name = tensor("input_179")]; + tensor input_183 = linear(bias = encoder_layers_9_fc2_bias, weight = encoder_layers_9_fc2_weight, x = input_179)[name = tensor("linear_59")]; + tensor input_185 = add(x = input_173, y = input_183)[name = tensor("input_185")]; + tensor hidden_states_61_axes_0 = const()[name = tensor("hidden_states_61_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_61")]; + tensor var_741_shape = shape(x = hidden_states_61)[name = tensor("op_741_shape")]; + tensor gather_36 = const()[name = tensor("gather_36"), val = tensor(1)]; + tensor gather_37_indices_0 = const()[name = tensor("gather_37_indices_0"), val = tensor(1)]; + tensor gather_37_axis_0 = const()[name = tensor("gather_37_axis_0"), val = tensor(0)]; + tensor gather_37_batch_dims_0 = const()[name = tensor("gather_37_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_37")]; + tensor 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("linear_60")]; + tensor concat_53x = const()[name = tensor("concat_53x"), val = tensor([1, -1, 16, 64])]; + tensor var_747 = reshape(shape = concat_53x, x = var_745)[name = tensor("op_747")]; + tensor query_21_perm_0 = const()[name = tensor("query_21_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_61")]; + tensor 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("linear_62")]; + tensor concat_54x = const()[name = tensor("concat_54x"), val = tensor([1, -1, 16, 64])]; + tensor var_756 = reshape(shape = concat_54x, x = key_states_21)[name = tensor("op_756")]; + tensor key_21_perm_0 = const()[name = tensor("key_21_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_55x = const()[name = tensor("concat_55x"), val = tensor([1, -1, 16, 64])]; + tensor var_759 = reshape(shape = concat_55x, x = value_states_21)[name = tensor("op_759")]; + tensor value_21_perm_0 = const()[name = tensor("value_21_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key_21 = transpose(perm = key_21_perm_0, x = var_756)[name = tensor("transpose_6")]; + tensor var_761_shape = shape(x = key_21)[name = tensor("op_761_shape")]; + tensor gather_38_indices_0 = const()[name = tensor("gather_38_indices_0"), val = tensor(2)]; + tensor gather_38_axis_0 = const()[name = tensor("gather_38_axis_0"), val = tensor(0)]; + tensor gather_38_batch_dims_0 = const()[name = tensor("gather_38_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_38")]; + tensor concat_56_values0_0 = const()[name = tensor("concat_56_values0_0"), val = tensor(0)]; + tensor concat_56_values1_0 = const()[name = tensor("concat_56_values1_0"), val = tensor(0)]; + tensor concat_56_values2_0 = const()[name = tensor("concat_56_values2_0"), val = tensor(0)]; + tensor concat_56_axis_0 = const()[name = tensor("concat_56_axis_0"), val = tensor(0)]; + tensor concat_56_interleave_0 = const()[name = tensor("concat_56_interleave_0"), val = tensor(false)]; + tensor 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("concat_56")]; + tensor attention_mask_25_begin_0 = const()[name = tensor("attention_mask_25_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_25_end_mask_0 = const()[name = tensor("attention_mask_25_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask_25")]; + tensor query_21 = transpose(perm = query_21_perm_0, x = var_747)[name = tensor("transpose_7")]; + tensor mul_10 = mul(x = query_21, y = var_7)[name = tensor("mul_10")]; + tensor matmul_10_transpose_y_0 = const()[name = tensor("matmul_10_transpose_y_0"), val = tensor(true)]; + tensor matmul_10_transpose_x_0 = const()[name = tensor("matmul_10_transpose_x_0"), val = tensor(false)]; + tensor 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("matmul_10")]; + tensor add_10 = add(x = matmul_10, y = attention_mask_25)[name = tensor("add_10")]; + tensor softmax_10_axis_0 = const()[name = tensor("softmax_10_axis_0"), val = tensor(-1)]; + tensor softmax_10 = softmax(axis = softmax_10_axis_0, x = add_10)[name = tensor("softmax_10")]; + tensor attn_output_41_transpose_x_0 = const()[name = tensor("attn_output_41_transpose_x_0"), val = tensor(false)]; + tensor attn_output_41_transpose_y_0 = const()[name = tensor("attn_output_41_transpose_y_0"), val = tensor(false)]; + tensor value_21 = transpose(perm = value_21_perm_0, x = var_759)[name = tensor("transpose_5")]; + tensor 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("attn_output_41")]; + tensor var_767_perm_0 = const()[name = tensor("op_767_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_57_axis_0 = const()[name = tensor("concat_57_axis_0"), val = tensor(0)]; + tensor concat_57_interleave_0 = const()[name = tensor("concat_57_interleave_0"), val = tensor(false)]; + tensor concat_57 = concat(axis = concat_57_axis_0, interleave = concat_57_interleave_0, values = (gather_36, gather_37, var_23))[name = tensor("concat_57")]; + tensor var_767 = transpose(perm = var_767_perm_0, x = attn_output_41)[name = tensor("transpose_4")]; + tensor var_770 = reshape(shape = concat_57, x = var_767)[name = tensor("op_770")]; + tensor 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("linear_63")]; + tensor input_191 = add(x = input_185, y = input_189)[name = tensor("input_191")]; + tensor input_193_axes_0 = const()[name = tensor("input_193_axes_0"), val = tensor([-1])]; + tensor 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("input_193")]; + tensor input_195 = linear(bias = encoder_layers_10_fc1_bias, weight = encoder_layers_10_fc1_weight, x = input_193)[name = tensor("linear_64")]; + tensor input_197 = relu(x = input_195)[name = tensor("input_197")]; + tensor input_201 = linear(bias = encoder_layers_10_fc2_bias, weight = encoder_layers_10_fc2_weight, x = input_197)[name = tensor("linear_65")]; + tensor input_203 = add(x = input_191, y = input_201)[name = tensor("input_203")]; + tensor hidden_states_67_axes_0 = const()[name = tensor("hidden_states_67_axes_0"), val = tensor([-1])]; + tensor 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("hidden_states_67")]; + tensor var_804_shape = shape(x = hidden_states_67)[name = tensor("op_804_shape")]; + tensor gather_39 = const()[name = tensor("gather_39"), val = tensor(1)]; + tensor gather_40_indices_0 = const()[name = tensor("gather_40_indices_0"), val = tensor(1)]; + tensor gather_40_axis_0 = const()[name = tensor("gather_40_axis_0"), val = tensor(0)]; + tensor gather_40_batch_dims_0 = const()[name = tensor("gather_40_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_40")]; + tensor 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("linear_66")]; + tensor concat_58x = const()[name = tensor("concat_58x"), val = tensor([1, -1, 16, 64])]; + tensor var_810 = reshape(shape = concat_58x, x = var_808)[name = tensor("op_810")]; + tensor query_perm_0 = const()[name = tensor("query_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor 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("linear_67")]; + tensor 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("linear_68")]; + tensor concat_59x = const()[name = tensor("concat_59x"), val = tensor([1, -1, 16, 64])]; + tensor var_819 = reshape(shape = concat_59x, x = key_states)[name = tensor("op_819")]; + tensor key_perm_0 = const()[name = tensor("key_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_60x = const()[name = tensor("concat_60x"), val = tensor([1, -1, 16, 64])]; + tensor var_822 = reshape(shape = concat_60x, x = value_states)[name = tensor("op_822")]; + tensor value_perm_0 = const()[name = tensor("value_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor key = transpose(perm = key_perm_0, x = var_819)[name = tensor("transpose_2")]; + tensor var_824_shape = shape(x = key)[name = tensor("op_824_shape")]; + tensor gather_41_indices_0 = const()[name = tensor("gather_41_indices_0"), val = tensor(2)]; + tensor gather_41_axis_0 = const()[name = tensor("gather_41_axis_0"), val = tensor(0)]; + tensor gather_41_batch_dims_0 = const()[name = tensor("gather_41_batch_dims_0"), val = tensor(0)]; + tensor 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("gather_41")]; + tensor concat_61_values0_0 = const()[name = tensor("concat_61_values0_0"), val = tensor(0)]; + tensor concat_61_values1_0 = const()[name = tensor("concat_61_values1_0"), val = tensor(0)]; + tensor concat_61_values2_0 = const()[name = tensor("concat_61_values2_0"), val = tensor(0)]; + tensor concat_61_axis_0 = const()[name = tensor("concat_61_axis_0"), val = tensor(0)]; + tensor concat_61_interleave_0 = const()[name = tensor("concat_61_interleave_0"), val = tensor(false)]; + tensor 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("concat_61")]; + tensor attention_mask_begin_0 = const()[name = tensor("attention_mask_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor attention_mask_end_mask_0 = const()[name = tensor("attention_mask_end_mask_0"), val = tensor([true, true, true, false])]; + tensor 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("attention_mask")]; + tensor query = transpose(perm = query_perm_0, x = var_810)[name = tensor("transpose_3")]; + tensor mul_11 = mul(x = query, y = var_7)[name = tensor("mul_11")]; + tensor matmul_11_transpose_y_0 = const()[name = tensor("matmul_11_transpose_y_0"), val = tensor(true)]; + tensor matmul_11_transpose_x_0 = const()[name = tensor("matmul_11_transpose_x_0"), val = tensor(false)]; + tensor matmul_11 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = mul_11, y = key)[name = tensor("matmul_11")]; + tensor add_11 = add(x = matmul_11, y = attention_mask_1)[name = tensor("add_11")]; + tensor softmax_11_axis_0 = const()[name = tensor("softmax_11_axis_0"), val = tensor(-1)]; + tensor softmax_11 = softmax(axis = softmax_11_axis_0, x = add_11)[name = tensor("softmax_11")]; + tensor attn_output_45_transpose_x_0 = const()[name = tensor("attn_output_45_transpose_x_0"), val = tensor(false)]; + tensor attn_output_45_transpose_y_0 = const()[name = tensor("attn_output_45_transpose_y_0"), val = tensor(false)]; + tensor value = transpose(perm = value_perm_0, x = var_822)[name = tensor("transpose_1")]; + tensor 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("attn_output_45")]; + tensor var_830_perm_0 = const()[name = tensor("op_830_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor concat_62_axis_0 = const()[name = tensor("concat_62_axis_0"), val = tensor(0)]; + tensor concat_62_interleave_0 = const()[name = tensor("concat_62_interleave_0"), val = tensor(false)]; + tensor concat_62 = concat(axis = concat_62_axis_0, interleave = concat_62_interleave_0, values = (gather_39, gather_40, var_23))[name = tensor("concat_62")]; + tensor var_830 = transpose(perm = var_830_perm_0, x = attn_output_45)[name = tensor("transpose_0")]; + tensor var_833 = reshape(shape = concat_62, x = var_830)[name = tensor("op_833")]; + tensor 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("linear_69")]; + tensor input_209 = add(x = input_203, y = input_207)[name = tensor("input_209")]; + tensor input_211_axes_0 = const()[name = tensor("input_211_axes_0"), val = tensor([-1])]; + tensor 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("input_211")]; + tensor input_213 = linear(bias = encoder_layers_11_fc1_bias, weight = encoder_layers_11_fc1_weight, x = input_211)[name = tensor("linear_70")]; + tensor input_215 = relu(x = input_213)[name = tensor("input_215")]; + tensor input_219 = linear(bias = encoder_layers_11_fc2_bias, weight = encoder_layers_11_fc2_weight, x = input_215)[name = tensor("linear_71")]; + tensor input = add(x = input_209, y = input_219)[name = tensor("input")]; + tensor var_857_axes_0 = const()[name = tensor("op_857_axes_0"), val = tensor([-1])]; + tensor 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("op_857")]; + } -> (output); +} \ No newline at end of file