| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.8.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] |
| { |
| func main<ios16>(tensor<int32, [1, ?]> encoder_attention_mask, tensor<fp32, [1, ?, 1024]> encoder_hidden_states, tensor<int32, [1, 1]> input_ids, tensor<fp16, [12, 1, 16, ?, 64]> past_cross_key, tensor<fp16, [12, 1, 16, ?, 64]> past_cross_value, tensor<fp16, [12, 1, 16, ?, 64]> past_self_key, tensor<fp16, [12, 1, 16, ?, 64]> past_self_value) [FlexibleShapeInformation = tuple<tuple<tensor<string, []>, dict<tensor<string, []>, tensor<int32, [?]>>>, tuple<tensor<string, []>, dict<tensor<string, []>, list<tensor<int32, [2]>, ?>>>>((("DefaultShapes", {{"encoder_attention_mask", [1, 1]}, {"encoder_hidden_states", [1, 1, 1024]}, {"past_cross_key", [12, 1, 16, 1, 64]}, {"past_cross_value", [12, 1, 16, 1, 64]}, {"past_self_key", [12, 1, 16, 1, 64]}, {"past_self_value", [12, 1, 16, 1, 64]}}), ("RangeDims", {{"encoder_attention_mask", [[1, 1], [1, 1024]]}, {"encoder_hidden_states", [[1, 1], [1, 1024], [1024, 1024]]}, {"past_cross_key", [[12, 12], [1, 1], [16, 16], [1, 1024], [64, 64]]}, {"past_cross_value", [[12, 12], [1, 1], [16, 16], [1, 1024], [64, 64]]}, {"past_self_key", [[12, 12], [1, 1], [16, 16], [1, 1023], [64, 64]]}, {"past_self_value", [[12, 12], [1, 1], [16, 16], [1, 1023], [64, 64]]}})))] { |
| tensor<string, []> cast_0_dtype_0 = const()[name = tensor<string, []>("cast_0_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<string, []> cast_1_dtype_0 = const()[name = tensor<string, []>("cast_1_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<string, []> cast_2_dtype_0 = const()[name = tensor<string, []>("cast_2_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<string, []> cast_3_dtype_0 = const()[name = tensor<string, []>("cast_3_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [256206, 1024]> decoder_embed_tokens_weight = const()[name = tensor<string, []>("decoder_embed_tokens_weight"), val = tensor<fp32, [256206, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<fp32, [1026, 1024]> decoder_embed_positions_weights = const()[name = tensor<string, []>("decoder_embed_positions_weights"), val = tensor<fp32, [1026, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1049419904)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_0_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053622464)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_0_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053626624)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053630784)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1053634944)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1057829312)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1057833472)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1062027840)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1062032000)))]; |
| tensor<fp32, [1024]> decoder_layers_0_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1066226368)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1066230528)))]; |
| tensor<fp32, [1024]> decoder_layers_0_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070424896)))]; |
| tensor<fp32, [1024]> decoder_layers_0_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070429056)))]; |
| tensor<fp32, [1024]> decoder_layers_0_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070433216)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070437376)))]; |
| tensor<fp32, [1024]> decoder_layers_0_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1074631744)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1074635904)))]; |
| tensor<fp32, [1024]> decoder_layers_0_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_0_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078830272)))]; |
| tensor<fp32, [1024]> decoder_layers_0_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_0_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078834432)))]; |
| tensor<fp32, [4096]> decoder_layers_0_fc1_bias = const()[name = tensor<string, []>("decoder_layers_0_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078838592)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_0_fc1_weight = const()[name = tensor<string, []>("decoder_layers_0_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078855040)))]; |
| tensor<fp32, [1024]> decoder_layers_0_fc2_bias = const()[name = tensor<string, []>("decoder_layers_0_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1095632320)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_0_fc2_weight = const()[name = tensor<string, []>("decoder_layers_0_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1095636480)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_1_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112413760)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_1_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112417920)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112422080)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112426240)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116620608)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116624768)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120819136)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1120823296)))]; |
| tensor<fp32, [1024]> decoder_layers_1_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1125017664)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1125021824)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129216192)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129220352)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129224512)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129228672)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1133423040)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1133427200)))]; |
| tensor<fp32, [1024]> decoder_layers_1_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_1_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137621568)))]; |
| tensor<fp32, [1024]> decoder_layers_1_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_1_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137625728)))]; |
| tensor<fp32, [4096]> decoder_layers_1_fc1_bias = const()[name = tensor<string, []>("decoder_layers_1_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137629888)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_1_fc1_weight = const()[name = tensor<string, []>("decoder_layers_1_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1137646336)))]; |
| tensor<fp32, [1024]> decoder_layers_1_fc2_bias = const()[name = tensor<string, []>("decoder_layers_1_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154423616)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_1_fc2_weight = const()[name = tensor<string, []>("decoder_layers_1_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154427776)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_2_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171205056)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_2_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171209216)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171213376)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171217536)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1175411904)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1175416064)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1179610432)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1179614592)))]; |
| tensor<fp32, [1024]> decoder_layers_2_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183808960)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1183813120)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188007488)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188011648)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188015808)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1188019968)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192214336)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1192218496)))]; |
| tensor<fp32, [1024]> decoder_layers_2_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_2_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1196412864)))]; |
| tensor<fp32, [1024]> decoder_layers_2_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_2_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1196417024)))]; |
| tensor<fp32, [4096]> decoder_layers_2_fc1_bias = const()[name = tensor<string, []>("decoder_layers_2_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1196421184)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_2_fc1_weight = const()[name = tensor<string, []>("decoder_layers_2_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1196437632)))]; |
| tensor<fp32, [1024]> decoder_layers_2_fc2_bias = const()[name = tensor<string, []>("decoder_layers_2_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213214912)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_2_fc2_weight = const()[name = tensor<string, []>("decoder_layers_2_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213219072)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_3_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1229996352)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_3_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1230000512)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1230004672)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1230008832)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234203200)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234207360)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238401728)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1238405888)))]; |
| tensor<fp32, [1024]> decoder_layers_3_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1242600256)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1242604416)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1246798784)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1246802944)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1246807104)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1246811264)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1251005632)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1251009792)))]; |
| tensor<fp32, [1024]> decoder_layers_3_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_3_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255204160)))]; |
| tensor<fp32, [1024]> decoder_layers_3_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_3_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255208320)))]; |
| tensor<fp32, [4096]> decoder_layers_3_fc1_bias = const()[name = tensor<string, []>("decoder_layers_3_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255212480)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_3_fc1_weight = const()[name = tensor<string, []>("decoder_layers_3_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1255228928)))]; |
| tensor<fp32, [1024]> decoder_layers_3_fc2_bias = const()[name = tensor<string, []>("decoder_layers_3_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1272006208)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_3_fc2_weight = const()[name = tensor<string, []>("decoder_layers_3_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1272010368)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_4_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288787648)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_4_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288791808)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288795968)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1288800128)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1292994496)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1292998656)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1297193024)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1297197184)))]; |
| tensor<fp32, [1024]> decoder_layers_4_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1301391552)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1301395712)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305590080)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305594240)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305598400)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1305602560)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309796928)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1309801088)))]; |
| tensor<fp32, [1024]> decoder_layers_4_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_4_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313995456)))]; |
| tensor<fp32, [1024]> decoder_layers_4_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_4_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1313999616)))]; |
| tensor<fp32, [4096]> decoder_layers_4_fc1_bias = const()[name = tensor<string, []>("decoder_layers_4_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1314003776)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_4_fc1_weight = const()[name = tensor<string, []>("decoder_layers_4_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1314020224)))]; |
| tensor<fp32, [1024]> decoder_layers_4_fc2_bias = const()[name = tensor<string, []>("decoder_layers_4_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1330797504)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_4_fc2_weight = const()[name = tensor<string, []>("decoder_layers_4_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1330801664)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_5_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347578944)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_5_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347583104)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347587264)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347591424)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1351785792)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1351789952)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355984320)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1355988480)))]; |
| tensor<fp32, [1024]> decoder_layers_5_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360182848)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1360187008)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364381376)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364385536)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364389696)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1364393856)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1368588224)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1368592384)))]; |
| tensor<fp32, [1024]> decoder_layers_5_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_5_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372786752)))]; |
| tensor<fp32, [1024]> decoder_layers_5_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_5_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372790912)))]; |
| tensor<fp32, [4096]> decoder_layers_5_fc1_bias = const()[name = tensor<string, []>("decoder_layers_5_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372795072)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_5_fc1_weight = const()[name = tensor<string, []>("decoder_layers_5_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1372811520)))]; |
| tensor<fp32, [1024]> decoder_layers_5_fc2_bias = const()[name = tensor<string, []>("decoder_layers_5_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389588800)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_5_fc2_weight = const()[name = tensor<string, []>("decoder_layers_5_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1389592960)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_6_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406370240)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_6_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406374400)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406378560)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1406382720)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410577088)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410581248)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414775616)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414779776)))]; |
| tensor<fp32, [1024]> decoder_layers_6_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1418974144)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1418978304)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423172672)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423176832)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423180992)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1423185152)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1427379520)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1427383680)))]; |
| tensor<fp32, [1024]> decoder_layers_6_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_6_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1431578048)))]; |
| tensor<fp32, [1024]> decoder_layers_6_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_6_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1431582208)))]; |
| tensor<fp32, [4096]> decoder_layers_6_fc1_bias = const()[name = tensor<string, []>("decoder_layers_6_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1431586368)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_6_fc1_weight = const()[name = tensor<string, []>("decoder_layers_6_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1431602816)))]; |
| tensor<fp32, [1024]> decoder_layers_6_fc2_bias = const()[name = tensor<string, []>("decoder_layers_6_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448380096)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_6_fc2_weight = const()[name = tensor<string, []>("decoder_layers_6_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1448384256)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_7_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465161536)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_7_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465165696)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465169856)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1465174016)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1469368384)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1469372544)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473566912)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1473571072)))]; |
| tensor<fp32, [1024]> decoder_layers_7_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1477765440)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1477769600)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481963968)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481968128)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481972288)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481976448)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1486170816)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1486174976)))]; |
| tensor<fp32, [1024]> decoder_layers_7_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_7_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490369344)))]; |
| tensor<fp32, [1024]> decoder_layers_7_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_7_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490373504)))]; |
| tensor<fp32, [4096]> decoder_layers_7_fc1_bias = const()[name = tensor<string, []>("decoder_layers_7_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490377664)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_7_fc1_weight = const()[name = tensor<string, []>("decoder_layers_7_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1490394112)))]; |
| tensor<fp32, [1024]> decoder_layers_7_fc2_bias = const()[name = tensor<string, []>("decoder_layers_7_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507171392)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_7_fc2_weight = const()[name = tensor<string, []>("decoder_layers_7_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1507175552)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_8_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523952832)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_8_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523956992)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523961152)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1523965312)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1528159680)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1528163840)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1532358208)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1532362368)))]; |
| tensor<fp32, [1024]> decoder_layers_8_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1536556736)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1536560896)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540755264)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540759424)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540763584)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1540767744)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1544962112)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1544966272)))]; |
| tensor<fp32, [1024]> decoder_layers_8_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_8_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549160640)))]; |
| tensor<fp32, [1024]> decoder_layers_8_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_8_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549164800)))]; |
| tensor<fp32, [4096]> decoder_layers_8_fc1_bias = const()[name = tensor<string, []>("decoder_layers_8_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549168960)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_8_fc1_weight = const()[name = tensor<string, []>("decoder_layers_8_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549185408)))]; |
| tensor<fp32, [1024]> decoder_layers_8_fc2_bias = const()[name = tensor<string, []>("decoder_layers_8_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565962688)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_8_fc2_weight = const()[name = tensor<string, []>("decoder_layers_8_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1565966848)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_9_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582744128)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_9_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582748288)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582752448)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1582756608)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586950976)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1586955136)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1591149504)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1591153664)))]; |
| tensor<fp32, [1024]> decoder_layers_9_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1595348032)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1595352192)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1599546560)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1599550720)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1599554880)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1599559040)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1603753408)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1603757568)))]; |
| tensor<fp32, [1024]> decoder_layers_9_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_9_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607951936)))]; |
| tensor<fp32, [1024]> decoder_layers_9_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_9_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607956096)))]; |
| tensor<fp32, [4096]> decoder_layers_9_fc1_bias = const()[name = tensor<string, []>("decoder_layers_9_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607960256)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_9_fc1_weight = const()[name = tensor<string, []>("decoder_layers_9_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1607976704)))]; |
| tensor<fp32, [1024]> decoder_layers_9_fc2_bias = const()[name = tensor<string, []>("decoder_layers_9_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624753984)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_9_fc2_weight = const()[name = tensor<string, []>("decoder_layers_9_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1624758144)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_10_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641535424)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_10_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641539584)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641543744)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1641547904)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645742272)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1645746432)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1649940800)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1649944960)))]; |
| tensor<fp32, [1024]> decoder_layers_10_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1654139328)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1654143488)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658337856)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658342016)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658346176)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1658350336)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1662544704)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1662548864)))]; |
| tensor<fp32, [1024]> decoder_layers_10_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_10_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1666743232)))]; |
| tensor<fp32, [1024]> decoder_layers_10_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_10_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1666747392)))]; |
| tensor<fp32, [4096]> decoder_layers_10_fc1_bias = const()[name = tensor<string, []>("decoder_layers_10_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1666751552)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_10_fc1_weight = const()[name = tensor<string, []>("decoder_layers_10_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1666768000)))]; |
| tensor<fp32, [1024]> decoder_layers_10_fc2_bias = const()[name = tensor<string, []>("decoder_layers_10_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1683545280)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_10_fc2_weight = const()[name = tensor<string, []>("decoder_layers_10_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1683549440)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_11_self_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1700326720)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_11_self_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1700330880)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_self_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1700335040)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_self_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_self_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1700339200)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_self_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704533568)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_self_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_self_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1704537728)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_self_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1708732096)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_self_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_self_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1708736256)))]; |
| tensor<fp32, [1024]> decoder_layers_11_self_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_self_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712930624)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_self_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_self_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1712934784)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717129152)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717133312)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_q_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_q_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717137472)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_encoder_attn_q_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_q_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1717141632)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_out_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_out_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1721336000)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_encoder_attn_out_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_out_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1721340160)))]; |
| tensor<fp32, [1024]> decoder_layers_11_final_layer_norm_bias = const()[name = tensor<string, []>("decoder_layers_11_final_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725534528)))]; |
| tensor<fp32, [1024]> decoder_layers_11_final_layer_norm_weight = const()[name = tensor<string, []>("decoder_layers_11_final_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725538688)))]; |
| tensor<fp32, [4096]> decoder_layers_11_fc1_bias = const()[name = tensor<string, []>("decoder_layers_11_fc1_bias"), val = tensor<fp32, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725542848)))]; |
| tensor<fp32, [4096, 1024]> decoder_layers_11_fc1_weight = const()[name = tensor<string, []>("decoder_layers_11_fc1_weight"), val = tensor<fp32, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1725559296)))]; |
| tensor<fp32, [1024]> decoder_layers_11_fc2_bias = const()[name = tensor<string, []>("decoder_layers_11_fc2_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1742336576)))]; |
| tensor<fp32, [1024, 4096]> decoder_layers_11_fc2_weight = const()[name = tensor<string, []>("decoder_layers_11_fc2_weight"), val = tensor<fp32, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1742340736)))]; |
| tensor<fp32, [1024]> decoder_layer_norm_bias = const()[name = tensor<string, []>("decoder_layer_norm_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1759118016)))]; |
| tensor<fp32, [1024]> decoder_layer_norm_weight = const()[name = tensor<string, []>("decoder_layer_norm_weight"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1759122176)))]; |
| tensor<int32, [5]> key_states_1_begin_0 = const()[name = tensor<string, []>("key_states_1_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_1_end_0 = const()[name = tensor<string, []>("key_states_1_end_0"), val = tensor<int32, [5]>([1, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_1_end_mask_0 = const()[name = tensor<string, []>("key_states_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_1_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [12, 1, 16, ?, 64]> cast_0 = cast(dtype = cast_0_dtype_0, x = past_self_key)[name = tensor<string, []>("cast_90")]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_1 = slice_by_index(begin = key_states_1_begin_0, end = key_states_1_end_0, end_mask = key_states_1_end_mask_0, squeeze_mask = key_states_1_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_1")]; |
| tensor<int32, [5]> value_states_1_begin_0 = const()[name = tensor<string, []>("value_states_1_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_1_end_0 = const()[name = tensor<string, []>("value_states_1_end_0"), val = tensor<int32, [5]>([1, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_1_end_mask_0 = const()[name = tensor<string, []>("value_states_1_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_1_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_1_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [12, 1, 16, ?, 64]> cast_1 = cast(dtype = cast_1_dtype_0, x = past_self_value)[name = tensor<string, []>("cast_89")]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_1 = slice_by_index(begin = value_states_1_begin_0, end = value_states_1_end_0, end_mask = value_states_1_end_mask_0, squeeze_mask = value_states_1_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_1")]; |
| tensor<int32, [5]> key_states_3_begin_0 = const()[name = tensor<string, []>("key_states_3_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_3_end_0 = const()[name = tensor<string, []>("key_states_3_end_0"), val = tensor<int32, [5]>([1, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_3_end_mask_0 = const()[name = tensor<string, []>("key_states_3_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_3_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_3_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [12, 1, 16, ?, 64]> cast_2 = cast(dtype = cast_2_dtype_0, x = past_cross_key)[name = tensor<string, []>("cast_88")]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_3 = slice_by_index(begin = key_states_3_begin_0, end = key_states_3_end_0, end_mask = key_states_3_end_mask_0, squeeze_mask = key_states_3_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_3")]; |
| tensor<int32, [5]> value_states_3_begin_0 = const()[name = tensor<string, []>("value_states_3_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_3_end_0 = const()[name = tensor<string, []>("value_states_3_end_0"), val = tensor<int32, [5]>([1, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_3_end_mask_0 = const()[name = tensor<string, []>("value_states_3_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_3_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_3_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [12, 1, 16, ?, 64]> cast_3 = cast(dtype = cast_3_dtype_0, x = past_cross_value)[name = tensor<string, []>("cast_87")]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_3 = slice_by_index(begin = value_states_3_begin_0, end = value_states_3_end_0, end_mask = value_states_3_end_mask_0, squeeze_mask = value_states_3_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_3")]; |
| tensor<int32, [5]> key_states_5_begin_0 = const()[name = tensor<string, []>("key_states_5_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_5_end_0 = const()[name = tensor<string, []>("key_states_5_end_0"), val = tensor<int32, [5]>([2, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_5_end_mask_0 = const()[name = tensor<string, []>("key_states_5_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_5_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_5_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_5 = slice_by_index(begin = key_states_5_begin_0, end = key_states_5_end_0, end_mask = key_states_5_end_mask_0, squeeze_mask = key_states_5_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_5")]; |
| tensor<int32, [5]> value_states_5_begin_0 = const()[name = tensor<string, []>("value_states_5_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_5_end_0 = const()[name = tensor<string, []>("value_states_5_end_0"), val = tensor<int32, [5]>([2, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_5_end_mask_0 = const()[name = tensor<string, []>("value_states_5_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_5_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_5_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_5 = slice_by_index(begin = value_states_5_begin_0, end = value_states_5_end_0, end_mask = value_states_5_end_mask_0, squeeze_mask = value_states_5_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_5")]; |
| tensor<int32, [5]> key_states_7_begin_0 = const()[name = tensor<string, []>("key_states_7_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_7_end_0 = const()[name = tensor<string, []>("key_states_7_end_0"), val = tensor<int32, [5]>([2, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_7_end_mask_0 = const()[name = tensor<string, []>("key_states_7_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_7_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_7_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_7 = slice_by_index(begin = key_states_7_begin_0, end = key_states_7_end_0, end_mask = key_states_7_end_mask_0, squeeze_mask = key_states_7_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_7")]; |
| tensor<int32, [5]> value_states_7_begin_0 = const()[name = tensor<string, []>("value_states_7_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_7_end_0 = const()[name = tensor<string, []>("value_states_7_end_0"), val = tensor<int32, [5]>([2, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_7_end_mask_0 = const()[name = tensor<string, []>("value_states_7_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_7_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_7_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_7 = slice_by_index(begin = value_states_7_begin_0, end = value_states_7_end_0, end_mask = value_states_7_end_mask_0, squeeze_mask = value_states_7_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_7")]; |
| tensor<int32, [5]> key_states_9_begin_0 = const()[name = tensor<string, []>("key_states_9_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_9_end_0 = const()[name = tensor<string, []>("key_states_9_end_0"), val = tensor<int32, [5]>([3, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_9_end_mask_0 = const()[name = tensor<string, []>("key_states_9_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_9_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_9_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_9 = slice_by_index(begin = key_states_9_begin_0, end = key_states_9_end_0, end_mask = key_states_9_end_mask_0, squeeze_mask = key_states_9_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_9")]; |
| tensor<int32, [5]> value_states_9_begin_0 = const()[name = tensor<string, []>("value_states_9_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_9_end_0 = const()[name = tensor<string, []>("value_states_9_end_0"), val = tensor<int32, [5]>([3, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_9_end_mask_0 = const()[name = tensor<string, []>("value_states_9_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_9_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_9_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_9 = slice_by_index(begin = value_states_9_begin_0, end = value_states_9_end_0, end_mask = value_states_9_end_mask_0, squeeze_mask = value_states_9_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_9")]; |
| tensor<int32, [5]> key_states_11_begin_0 = const()[name = tensor<string, []>("key_states_11_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_11_end_0 = const()[name = tensor<string, []>("key_states_11_end_0"), val = tensor<int32, [5]>([3, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_11_end_mask_0 = const()[name = tensor<string, []>("key_states_11_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_11_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_11_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_11 = slice_by_index(begin = key_states_11_begin_0, end = key_states_11_end_0, end_mask = key_states_11_end_mask_0, squeeze_mask = key_states_11_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_11")]; |
| tensor<int32, [5]> value_states_11_begin_0 = const()[name = tensor<string, []>("value_states_11_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_11_end_0 = const()[name = tensor<string, []>("value_states_11_end_0"), val = tensor<int32, [5]>([3, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_11_end_mask_0 = const()[name = tensor<string, []>("value_states_11_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_11_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_11_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_11 = slice_by_index(begin = value_states_11_begin_0, end = value_states_11_end_0, end_mask = value_states_11_end_mask_0, squeeze_mask = value_states_11_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_11")]; |
| tensor<int32, [5]> key_states_13_begin_0 = const()[name = tensor<string, []>("key_states_13_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_13_end_0 = const()[name = tensor<string, []>("key_states_13_end_0"), val = tensor<int32, [5]>([4, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_13_end_mask_0 = const()[name = tensor<string, []>("key_states_13_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_13_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_13_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_13 = slice_by_index(begin = key_states_13_begin_0, end = key_states_13_end_0, end_mask = key_states_13_end_mask_0, squeeze_mask = key_states_13_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_13")]; |
| tensor<int32, [5]> value_states_13_begin_0 = const()[name = tensor<string, []>("value_states_13_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_13_end_0 = const()[name = tensor<string, []>("value_states_13_end_0"), val = tensor<int32, [5]>([4, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_13_end_mask_0 = const()[name = tensor<string, []>("value_states_13_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_13_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_13_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_13 = slice_by_index(begin = value_states_13_begin_0, end = value_states_13_end_0, end_mask = value_states_13_end_mask_0, squeeze_mask = value_states_13_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_13")]; |
| tensor<int32, [5]> key_states_15_begin_0 = const()[name = tensor<string, []>("key_states_15_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_15_end_0 = const()[name = tensor<string, []>("key_states_15_end_0"), val = tensor<int32, [5]>([4, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_15_end_mask_0 = const()[name = tensor<string, []>("key_states_15_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_15_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_15_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_15 = slice_by_index(begin = key_states_15_begin_0, end = key_states_15_end_0, end_mask = key_states_15_end_mask_0, squeeze_mask = key_states_15_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_15")]; |
| tensor<int32, [5]> value_states_15_begin_0 = const()[name = tensor<string, []>("value_states_15_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_15_end_0 = const()[name = tensor<string, []>("value_states_15_end_0"), val = tensor<int32, [5]>([4, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_15_end_mask_0 = const()[name = tensor<string, []>("value_states_15_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_15_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_15_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_15 = slice_by_index(begin = value_states_15_begin_0, end = value_states_15_end_0, end_mask = value_states_15_end_mask_0, squeeze_mask = value_states_15_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_15")]; |
| tensor<int32, [5]> key_states_17_begin_0 = const()[name = tensor<string, []>("key_states_17_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_17_end_0 = const()[name = tensor<string, []>("key_states_17_end_0"), val = tensor<int32, [5]>([5, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_17_end_mask_0 = const()[name = tensor<string, []>("key_states_17_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_17_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_17_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_17 = slice_by_index(begin = key_states_17_begin_0, end = key_states_17_end_0, end_mask = key_states_17_end_mask_0, squeeze_mask = key_states_17_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_17")]; |
| tensor<int32, [5]> value_states_17_begin_0 = const()[name = tensor<string, []>("value_states_17_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_17_end_0 = const()[name = tensor<string, []>("value_states_17_end_0"), val = tensor<int32, [5]>([5, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_17_end_mask_0 = const()[name = tensor<string, []>("value_states_17_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_17_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_17_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_17 = slice_by_index(begin = value_states_17_begin_0, end = value_states_17_end_0, end_mask = value_states_17_end_mask_0, squeeze_mask = value_states_17_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_17")]; |
| tensor<int32, [5]> key_states_19_begin_0 = const()[name = tensor<string, []>("key_states_19_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_19_end_0 = const()[name = tensor<string, []>("key_states_19_end_0"), val = tensor<int32, [5]>([5, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_19_end_mask_0 = const()[name = tensor<string, []>("key_states_19_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_19_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_19_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_19 = slice_by_index(begin = key_states_19_begin_0, end = key_states_19_end_0, end_mask = key_states_19_end_mask_0, squeeze_mask = key_states_19_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_19")]; |
| tensor<int32, [5]> value_states_19_begin_0 = const()[name = tensor<string, []>("value_states_19_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_19_end_0 = const()[name = tensor<string, []>("value_states_19_end_0"), val = tensor<int32, [5]>([5, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_19_end_mask_0 = const()[name = tensor<string, []>("value_states_19_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_19_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_19_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_19 = slice_by_index(begin = value_states_19_begin_0, end = value_states_19_end_0, end_mask = value_states_19_end_mask_0, squeeze_mask = value_states_19_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_19")]; |
| tensor<int32, [5]> key_states_21_begin_0 = const()[name = tensor<string, []>("key_states_21_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_21_end_0 = const()[name = tensor<string, []>("key_states_21_end_0"), val = tensor<int32, [5]>([6, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_21_end_mask_0 = const()[name = tensor<string, []>("key_states_21_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_21_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_21_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_21 = slice_by_index(begin = key_states_21_begin_0, end = key_states_21_end_0, end_mask = key_states_21_end_mask_0, squeeze_mask = key_states_21_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_21")]; |
| tensor<int32, [5]> value_states_21_begin_0 = const()[name = tensor<string, []>("value_states_21_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_21_end_0 = const()[name = tensor<string, []>("value_states_21_end_0"), val = tensor<int32, [5]>([6, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_21_end_mask_0 = const()[name = tensor<string, []>("value_states_21_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_21_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_21_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_21 = slice_by_index(begin = value_states_21_begin_0, end = value_states_21_end_0, end_mask = value_states_21_end_mask_0, squeeze_mask = value_states_21_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_21")]; |
| tensor<int32, [5]> key_states_23_begin_0 = const()[name = tensor<string, []>("key_states_23_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_23_end_0 = const()[name = tensor<string, []>("key_states_23_end_0"), val = tensor<int32, [5]>([6, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_23_end_mask_0 = const()[name = tensor<string, []>("key_states_23_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_23_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_23_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_23 = slice_by_index(begin = key_states_23_begin_0, end = key_states_23_end_0, end_mask = key_states_23_end_mask_0, squeeze_mask = key_states_23_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_23")]; |
| tensor<int32, [5]> value_states_23_begin_0 = const()[name = tensor<string, []>("value_states_23_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_23_end_0 = const()[name = tensor<string, []>("value_states_23_end_0"), val = tensor<int32, [5]>([6, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_23_end_mask_0 = const()[name = tensor<string, []>("value_states_23_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_23_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_23_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_23 = slice_by_index(begin = value_states_23_begin_0, end = value_states_23_end_0, end_mask = value_states_23_end_mask_0, squeeze_mask = value_states_23_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_23")]; |
| tensor<int32, [5]> key_states_25_begin_0 = const()[name = tensor<string, []>("key_states_25_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_25_end_0 = const()[name = tensor<string, []>("key_states_25_end_0"), val = tensor<int32, [5]>([7, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_25_end_mask_0 = const()[name = tensor<string, []>("key_states_25_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_25_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_25_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_25 = slice_by_index(begin = key_states_25_begin_0, end = key_states_25_end_0, end_mask = key_states_25_end_mask_0, squeeze_mask = key_states_25_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_25")]; |
| tensor<int32, [5]> value_states_25_begin_0 = const()[name = tensor<string, []>("value_states_25_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_25_end_0 = const()[name = tensor<string, []>("value_states_25_end_0"), val = tensor<int32, [5]>([7, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_25_end_mask_0 = const()[name = tensor<string, []>("value_states_25_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_25_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_25_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_25 = slice_by_index(begin = value_states_25_begin_0, end = value_states_25_end_0, end_mask = value_states_25_end_mask_0, squeeze_mask = value_states_25_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_25")]; |
| tensor<int32, [5]> key_states_27_begin_0 = const()[name = tensor<string, []>("key_states_27_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_27_end_0 = const()[name = tensor<string, []>("key_states_27_end_0"), val = tensor<int32, [5]>([7, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_27_end_mask_0 = const()[name = tensor<string, []>("key_states_27_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_27_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_27_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_27 = slice_by_index(begin = key_states_27_begin_0, end = key_states_27_end_0, end_mask = key_states_27_end_mask_0, squeeze_mask = key_states_27_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_27")]; |
| tensor<int32, [5]> value_states_27_begin_0 = const()[name = tensor<string, []>("value_states_27_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_27_end_0 = const()[name = tensor<string, []>("value_states_27_end_0"), val = tensor<int32, [5]>([7, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_27_end_mask_0 = const()[name = tensor<string, []>("value_states_27_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_27_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_27_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_27 = slice_by_index(begin = value_states_27_begin_0, end = value_states_27_end_0, end_mask = value_states_27_end_mask_0, squeeze_mask = value_states_27_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_27")]; |
| tensor<int32, [5]> key_states_29_begin_0 = const()[name = tensor<string, []>("key_states_29_begin_0"), val = tensor<int32, [5]>([7, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_29_end_0 = const()[name = tensor<string, []>("key_states_29_end_0"), val = tensor<int32, [5]>([8, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_29_end_mask_0 = const()[name = tensor<string, []>("key_states_29_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_29_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_29_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_29 = slice_by_index(begin = key_states_29_begin_0, end = key_states_29_end_0, end_mask = key_states_29_end_mask_0, squeeze_mask = key_states_29_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_29")]; |
| tensor<int32, [5]> value_states_29_begin_0 = const()[name = tensor<string, []>("value_states_29_begin_0"), val = tensor<int32, [5]>([7, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_29_end_0 = const()[name = tensor<string, []>("value_states_29_end_0"), val = tensor<int32, [5]>([8, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_29_end_mask_0 = const()[name = tensor<string, []>("value_states_29_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_29_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_29_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_29 = slice_by_index(begin = value_states_29_begin_0, end = value_states_29_end_0, end_mask = value_states_29_end_mask_0, squeeze_mask = value_states_29_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_29")]; |
| tensor<int32, [5]> key_states_31_begin_0 = const()[name = tensor<string, []>("key_states_31_begin_0"), val = tensor<int32, [5]>([7, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_31_end_0 = const()[name = tensor<string, []>("key_states_31_end_0"), val = tensor<int32, [5]>([8, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_31_end_mask_0 = const()[name = tensor<string, []>("key_states_31_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_31_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_31_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_31 = slice_by_index(begin = key_states_31_begin_0, end = key_states_31_end_0, end_mask = key_states_31_end_mask_0, squeeze_mask = key_states_31_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_31")]; |
| tensor<int32, [5]> value_states_31_begin_0 = const()[name = tensor<string, []>("value_states_31_begin_0"), val = tensor<int32, [5]>([7, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_31_end_0 = const()[name = tensor<string, []>("value_states_31_end_0"), val = tensor<int32, [5]>([8, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_31_end_mask_0 = const()[name = tensor<string, []>("value_states_31_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_31_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_31_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_31 = slice_by_index(begin = value_states_31_begin_0, end = value_states_31_end_0, end_mask = value_states_31_end_mask_0, squeeze_mask = value_states_31_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_31")]; |
| tensor<int32, [5]> key_states_33_begin_0 = const()[name = tensor<string, []>("key_states_33_begin_0"), val = tensor<int32, [5]>([8, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_33_end_0 = const()[name = tensor<string, []>("key_states_33_end_0"), val = tensor<int32, [5]>([9, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_33_end_mask_0 = const()[name = tensor<string, []>("key_states_33_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_33_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_33_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_33 = slice_by_index(begin = key_states_33_begin_0, end = key_states_33_end_0, end_mask = key_states_33_end_mask_0, squeeze_mask = key_states_33_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_33")]; |
| tensor<int32, [5]> value_states_33_begin_0 = const()[name = tensor<string, []>("value_states_33_begin_0"), val = tensor<int32, [5]>([8, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_33_end_0 = const()[name = tensor<string, []>("value_states_33_end_0"), val = tensor<int32, [5]>([9, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_33_end_mask_0 = const()[name = tensor<string, []>("value_states_33_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_33_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_33_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_33 = slice_by_index(begin = value_states_33_begin_0, end = value_states_33_end_0, end_mask = value_states_33_end_mask_0, squeeze_mask = value_states_33_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_33")]; |
| tensor<int32, [5]> key_states_35_begin_0 = const()[name = tensor<string, []>("key_states_35_begin_0"), val = tensor<int32, [5]>([8, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_35_end_0 = const()[name = tensor<string, []>("key_states_35_end_0"), val = tensor<int32, [5]>([9, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_35_end_mask_0 = const()[name = tensor<string, []>("key_states_35_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_35_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_35_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_35 = slice_by_index(begin = key_states_35_begin_0, end = key_states_35_end_0, end_mask = key_states_35_end_mask_0, squeeze_mask = key_states_35_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_35")]; |
| tensor<int32, [5]> value_states_35_begin_0 = const()[name = tensor<string, []>("value_states_35_begin_0"), val = tensor<int32, [5]>([8, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_35_end_0 = const()[name = tensor<string, []>("value_states_35_end_0"), val = tensor<int32, [5]>([9, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_35_end_mask_0 = const()[name = tensor<string, []>("value_states_35_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_35_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_35_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_35 = slice_by_index(begin = value_states_35_begin_0, end = value_states_35_end_0, end_mask = value_states_35_end_mask_0, squeeze_mask = value_states_35_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_35")]; |
| tensor<int32, [5]> key_states_37_begin_0 = const()[name = tensor<string, []>("key_states_37_begin_0"), val = tensor<int32, [5]>([9, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_37_end_0 = const()[name = tensor<string, []>("key_states_37_end_0"), val = tensor<int32, [5]>([10, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_37_end_mask_0 = const()[name = tensor<string, []>("key_states_37_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_37_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_37_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_37 = slice_by_index(begin = key_states_37_begin_0, end = key_states_37_end_0, end_mask = key_states_37_end_mask_0, squeeze_mask = key_states_37_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_37")]; |
| tensor<int32, [5]> value_states_37_begin_0 = const()[name = tensor<string, []>("value_states_37_begin_0"), val = tensor<int32, [5]>([9, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_37_end_0 = const()[name = tensor<string, []>("value_states_37_end_0"), val = tensor<int32, [5]>([10, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_37_end_mask_0 = const()[name = tensor<string, []>("value_states_37_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_37_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_37_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_37 = slice_by_index(begin = value_states_37_begin_0, end = value_states_37_end_0, end_mask = value_states_37_end_mask_0, squeeze_mask = value_states_37_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_37")]; |
| tensor<int32, [5]> key_states_39_begin_0 = const()[name = tensor<string, []>("key_states_39_begin_0"), val = tensor<int32, [5]>([9, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_39_end_0 = const()[name = tensor<string, []>("key_states_39_end_0"), val = tensor<int32, [5]>([10, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_39_end_mask_0 = const()[name = tensor<string, []>("key_states_39_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_39_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_39_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_39 = slice_by_index(begin = key_states_39_begin_0, end = key_states_39_end_0, end_mask = key_states_39_end_mask_0, squeeze_mask = key_states_39_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_39")]; |
| tensor<int32, [5]> value_states_39_begin_0 = const()[name = tensor<string, []>("value_states_39_begin_0"), val = tensor<int32, [5]>([9, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_39_end_0 = const()[name = tensor<string, []>("value_states_39_end_0"), val = tensor<int32, [5]>([10, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_39_end_mask_0 = const()[name = tensor<string, []>("value_states_39_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_39_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_39_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_39 = slice_by_index(begin = value_states_39_begin_0, end = value_states_39_end_0, end_mask = value_states_39_end_mask_0, squeeze_mask = value_states_39_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_39")]; |
| tensor<int32, [5]> key_states_41_begin_0 = const()[name = tensor<string, []>("key_states_41_begin_0"), val = tensor<int32, [5]>([10, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_41_end_0 = const()[name = tensor<string, []>("key_states_41_end_0"), val = tensor<int32, [5]>([11, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_41_end_mask_0 = const()[name = tensor<string, []>("key_states_41_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_41_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_41_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_41 = slice_by_index(begin = key_states_41_begin_0, end = key_states_41_end_0, end_mask = key_states_41_end_mask_0, squeeze_mask = key_states_41_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_41")]; |
| tensor<int32, [5]> value_states_41_begin_0 = const()[name = tensor<string, []>("value_states_41_begin_0"), val = tensor<int32, [5]>([10, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_41_end_0 = const()[name = tensor<string, []>("value_states_41_end_0"), val = tensor<int32, [5]>([11, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_41_end_mask_0 = const()[name = tensor<string, []>("value_states_41_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_41_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_41_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_41 = slice_by_index(begin = value_states_41_begin_0, end = value_states_41_end_0, end_mask = value_states_41_end_mask_0, squeeze_mask = value_states_41_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_41")]; |
| tensor<int32, [5]> key_states_43_begin_0 = const()[name = tensor<string, []>("key_states_43_begin_0"), val = tensor<int32, [5]>([10, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_43_end_0 = const()[name = tensor<string, []>("key_states_43_end_0"), val = tensor<int32, [5]>([11, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_43_end_mask_0 = const()[name = tensor<string, []>("key_states_43_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_43_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_43_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_43 = slice_by_index(begin = key_states_43_begin_0, end = key_states_43_end_0, end_mask = key_states_43_end_mask_0, squeeze_mask = key_states_43_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_43")]; |
| tensor<int32, [5]> value_states_43_begin_0 = const()[name = tensor<string, []>("value_states_43_begin_0"), val = tensor<int32, [5]>([10, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_43_end_0 = const()[name = tensor<string, []>("value_states_43_end_0"), val = tensor<int32, [5]>([11, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_43_end_mask_0 = const()[name = tensor<string, []>("value_states_43_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_43_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_43_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_43 = slice_by_index(begin = value_states_43_begin_0, end = value_states_43_end_0, end_mask = value_states_43_end_mask_0, squeeze_mask = value_states_43_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_43")]; |
| tensor<int32, [5]> key_states_45_begin_0 = const()[name = tensor<string, []>("key_states_45_begin_0"), val = tensor<int32, [5]>([11, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_45_end_0 = const()[name = tensor<string, []>("key_states_45_end_0"), val = tensor<int32, [5]>([12, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_45_end_mask_0 = const()[name = tensor<string, []>("key_states_45_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_45_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_45_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_45 = slice_by_index(begin = key_states_45_begin_0, end = key_states_45_end_0, end_mask = key_states_45_end_mask_0, squeeze_mask = key_states_45_squeeze_mask_0, x = cast_0)[name = tensor<string, []>("key_states_45")]; |
| tensor<int32, [5]> value_states_45_begin_0 = const()[name = tensor<string, []>("value_states_45_begin_0"), val = tensor<int32, [5]>([11, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_45_end_0 = const()[name = tensor<string, []>("value_states_45_end_0"), val = tensor<int32, [5]>([12, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_45_end_mask_0 = const()[name = tensor<string, []>("value_states_45_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_45_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_45_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_45 = slice_by_index(begin = value_states_45_begin_0, end = value_states_45_end_0, end_mask = value_states_45_end_mask_0, squeeze_mask = value_states_45_squeeze_mask_0, x = cast_1)[name = tensor<string, []>("value_states_45")]; |
| tensor<int32, [5]> key_states_47_begin_0 = const()[name = tensor<string, []>("key_states_47_begin_0"), val = tensor<int32, [5]>([11, 0, 0, 0, 0])]; |
| tensor<int32, [5]> key_states_47_end_0 = const()[name = tensor<string, []>("key_states_47_end_0"), val = tensor<int32, [5]>([12, 1, 16, 0, 64])]; |
| tensor<bool, [5]> key_states_47_end_mask_0 = const()[name = tensor<string, []>("key_states_47_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> key_states_47_squeeze_mask_0 = const()[name = tensor<string, []>("key_states_47_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_47 = slice_by_index(begin = key_states_47_begin_0, end = key_states_47_end_0, end_mask = key_states_47_end_mask_0, squeeze_mask = key_states_47_squeeze_mask_0, x = cast_2)[name = tensor<string, []>("key_states_47")]; |
| tensor<int32, [5]> value_states_47_begin_0 = const()[name = tensor<string, []>("value_states_47_begin_0"), val = tensor<int32, [5]>([11, 0, 0, 0, 0])]; |
| tensor<int32, [5]> value_states_47_end_0 = const()[name = tensor<string, []>("value_states_47_end_0"), val = tensor<int32, [5]>([12, 1, 16, 0, 64])]; |
| tensor<bool, [5]> value_states_47_end_mask_0 = const()[name = tensor<string, []>("value_states_47_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])]; |
| tensor<bool, [5]> value_states_47_squeeze_mask_0 = const()[name = tensor<string, []>("value_states_47_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_47 = slice_by_index(begin = value_states_47_begin_0, end = value_states_47_end_0, end_mask = value_states_47_end_mask_0, squeeze_mask = value_states_47_squeeze_mask_0, x = cast_3)[name = tensor<string, []>("value_states_47")]; |
| tensor<int32, []> var_173 = const()[name = tensor<string, []>("op_173"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_174_interleave_0 = const()[name = tensor<string, []>("op_174_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_174 = concat(axis = var_173, interleave = var_174_interleave_0, values = key_states_1)[name = tensor<string, []>("op_174")]; |
| tensor<int32, []> var_176 = const()[name = tensor<string, []>("op_176"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_177_interleave_0 = const()[name = tensor<string, []>("op_177_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_177 = concat(axis = var_176, interleave = var_177_interleave_0, values = value_states_1)[name = tensor<string, []>("op_177")]; |
| tensor<int32, []> var_195 = const()[name = tensor<string, []>("op_195"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_3_interleave_0 = const()[name = tensor<string, []>("key_3_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_3 = concat(axis = var_195, interleave = key_3_interleave_0, values = key_states_3)[name = tensor<string, []>("key_3")]; |
| tensor<int32, []> var_198 = const()[name = tensor<string, []>("op_198"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_3_interleave_0 = const()[name = tensor<string, []>("value_3_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_3 = concat(axis = var_198, interleave = value_3_interleave_0, values = value_states_3)[name = tensor<string, []>("value_3")]; |
| tensor<int32, []> var_217 = const()[name = tensor<string, []>("op_217"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_218_interleave_0 = const()[name = tensor<string, []>("op_218_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_218 = concat(axis = var_217, interleave = var_218_interleave_0, values = key_states_5)[name = tensor<string, []>("op_218")]; |
| tensor<int32, []> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_221_interleave_0 = const()[name = tensor<string, []>("op_221_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_221 = concat(axis = var_220, interleave = var_221_interleave_0, values = value_states_5)[name = tensor<string, []>("op_221")]; |
| tensor<int32, []> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_7_interleave_0 = const()[name = tensor<string, []>("key_7_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_7 = concat(axis = var_239, interleave = key_7_interleave_0, values = key_states_7)[name = tensor<string, []>("key_7")]; |
| tensor<int32, []> var_242 = const()[name = tensor<string, []>("op_242"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_7_interleave_0 = const()[name = tensor<string, []>("value_7_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_7 = concat(axis = var_242, interleave = value_7_interleave_0, values = value_states_7)[name = tensor<string, []>("value_7")]; |
| tensor<int32, []> var_261 = const()[name = tensor<string, []>("op_261"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_262_interleave_0 = const()[name = tensor<string, []>("op_262_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_262 = concat(axis = var_261, interleave = var_262_interleave_0, values = key_states_9)[name = tensor<string, []>("op_262")]; |
| tensor<int32, []> var_264 = const()[name = tensor<string, []>("op_264"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_265_interleave_0 = const()[name = tensor<string, []>("op_265_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_265 = concat(axis = var_264, interleave = var_265_interleave_0, values = value_states_9)[name = tensor<string, []>("op_265")]; |
| tensor<int32, []> var_283 = const()[name = tensor<string, []>("op_283"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_11_interleave_0 = const()[name = tensor<string, []>("key_11_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_11 = concat(axis = var_283, interleave = key_11_interleave_0, values = key_states_11)[name = tensor<string, []>("key_11")]; |
| tensor<int32, []> var_286 = const()[name = tensor<string, []>("op_286"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_11_interleave_0 = const()[name = tensor<string, []>("value_11_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_11 = concat(axis = var_286, interleave = value_11_interleave_0, values = value_states_11)[name = tensor<string, []>("value_11")]; |
| tensor<int32, []> var_305 = const()[name = tensor<string, []>("op_305"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_306_interleave_0 = const()[name = tensor<string, []>("op_306_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_306 = concat(axis = var_305, interleave = var_306_interleave_0, values = key_states_13)[name = tensor<string, []>("op_306")]; |
| tensor<int32, []> var_308 = const()[name = tensor<string, []>("op_308"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_309_interleave_0 = const()[name = tensor<string, []>("op_309_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_309 = concat(axis = var_308, interleave = var_309_interleave_0, values = value_states_13)[name = tensor<string, []>("op_309")]; |
| tensor<int32, []> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_15_interleave_0 = const()[name = tensor<string, []>("key_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_15 = concat(axis = var_327, interleave = key_15_interleave_0, values = key_states_15)[name = tensor<string, []>("key_15")]; |
| tensor<int32, []> var_330 = const()[name = tensor<string, []>("op_330"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_15_interleave_0 = const()[name = tensor<string, []>("value_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_15 = concat(axis = var_330, interleave = value_15_interleave_0, values = value_states_15)[name = tensor<string, []>("value_15")]; |
| tensor<int32, []> var_349 = const()[name = tensor<string, []>("op_349"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_350_interleave_0 = const()[name = tensor<string, []>("op_350_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_350 = concat(axis = var_349, interleave = var_350_interleave_0, values = key_states_17)[name = tensor<string, []>("op_350")]; |
| tensor<int32, []> var_352 = const()[name = tensor<string, []>("op_352"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_353_interleave_0 = const()[name = tensor<string, []>("op_353_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_353 = concat(axis = var_352, interleave = var_353_interleave_0, values = value_states_17)[name = tensor<string, []>("op_353")]; |
| tensor<int32, []> var_371 = const()[name = tensor<string, []>("op_371"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_19_interleave_0 = const()[name = tensor<string, []>("key_19_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_19 = concat(axis = var_371, interleave = key_19_interleave_0, values = key_states_19)[name = tensor<string, []>("key_19")]; |
| tensor<int32, []> var_374 = const()[name = tensor<string, []>("op_374"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_19_interleave_0 = const()[name = tensor<string, []>("value_19_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_19 = concat(axis = var_374, interleave = value_19_interleave_0, values = value_states_19)[name = tensor<string, []>("value_19")]; |
| tensor<int32, []> var_393 = const()[name = tensor<string, []>("op_393"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_394_interleave_0 = const()[name = tensor<string, []>("op_394_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_394 = concat(axis = var_393, interleave = var_394_interleave_0, values = key_states_21)[name = tensor<string, []>("op_394")]; |
| tensor<int32, []> var_396 = const()[name = tensor<string, []>("op_396"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_397_interleave_0 = const()[name = tensor<string, []>("op_397_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_397 = concat(axis = var_396, interleave = var_397_interleave_0, values = value_states_21)[name = tensor<string, []>("op_397")]; |
| tensor<int32, []> var_415 = const()[name = tensor<string, []>("op_415"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_23_interleave_0 = const()[name = tensor<string, []>("key_23_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_23 = concat(axis = var_415, interleave = key_23_interleave_0, values = key_states_23)[name = tensor<string, []>("key_23")]; |
| tensor<int32, []> var_418 = const()[name = tensor<string, []>("op_418"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_23_interleave_0 = const()[name = tensor<string, []>("value_23_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_23 = concat(axis = var_418, interleave = value_23_interleave_0, values = value_states_23)[name = tensor<string, []>("value_23")]; |
| tensor<int32, []> var_437 = const()[name = tensor<string, []>("op_437"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_438_interleave_0 = const()[name = tensor<string, []>("op_438_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_438 = concat(axis = var_437, interleave = var_438_interleave_0, values = key_states_25)[name = tensor<string, []>("op_438")]; |
| tensor<int32, []> var_440 = const()[name = tensor<string, []>("op_440"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_441_interleave_0 = const()[name = tensor<string, []>("op_441_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_441 = concat(axis = var_440, interleave = var_441_interleave_0, values = value_states_25)[name = tensor<string, []>("op_441")]; |
| tensor<int32, []> var_459 = const()[name = tensor<string, []>("op_459"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_27_interleave_0 = const()[name = tensor<string, []>("key_27_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_27 = concat(axis = var_459, interleave = key_27_interleave_0, values = key_states_27)[name = tensor<string, []>("key_27")]; |
| tensor<int32, []> var_462 = const()[name = tensor<string, []>("op_462"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_27_interleave_0 = const()[name = tensor<string, []>("value_27_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_27 = concat(axis = var_462, interleave = value_27_interleave_0, values = value_states_27)[name = tensor<string, []>("value_27")]; |
| tensor<int32, []> var_481 = const()[name = tensor<string, []>("op_481"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_482_interleave_0 = const()[name = tensor<string, []>("op_482_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_482 = concat(axis = var_481, interleave = var_482_interleave_0, values = key_states_29)[name = tensor<string, []>("op_482")]; |
| tensor<int32, []> var_484 = const()[name = tensor<string, []>("op_484"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_485_interleave_0 = const()[name = tensor<string, []>("op_485_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_485 = concat(axis = var_484, interleave = var_485_interleave_0, values = value_states_29)[name = tensor<string, []>("op_485")]; |
| tensor<int32, []> var_503 = const()[name = tensor<string, []>("op_503"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_31_interleave_0 = const()[name = tensor<string, []>("key_31_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_31 = concat(axis = var_503, interleave = key_31_interleave_0, values = key_states_31)[name = tensor<string, []>("key_31")]; |
| tensor<int32, []> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_31_interleave_0 = const()[name = tensor<string, []>("value_31_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_31 = concat(axis = var_506, interleave = value_31_interleave_0, values = value_states_31)[name = tensor<string, []>("value_31")]; |
| tensor<int32, []> var_525 = const()[name = tensor<string, []>("op_525"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_526_interleave_0 = const()[name = tensor<string, []>("op_526_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_526 = concat(axis = var_525, interleave = var_526_interleave_0, values = key_states_33)[name = tensor<string, []>("op_526")]; |
| tensor<int32, []> var_528 = const()[name = tensor<string, []>("op_528"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_529_interleave_0 = const()[name = tensor<string, []>("op_529_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_529 = concat(axis = var_528, interleave = var_529_interleave_0, values = value_states_33)[name = tensor<string, []>("op_529")]; |
| tensor<int32, []> var_547 = const()[name = tensor<string, []>("op_547"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_35_interleave_0 = const()[name = tensor<string, []>("key_35_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_35 = concat(axis = var_547, interleave = key_35_interleave_0, values = key_states_35)[name = tensor<string, []>("key_35")]; |
| tensor<int32, []> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_35_interleave_0 = const()[name = tensor<string, []>("value_35_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_35 = concat(axis = var_550, interleave = value_35_interleave_0, values = value_states_35)[name = tensor<string, []>("value_35")]; |
| tensor<int32, []> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_570_interleave_0 = const()[name = tensor<string, []>("op_570_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_570 = concat(axis = var_569, interleave = var_570_interleave_0, values = key_states_37)[name = tensor<string, []>("op_570")]; |
| tensor<int32, []> var_572 = const()[name = tensor<string, []>("op_572"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_573_interleave_0 = const()[name = tensor<string, []>("op_573_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_573 = concat(axis = var_572, interleave = var_573_interleave_0, values = value_states_37)[name = tensor<string, []>("op_573")]; |
| tensor<int32, []> var_591 = const()[name = tensor<string, []>("op_591"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_39_interleave_0 = const()[name = tensor<string, []>("key_39_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_39 = concat(axis = var_591, interleave = key_39_interleave_0, values = key_states_39)[name = tensor<string, []>("key_39")]; |
| tensor<int32, []> var_594 = const()[name = tensor<string, []>("op_594"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_39_interleave_0 = const()[name = tensor<string, []>("value_39_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_39 = concat(axis = var_594, interleave = value_39_interleave_0, values = value_states_39)[name = tensor<string, []>("value_39")]; |
| tensor<int32, []> var_613 = const()[name = tensor<string, []>("op_613"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_614_interleave_0 = const()[name = tensor<string, []>("op_614_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_614 = concat(axis = var_613, interleave = var_614_interleave_0, values = key_states_41)[name = tensor<string, []>("op_614")]; |
| tensor<int32, []> var_616 = const()[name = tensor<string, []>("op_616"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_617_interleave_0 = const()[name = tensor<string, []>("op_617_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_617 = concat(axis = var_616, interleave = var_617_interleave_0, values = value_states_41)[name = tensor<string, []>("op_617")]; |
| tensor<int32, []> var_635 = const()[name = tensor<string, []>("op_635"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_43_interleave_0 = const()[name = tensor<string, []>("key_43_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_43 = concat(axis = var_635, interleave = key_43_interleave_0, values = key_states_43)[name = tensor<string, []>("key_43")]; |
| tensor<int32, []> var_638 = const()[name = tensor<string, []>("op_638"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_43_interleave_0 = const()[name = tensor<string, []>("value_43_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_43 = concat(axis = var_638, interleave = value_43_interleave_0, values = value_states_43)[name = tensor<string, []>("value_43")]; |
| tensor<int32, []> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_658_interleave_0 = const()[name = tensor<string, []>("op_658_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_658 = concat(axis = var_657, interleave = var_658_interleave_0, values = key_states_45)[name = tensor<string, []>("op_658")]; |
| tensor<int32, []> var_660 = const()[name = tensor<string, []>("op_660"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> var_661_interleave_0 = const()[name = tensor<string, []>("op_661_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> var_661 = concat(axis = var_660, interleave = var_661_interleave_0, values = value_states_45)[name = tensor<string, []>("op_661")]; |
| tensor<int32, []> var_679 = const()[name = tensor<string, []>("op_679"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> key_interleave_0 = const()[name = tensor<string, []>("key_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key = concat(axis = var_679, interleave = key_interleave_0, values = key_states_47)[name = tensor<string, []>("key")]; |
| tensor<int32, []> var_682 = const()[name = tensor<string, []>("op_682"), val = tensor<int32, []>(-2)]; |
| tensor<bool, []> value_interleave_0 = const()[name = tensor<string, []>("value_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value = concat(axis = var_682, interleave = value_interleave_0, values = value_states_47)[name = tensor<string, []>("value")]; |
| tensor<fp32, []> var_685 = const()[name = tensor<string, []>("op_685"), val = tensor<fp32, []>(0x1.4f8b58p-17)]; |
| tensor<fp32, []> var_687 = const()[name = tensor<string, []>("op_687"), val = tensor<fp32, []>(0x1p-3)]; |
| tensor<int32, []> var_689 = const()[name = tensor<string, []>("op_689"), val = tensor<int32, []>(-2)]; |
| tensor<fp32, []> var_698 = const()[name = tensor<string, []>("op_698"), val = tensor<fp32, []>(-0x1.fffffep+127)]; |
| tensor<int32, []> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_705 = const()[name = tensor<string, []>("op_705"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> const_48 = const()[name = tensor<string, []>("const_48"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> var_737_axis_0 = const()[name = tensor<string, []>("op_737_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_737_batch_dims_0 = const()[name = tensor<string, []>("op_737_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [1, 1, 1024]> var_737 = gather(axis = var_737_axis_0, batch_dims = var_737_batch_dims_0, indices = input_ids, x = decoder_embed_tokens_weight)[name = tensor<string, []>("op_737")]; |
| tensor<fp32, []> var_738 = const()[name = tensor<string, []>("op_738"), val = tensor<fp32, []>(0x1p+5)]; |
| tensor<fp32, [1, 1, 1024]> inputs_embeds = mul(x = var_737, y = var_738)[name = tensor<string, []>("inputs_embeds")]; |
| tensor<int32, []> const_49 = const()[name = tensor<string, []>("const_49"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> seq_length = const()[name = tensor<string, []>("seq_length"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [4]> var_743_shape = shape(x = var_174)[name = tensor<string, []>("op_743_shape")]; |
| tensor<int32, []> gather_0_indices_0 = const()[name = tensor<string, []>("gather_0_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_0_axis_0 = const()[name = tensor<string, []>("gather_0_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_0_batch_dims_0 = const()[name = tensor<string, []>("gather_0_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_0 = gather(axis = gather_0_axis_0, batch_dims = gather_0_batch_dims_0, indices = gather_0_indices_0, x = var_743_shape)[name = tensor<string, []>("gather_0")]; |
| tensor<int32, [1]> var_745 = add(x = gather_0, y = seq_length)[name = tensor<string, []>("op_745")]; |
| tensor<int32, []> var_746 = squeeze(x = var_745)[name = tensor<string, []>("op_746")]; |
| tensor<int32, []> const_51 = const()[name = tensor<string, []>("const_51"), val = tensor<int32, []>(1)]; |
| tensor<int32, [?]> cache_position = range_1d(end = var_746, start = gather_0, step = const_51)[name = tensor<string, []>("cache_position")]; |
| tensor<int32, []> concat_0_axis_0 = const()[name = tensor<string, []>("concat_0_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_0_interleave_0 = const()[name = tensor<string, []>("concat_0_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [2]> concat_0 = concat(axis = concat_0_axis_0, interleave = concat_0_interleave_0, values = (const_49, var_746))[name = tensor<string, []>("concat_0")]; |
| tensor<fp32, []> fill_0_value_0 = const()[name = tensor<string, []>("fill_0_value_0"), val = tensor<fp32, []>(0x1p+0)]; |
| tensor<fp32, [1, ?]> fill_0 = fill(shape = concat_0, value = fill_0_value_0)[name = tensor<string, []>("fill_0")]; |
| tensor<int32, []> const_52 = const()[name = tensor<string, []>("const_52"), val = tensor<int32, []>(1)]; |
| tensor<int32, [2]> var_753_shape = shape(x = fill_0)[name = tensor<string, []>("op_753_shape")]; |
| tensor<int32, []> gather_1_indices_0 = const()[name = tensor<string, []>("gather_1_indices_0"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> gather_1_axis_0 = const()[name = tensor<string, []>("gather_1_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_1_batch_dims_0 = const()[name = tensor<string, []>("gather_1_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_1 = gather(axis = gather_1_axis_0, batch_dims = gather_1_batch_dims_0, indices = gather_1_indices_0, x = var_753_shape)[name = tensor<string, []>("gather_1")]; |
| tensor<int32, []> concat_1_axis_0 = const()[name = tensor<string, []>("concat_1_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_1_interleave_0 = const()[name = tensor<string, []>("concat_1_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [2]> concat_1 = concat(axis = concat_1_axis_0, interleave = concat_1_interleave_0, values = (const_52, gather_1))[name = tensor<string, []>("concat_1")]; |
| tensor<fp32, []> causal_mask_1_value_0 = const()[name = tensor<string, []>("causal_mask_1_value_0"), val = tensor<fp32, []>(-0x1.fffffep+127)]; |
| tensor<fp32, [1, ?]> causal_mask_1 = fill(shape = concat_1, value = causal_mask_1_value_0)[name = tensor<string, []>("causal_mask_1")]; |
| tensor<int32, []> const_54 = const()[name = tensor<string, []>("const_54"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> const_55 = const()[name = tensor<string, []>("const_55"), val = tensor<int32, []>(1)]; |
| tensor<int32, [?]> var_757 = range_1d(end = gather_1, start = const_54, step = const_55)[name = tensor<string, []>("op_757")]; |
| tensor<int32, [2]> var_758 = const()[name = tensor<string, []>("op_758"), val = tensor<int32, [2]>([-1, 1])]; |
| tensor<int32, [?, 1]> var_759 = reshape(shape = var_758, x = cache_position)[name = tensor<string, []>("op_759")]; |
| tensor<bool, [?, ?]> var_760 = greater(x = var_757, y = var_759)[name = tensor<string, []>("op_760")]; |
| tensor<string, []> var_760_promoted_dtype_0 = const()[name = tensor<string, []>("op_760_promoted_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [?, ?]> var_760_promoted = cast(dtype = var_760_promoted_dtype_0, x = var_760)[name = tensor<string, []>("cast_86")]; |
| tensor<fp32, [?, ?]> causal_mask_3 = mul(x = causal_mask_1, y = var_760_promoted)[name = tensor<string, []>("causal_mask_3")]; |
| tensor<int32, [1]> var_762_axes_0 = const()[name = tensor<string, []>("op_762_axes_0"), val = tensor<int32, [1]>([0])]; |
| tensor<fp32, [1, ?, ?]> var_762 = expand_dims(axes = var_762_axes_0, x = causal_mask_3)[name = tensor<string, []>("op_762")]; |
| tensor<int32, [1]> var_763_axes_0 = const()[name = tensor<string, []>("op_763_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp32, [1, 1, ?, ?]> var_763 = expand_dims(axes = var_763_axes_0, x = var_762)[name = tensor<string, []>("op_763")]; |
| tensor<int32, [4]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [4]>([1, 1, -1, -1])]; |
| tensor<int32, [4]> shape_0 = shape(x = var_763)[name = tensor<string, []>("shape_0")]; |
| tensor<bool, [4]> equal_0 = const()[name = tensor<string, []>("equal_0"), val = tensor<bool, [4]>([false, false, true, true])]; |
| tensor<int32, [4]> select_0 = select(a = shape_0, b = concat_2, cond = equal_0)[name = tensor<string, []>("select_0")]; |
| tensor<int32, [4]> real_div_0 = real_div(x = select_0, y = shape_0)[name = tensor<string, []>("real_div_0")]; |
| tensor<fp32, [?, ?, ?, ?]> causal_mask_5 = tile(reps = real_div_0, x = var_763)[name = tensor<string, []>("causal_mask_5")]; |
| tensor<int32, []> concat_3_values0_0 = const()[name = tensor<string, []>("concat_3_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_3_values1_0 = const()[name = tensor<string, []>("concat_3_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_3_values2_0 = const()[name = tensor<string, []>("concat_3_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_3_axis_0 = const()[name = tensor<string, []>("concat_3_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_3_interleave_0 = const()[name = tensor<string, []>("concat_3_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_3 = concat(axis = concat_3_axis_0, interleave = concat_3_interleave_0, values = (concat_3_values0_0, concat_3_values1_0, concat_3_values2_0, gather_1))[name = tensor<string, []>("concat_3")]; |
| tensor<int32, [4]> var_773_begin_0 = const()[name = tensor<string, []>("op_773_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> var_773_end_mask_0 = const()[name = tensor<string, []>("op_773_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> var_773 = slice_by_index(begin = var_773_begin_0, end = concat_3, end_mask = var_773_end_mask_0, x = causal_mask_5)[name = tensor<string, []>("op_773")]; |
| tensor<int32, [1]> var_775_axes_0 = const()[name = tensor<string, []>("op_775_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp32, [1, 1, ?]> var_775 = expand_dims(axes = var_775_axes_0, x = fill_0)[name = tensor<string, []>("op_775")]; |
| tensor<int32, [1]> var_776_axes_0 = const()[name = tensor<string, []>("op_776_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp32, [1, 1, 1, ?]> var_776 = expand_dims(axes = var_776_axes_0, x = var_775)[name = tensor<string, []>("op_776")]; |
| tensor<fp32, [?, ?, ?, ?]> padding_mask_1 = add(x = var_773, y = var_776)[name = tensor<string, []>("padding_mask_1")]; |
| tensor<fp32, []> var_702_promoted = const()[name = tensor<string, []>("op_702_promoted"), val = tensor<fp32, []>(0x0p+0)]; |
| tensor<bool, [?, ?, ?, ?]> padding_mask = equal(x = padding_mask_1, y = var_702_promoted)[name = tensor<string, []>("padding_mask")]; |
| tensor<fp32, [?, ?, ?, ?]> var_785 = select(a = var_698, b = var_773, cond = padding_mask)[name = tensor<string, []>("op_785")]; |
| tensor<int32, [1]> expand_dims_4_axes_0 = const()[name = tensor<string, []>("expand_dims_4_axes_0"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, [1]> expand_dims_4 = expand_dims(axes = expand_dims_4_axes_0, x = gather_1)[name = tensor<string, []>("expand_dims_4")]; |
| tensor<int32, [4]> concat_6 = const()[name = tensor<string, []>("concat_6"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [1]> concat_7_values0_0 = const()[name = tensor<string, []>("concat_7_values0_0"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, [1]> concat_7_values1_0 = const()[name = tensor<string, []>("concat_7_values1_0"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, [1]> concat_7_values2_0 = const()[name = tensor<string, []>("concat_7_values2_0"), val = tensor<int32, [1]>([0])]; |
| tensor<int32, []> concat_7_axis_0 = const()[name = tensor<string, []>("concat_7_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_7_interleave_0 = const()[name = tensor<string, []>("concat_7_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_7 = concat(axis = concat_7_axis_0, interleave = concat_7_interleave_0, values = (concat_7_values0_0, concat_7_values1_0, concat_7_values2_0, expand_dims_4))[name = tensor<string, []>("concat_7")]; |
| tensor<int32, [4]> causal_mask_internal_tensor_assign_1_stride_0 = const()[name = tensor<string, []>("causal_mask_internal_tensor_assign_1_stride_0"), val = tensor<int32, [4]>([1, 1, 1, 1])]; |
| tensor<bool, [4]> causal_mask_internal_tensor_assign_1_begin_mask_0 = const()[name = tensor<string, []>("causal_mask_internal_tensor_assign_1_begin_mask_0"), val = tensor<bool, [4]>([false, false, false, false])]; |
| tensor<bool, [4]> causal_mask_internal_tensor_assign_1_end_mask_0 = const()[name = tensor<string, []>("causal_mask_internal_tensor_assign_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<bool, [4]> causal_mask_internal_tensor_assign_1_squeeze_mask_0 = const()[name = tensor<string, []>("causal_mask_internal_tensor_assign_1_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, false, false])]; |
| tensor<int32, [4]> shape_2 = shape(x = causal_mask_5)[name = tensor<string, []>("shape_2")]; |
| tensor<bool, []> reduce_prod_0_keep_dims_0 = const()[name = tensor<string, []>("reduce_prod_0_keep_dims_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> reduce_prod_0 = reduce_prod(keep_dims = reduce_prod_0_keep_dims_0, x = shape_2)[name = tensor<string, []>("reduce_prod_0")]; |
| tensor<int32, []> range_1d_0_start_0 = const()[name = tensor<string, []>("range_1d_0_start_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> range_1d_0_step_0 = const()[name = tensor<string, []>("range_1d_0_step_0"), val = tensor<int32, []>(1)]; |
| tensor<int32, [?]> range_1d_0 = range_1d(end = reduce_prod_0, start = range_1d_0_start_0, step = range_1d_0_step_0)[name = tensor<string, []>("range_1d_0")]; |
| tensor<int32, [?, ?, ?, ?]> reshape_0 = reshape(shape = shape_2, x = range_1d_0)[name = tensor<string, []>("reshape_0")]; |
| tensor<int32, [?, ?, ?, ?]> slice_by_index_0 = slice_by_index(begin = concat_6, begin_mask = causal_mask_internal_tensor_assign_1_begin_mask_0, end = concat_7, end_mask = causal_mask_internal_tensor_assign_1_end_mask_0, squeeze_mask = causal_mask_internal_tensor_assign_1_squeeze_mask_0, stride = causal_mask_internal_tensor_assign_1_stride_0, x = reshape_0)[name = tensor<string, []>("slice_by_index_0")]; |
| tensor<int32, [1]> reshape_1_shape_0 = const()[name = tensor<string, []>("reshape_1_shape_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<int32, [?]> reshape_1 = reshape(shape = reshape_1_shape_0, x = slice_by_index_0)[name = tensor<string, []>("reshape_1")]; |
| tensor<int32, [1]> reshape_2_shape_0 = const()[name = tensor<string, []>("reshape_2_shape_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [?]> reshape_2 = reshape(shape = reshape_2_shape_0, x = var_785)[name = tensor<string, []>("reshape_2")]; |
| tensor<int32, [1]> reshape_3_shape_0 = const()[name = tensor<string, []>("reshape_3_shape_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [?]> reshape_3 = reshape(shape = reshape_3_shape_0, x = causal_mask_5)[name = tensor<string, []>("reshape_3")]; |
| tensor<string, []> scatter_0_mode_0 = const()[name = tensor<string, []>("scatter_0_mode_0"), val = tensor<string, []>("update")]; |
| tensor<int32, []> scatter_0_axis_0 = const()[name = tensor<string, []>("scatter_0_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [?]> scatter_0 = scatter(axis = scatter_0_axis_0, data = reshape_3, indices = reshape_1, mode = scatter_0_mode_0, updates = reshape_2)[name = tensor<string, []>("scatter_0")]; |
| tensor<fp32, [?, ?, ?, ?]> reshape_4 = reshape(shape = shape_2, x = scatter_0)[name = tensor<string, []>("reshape_4")]; |
| tensor<int32, [2]> var_791_shape = shape(x = encoder_attention_mask)[name = tensor<string, []>("op_791_shape")]; |
| tensor<int32, []> gather_3 = const()[name = tensor<string, []>("gather_3"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> gather_4_indices_0 = const()[name = tensor<string, []>("gather_4_indices_0"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> gather_4_axis_0 = const()[name = tensor<string, []>("gather_4_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_4_batch_dims_0 = const()[name = tensor<string, []>("gather_4_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_4 = gather(axis = gather_4_axis_0, batch_dims = gather_4_batch_dims_0, indices = gather_4_indices_0, x = var_791_shape)[name = tensor<string, []>("gather_4")]; |
| tensor<int32, [1]> var_794_axes_0 = const()[name = tensor<string, []>("op_794_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1, 1, ?]> var_794 = expand_dims(axes = var_794_axes_0, x = encoder_attention_mask)[name = tensor<string, []>("op_794")]; |
| tensor<int32, [1]> var_795_axes_0 = const()[name = tensor<string, []>("op_795_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1, 1, 1, ?]> var_795 = expand_dims(axes = var_795_axes_0, x = var_794)[name = tensor<string, []>("op_795")]; |
| tensor<int32, []> concat_8_axis_0 = const()[name = tensor<string, []>("concat_8_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_8_interleave_0 = const()[name = tensor<string, []>("concat_8_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_8 = concat(axis = concat_8_axis_0, interleave = concat_8_interleave_0, values = (gather_3, var_705, const_48, gather_4))[name = tensor<string, []>("concat_8")]; |
| tensor<int32, [4]> shape_1 = shape(x = var_795)[name = tensor<string, []>("shape_1")]; |
| tensor<int32, []> equal_1_y_0 = const()[name = tensor<string, []>("equal_1_y_0"), val = tensor<int32, []>(-1)]; |
| tensor<bool, [4]> equal_1 = equal(x = concat_8, y = equal_1_y_0)[name = tensor<string, []>("equal_1")]; |
| tensor<int32, [4]> select_1 = select(a = shape_1, b = concat_8, cond = equal_1)[name = tensor<string, []>("select_1")]; |
| tensor<int32, [4]> real_div_1 = real_div(x = select_1, y = shape_1)[name = tensor<string, []>("real_div_1")]; |
| tensor<int32, [?, ?, ?, ?]> var_798 = tile(reps = real_div_1, x = var_795)[name = tensor<string, []>("op_798")]; |
| tensor<string, []> expanded_mask_dtype_0 = const()[name = tensor<string, []>("expanded_mask_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, []> const_56 = const()[name = tensor<string, []>("const_56"), val = tensor<fp32, []>(0x1p+0)]; |
| tensor<fp32, [?, ?, ?, ?]> expanded_mask = cast(dtype = expanded_mask_dtype_0, x = var_798)[name = tensor<string, []>("cast_85")]; |
| tensor<fp32, [?, ?, ?, ?]> inverted_mask = sub(x = const_56, y = expanded_mask)[name = tensor<string, []>("inverted_mask")]; |
| tensor<string, []> var_803_dtype_0 = const()[name = tensor<string, []>("op_803_dtype_0"), val = tensor<string, []>("bool")]; |
| tensor<bool, [?, ?, ?, ?]> var_803 = cast(dtype = var_803_dtype_0, x = inverted_mask)[name = tensor<string, []>("cast_84")]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_5 = select(a = var_698, b = inverted_mask, cond = var_803)[name = tensor<string, []>("attention_mask_5")]; |
| tensor<bool, [1, 1]> var_808 = not_equal(x = input_ids, y = var_705)[name = tensor<string, []>("op_808")]; |
| tensor<string, []> mask_dtype_0 = const()[name = tensor<string, []>("mask_dtype_0"), val = tensor<string, []>("int32")]; |
| tensor<bool, []> var_810_exclusive_0 = const()[name = tensor<string, []>("op_810_exclusive_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_810_reverse_0 = const()[name = tensor<string, []>("op_810_reverse_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [1, 1]> mask = cast(dtype = mask_dtype_0, x = var_808)[name = tensor<string, []>("cast_83")]; |
| tensor<int32, [1, 1]> var_810 = cumsum(axis = var_705, exclusive = var_810_exclusive_0, reverse = var_810_reverse_0, x = mask)[name = tensor<string, []>("op_810")]; |
| tensor<int32, [1, 1]> var_812 = add(x = var_810, y = gather_0)[name = tensor<string, []>("op_812")]; |
| tensor<int32, [1, 1]> incremental_indices = mul(x = var_812, y = mask)[name = tensor<string, []>("incremental_indices")]; |
| tensor<int32, []> var_815 = const()[name = tensor<string, []>("op_815"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1, 1]> var_816 = add(x = incremental_indices, y = var_815)[name = tensor<string, []>("op_816")]; |
| tensor<int32, [1]> var_818 = const()[name = tensor<string, []>("op_818"), val = tensor<int32, [1]>([-1])]; |
| tensor<int32, [1]> var_819 = reshape(shape = var_818, x = var_816)[name = tensor<string, []>("op_819")]; |
| tensor<int32, []> var_820_batch_dims_0 = const()[name = tensor<string, []>("op_820_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [1, 1024]> var_820 = gather(axis = var_702, batch_dims = var_820_batch_dims_0, indices = var_819, x = decoder_embed_positions_weights)[name = tensor<string, []>("op_820")]; |
| tensor<int32, [3]> var_822 = const()[name = tensor<string, []>("op_822"), val = tensor<int32, [3]>([1, 1, 1024])]; |
| tensor<fp32, [1, 1, 1024]> var_823 = reshape(shape = var_822, x = var_820)[name = tensor<string, []>("op_823")]; |
| tensor<fp32, [1, 1, 1024]> input_3 = add(x = inputs_embeds, y = var_823)[name = tensor<string, []>("input_3")]; |
| tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_1 = layer_norm(axes = hidden_states_1_axes_0, beta = decoder_layers_0_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_0_self_attn_layer_norm_weight, x = input_3)[name = tensor<string, []>("hidden_states_1")]; |
| tensor<fp32, [1, 1, 1024]> var_847 = linear(bias = decoder_layers_0_self_attn_q_proj_bias, weight = decoder_layers_0_self_attn_q_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_0")]; |
| tensor<int32, [4]> var_848 = const()[name = tensor<string, []>("op_848"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_849 = reshape(shape = var_848, x = var_847)[name = tensor<string, []>("op_849")]; |
| tensor<int32, [4]> query_1_perm_0 = const()[name = tensor<string, []>("query_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_49 = linear(bias = decoder_layers_0_self_attn_k_proj_bias, weight = decoder_layers_0_self_attn_k_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_1")]; |
| tensor<fp32, [1, 1, 1024]> value_states_49 = linear(bias = decoder_layers_0_self_attn_v_proj_bias, weight = decoder_layers_0_self_attn_v_proj_weight, x = hidden_states_1)[name = tensor<string, []>("linear_2")]; |
| tensor<int32, [4]> var_857 = const()[name = tensor<string, []>("op_857"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_858 = reshape(shape = var_857, x = key_states_49)[name = tensor<string, []>("op_858")]; |
| tensor<int32, [4]> key_states_51_perm_0 = const()[name = tensor<string, []>("key_states_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_860 = const()[name = tensor<string, []>("op_860"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_861 = reshape(shape = var_860, x = value_states_49)[name = tensor<string, []>("op_861")]; |
| tensor<int32, [4]> value_states_51_perm_0 = const()[name = tensor<string, []>("value_states_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_1_interleave_0 = const()[name = tensor<string, []>("key_1_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_51 = transpose(perm = key_states_51_perm_0, x = var_858)[name = tensor<string, []>("transpose_70")]; |
| tensor<fp32, [1, 16, ?, 64]> key_1 = concat(axis = var_689, interleave = key_1_interleave_0, values = (var_174, key_states_51))[name = tensor<string, []>("key_1")]; |
| tensor<bool, []> value_1_interleave_0 = const()[name = tensor<string, []>("value_1_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_51 = transpose(perm = value_states_51_perm_0, x = var_861)[name = tensor<string, []>("transpose_69")]; |
| tensor<fp32, [1, 16, ?, 64]> value_1 = concat(axis = var_689, interleave = value_1_interleave_0, values = (var_177, value_states_51))[name = tensor<string, []>("value_1")]; |
| tensor<int32, [4]> var_867_shape = shape(x = key_1)[name = tensor<string, []>("op_867_shape")]; |
| tensor<int32, []> gather_5_indices_0 = const()[name = tensor<string, []>("gather_5_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_5_axis_0 = const()[name = tensor<string, []>("gather_5_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_5_batch_dims_0 = const()[name = tensor<string, []>("gather_5_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_5 = gather(axis = gather_5_axis_0, batch_dims = gather_5_batch_dims_0, indices = gather_5_indices_0, x = var_867_shape)[name = tensor<string, []>("gather_5")]; |
| tensor<int32, []> concat_9_values0_0 = const()[name = tensor<string, []>("concat_9_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_9_values1_0 = const()[name = tensor<string, []>("concat_9_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_9_values2_0 = const()[name = tensor<string, []>("concat_9_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_9_axis_0 = const()[name = tensor<string, []>("concat_9_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_9_interleave_0 = const()[name = tensor<string, []>("concat_9_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_9 = concat(axis = concat_9_axis_0, interleave = concat_9_interleave_0, values = (concat_9_values0_0, concat_9_values1_0, concat_9_values2_0, gather_5))[name = tensor<string, []>("concat_9")]; |
| tensor<int32, [4]> attention_mask_3_begin_0 = const()[name = tensor<string, []>("attention_mask_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_3_end_mask_0 = const()[name = tensor<string, []>("attention_mask_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_3 = slice_by_index(begin = attention_mask_3_begin_0, end = concat_9, end_mask = attention_mask_3_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_3")]; |
| tensor<fp32, [1, 16, 1, 64]> query_1 = transpose(perm = query_1_perm_0, x = var_849)[name = tensor<string, []>("transpose_71")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_0 = mul(x = query_1, y = var_687)[name = tensor<string, []>("mul_0")]; |
| tensor<bool, []> matmul_0_transpose_y_0 = const()[name = tensor<string, []>("matmul_0_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_0_transpose_x_0 = const()[name = tensor<string, []>("matmul_0_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_0 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = mul_0, y = key_1)[name = tensor<string, []>("matmul_0")]; |
| tensor<fp32, [?, 16, ?, ?]> add_0 = add(x = matmul_0, y = attention_mask_3)[name = tensor<string, []>("add_0")]; |
| tensor<int32, []> softmax_0_axis_0 = const()[name = tensor<string, []>("softmax_0_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_0 = softmax(axis = softmax_0_axis_0, x = add_0)[name = tensor<string, []>("softmax_0")]; |
| tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_1 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = softmax_0, y = value_1)[name = tensor<string, []>("attn_output_1")]; |
| tensor<int32, [4]> var_873_perm_0 = const()[name = tensor<string, []>("op_873_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_873 = transpose(perm = var_873_perm_0, x = attn_output_1)[name = tensor<string, []>("transpose_68")]; |
| tensor<fp32, [1, 1, ?]> var_876 = reshape(shape = var_875, x = var_873)[name = tensor<string, []>("op_876")]; |
| tensor<fp32, [1, 1, 1024]> input_9 = linear(bias = decoder_layers_0_self_attn_out_proj_bias, weight = decoder_layers_0_self_attn_out_proj_weight, x = var_876)[name = tensor<string, []>("linear_3")]; |
| tensor<fp32, [1, 1, 1024]> input_11 = add(x = input_3, y = input_9)[name = tensor<string, []>("input_11")]; |
| tensor<int32, [1]> hidden_states_5_axes_0 = const()[name = tensor<string, []>("hidden_states_5_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_5 = layer_norm(axes = hidden_states_5_axes_0, beta = decoder_layers_0_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_0_encoder_attn_layer_norm_weight, x = input_11)[name = tensor<string, []>("hidden_states_5")]; |
| tensor<fp32, [1, 1, 1024]> var_897 = linear(bias = decoder_layers_0_encoder_attn_q_proj_bias, weight = decoder_layers_0_encoder_attn_q_proj_weight, x = hidden_states_5)[name = tensor<string, []>("linear_4")]; |
| tensor<int32, [4]> var_898 = const()[name = tensor<string, []>("op_898"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_899 = reshape(shape = var_898, x = var_897)[name = tensor<string, []>("op_899")]; |
| tensor<int32, [4]> query_3_perm_0 = const()[name = tensor<string, []>("query_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_901_shape = shape(x = key_3)[name = tensor<string, []>("op_901_shape")]; |
| tensor<int32, []> gather_6_indices_0 = const()[name = tensor<string, []>("gather_6_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_6_axis_0 = const()[name = tensor<string, []>("gather_6_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_6_batch_dims_0 = const()[name = tensor<string, []>("gather_6_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_6 = gather(axis = gather_6_axis_0, batch_dims = gather_6_batch_dims_0, indices = gather_6_indices_0, x = var_901_shape)[name = tensor<string, []>("gather_6")]; |
| tensor<int32, []> concat_10_values0_0 = const()[name = tensor<string, []>("concat_10_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_10_values1_0 = const()[name = tensor<string, []>("concat_10_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_10_values2_0 = const()[name = tensor<string, []>("concat_10_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_10_axis_0 = const()[name = tensor<string, []>("concat_10_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_10_interleave_0 = const()[name = tensor<string, []>("concat_10_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_10 = concat(axis = concat_10_axis_0, interleave = concat_10_interleave_0, values = (concat_10_values0_0, concat_10_values1_0, concat_10_values2_0, gather_6))[name = tensor<string, []>("concat_10")]; |
| tensor<int32, [4]> attention_mask_7_begin_0 = const()[name = tensor<string, []>("attention_mask_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_7_end_mask_0 = const()[name = tensor<string, []>("attention_mask_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_7 = slice_by_index(begin = attention_mask_7_begin_0, end = concat_10, end_mask = attention_mask_7_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_7")]; |
| tensor<fp32, [1, 16, 1, 64]> query_3 = transpose(perm = query_3_perm_0, x = var_899)[name = tensor<string, []>("transpose_67")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_1 = mul(x = query_3, y = var_687)[name = tensor<string, []>("mul_1")]; |
| tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_1 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = mul_1, y = key_3)[name = tensor<string, []>("matmul_1")]; |
| tensor<fp32, [?, 16, ?, ?]> add_1 = add(x = matmul_1, y = attention_mask_7)[name = tensor<string, []>("add_1")]; |
| tensor<int32, []> softmax_1_axis_0 = const()[name = tensor<string, []>("softmax_1_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_1 = softmax(axis = softmax_1_axis_0, x = add_1)[name = tensor<string, []>("softmax_1")]; |
| tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_5 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = softmax_1, y = value_3)[name = tensor<string, []>("attn_output_5")]; |
| tensor<int32, [4]> var_907_perm_0 = const()[name = tensor<string, []>("op_907_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_909 = const()[name = tensor<string, []>("op_909"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_907 = transpose(perm = var_907_perm_0, x = attn_output_5)[name = tensor<string, []>("transpose_66")]; |
| tensor<fp32, [1, 1, ?]> var_910 = reshape(shape = var_909, x = var_907)[name = tensor<string, []>("op_910")]; |
| tensor<fp32, [1, 1, 1024]> input_15 = linear(bias = decoder_layers_0_encoder_attn_out_proj_bias, weight = decoder_layers_0_encoder_attn_out_proj_weight, x = var_910)[name = tensor<string, []>("linear_5")]; |
| tensor<fp32, [1, 1, 1024]> input_17 = add(x = input_11, y = input_15)[name = tensor<string, []>("input_17")]; |
| tensor<int32, [1]> input_19_axes_0 = const()[name = tensor<string, []>("input_19_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_19 = layer_norm(axes = input_19_axes_0, beta = decoder_layers_0_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_0_final_layer_norm_weight, x = input_17)[name = tensor<string, []>("input_19")]; |
| tensor<fp32, [1, 1, 4096]> input_21 = linear(bias = decoder_layers_0_fc1_bias, weight = decoder_layers_0_fc1_weight, x = input_19)[name = tensor<string, []>("linear_6")]; |
| tensor<fp32, [1, 1, 4096]> input_23 = relu(x = input_21)[name = tensor<string, []>("input_23")]; |
| tensor<fp32, [1, 1, 1024]> input_27 = linear(bias = decoder_layers_0_fc2_bias, weight = decoder_layers_0_fc2_weight, x = input_23)[name = tensor<string, []>("linear_7")]; |
| tensor<fp32, [1, 1, 1024]> input_29 = add(x = input_17, y = input_27)[name = tensor<string, []>("input_29")]; |
| tensor<int32, [1]> hidden_states_11_axes_0 = const()[name = tensor<string, []>("hidden_states_11_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_11 = layer_norm(axes = hidden_states_11_axes_0, beta = decoder_layers_1_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_1_self_attn_layer_norm_weight, x = input_29)[name = tensor<string, []>("hidden_states_11")]; |
| tensor<fp32, [1, 1, 1024]> var_954 = linear(bias = decoder_layers_1_self_attn_q_proj_bias, weight = decoder_layers_1_self_attn_q_proj_weight, x = hidden_states_11)[name = tensor<string, []>("linear_8")]; |
| tensor<int32, [4]> var_955 = const()[name = tensor<string, []>("op_955"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_956 = reshape(shape = var_955, x = var_954)[name = tensor<string, []>("op_956")]; |
| tensor<int32, [4]> query_5_perm_0 = const()[name = tensor<string, []>("query_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_53 = linear(bias = decoder_layers_1_self_attn_k_proj_bias, weight = decoder_layers_1_self_attn_k_proj_weight, x = hidden_states_11)[name = tensor<string, []>("linear_9")]; |
| tensor<fp32, [1, 1, 1024]> value_states_53 = linear(bias = decoder_layers_1_self_attn_v_proj_bias, weight = decoder_layers_1_self_attn_v_proj_weight, x = hidden_states_11)[name = tensor<string, []>("linear_10")]; |
| tensor<int32, [4]> var_964 = const()[name = tensor<string, []>("op_964"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_965 = reshape(shape = var_964, x = key_states_53)[name = tensor<string, []>("op_965")]; |
| tensor<int32, [4]> key_states_55_perm_0 = const()[name = tensor<string, []>("key_states_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_967 = const()[name = tensor<string, []>("op_967"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_968 = reshape(shape = var_967, x = value_states_53)[name = tensor<string, []>("op_968")]; |
| tensor<int32, [4]> value_states_55_perm_0 = const()[name = tensor<string, []>("value_states_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_5_interleave_0 = const()[name = tensor<string, []>("key_5_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_55 = transpose(perm = key_states_55_perm_0, x = var_965)[name = tensor<string, []>("transpose_64")]; |
| tensor<fp32, [1, 16, ?, 64]> key_5 = concat(axis = var_689, interleave = key_5_interleave_0, values = (var_218, key_states_55))[name = tensor<string, []>("key_5")]; |
| tensor<bool, []> value_5_interleave_0 = const()[name = tensor<string, []>("value_5_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_55 = transpose(perm = value_states_55_perm_0, x = var_968)[name = tensor<string, []>("transpose_63")]; |
| tensor<fp32, [1, 16, ?, 64]> value_5 = concat(axis = var_689, interleave = value_5_interleave_0, values = (var_221, value_states_55))[name = tensor<string, []>("value_5")]; |
| tensor<int32, [4]> var_974_shape = shape(x = key_5)[name = tensor<string, []>("op_974_shape")]; |
| tensor<int32, []> gather_7_indices_0 = const()[name = tensor<string, []>("gather_7_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_7_axis_0 = const()[name = tensor<string, []>("gather_7_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_7_batch_dims_0 = const()[name = tensor<string, []>("gather_7_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_7 = gather(axis = gather_7_axis_0, batch_dims = gather_7_batch_dims_0, indices = gather_7_indices_0, x = var_974_shape)[name = tensor<string, []>("gather_7")]; |
| tensor<int32, []> concat_11_values0_0 = const()[name = tensor<string, []>("concat_11_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_11_values1_0 = const()[name = tensor<string, []>("concat_11_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_11_values2_0 = const()[name = tensor<string, []>("concat_11_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_11_axis_0 = const()[name = tensor<string, []>("concat_11_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_11_interleave_0 = const()[name = tensor<string, []>("concat_11_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_11 = concat(axis = concat_11_axis_0, interleave = concat_11_interleave_0, values = (concat_11_values0_0, concat_11_values1_0, concat_11_values2_0, gather_7))[name = tensor<string, []>("concat_11")]; |
| tensor<int32, [4]> attention_mask_9_begin_0 = const()[name = tensor<string, []>("attention_mask_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_9_end_mask_0 = const()[name = tensor<string, []>("attention_mask_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_9 = slice_by_index(begin = attention_mask_9_begin_0, end = concat_11, end_mask = attention_mask_9_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_9")]; |
| tensor<fp32, [1, 16, 1, 64]> query_5 = transpose(perm = query_5_perm_0, x = var_956)[name = tensor<string, []>("transpose_65")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_2 = mul(x = query_5, y = var_687)[name = tensor<string, []>("mul_2")]; |
| tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_2 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = mul_2, y = key_5)[name = tensor<string, []>("matmul_2")]; |
| tensor<fp32, [?, 16, ?, ?]> add_2 = add(x = matmul_2, y = attention_mask_9)[name = tensor<string, []>("add_2")]; |
| tensor<int32, []> softmax_2_axis_0 = const()[name = tensor<string, []>("softmax_2_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_2 = softmax(axis = softmax_2_axis_0, x = add_2)[name = tensor<string, []>("softmax_2")]; |
| tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_9 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = softmax_2, y = value_5)[name = tensor<string, []>("attn_output_9")]; |
| tensor<int32, [4]> var_980_perm_0 = const()[name = tensor<string, []>("op_980_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_982 = const()[name = tensor<string, []>("op_982"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_980 = transpose(perm = var_980_perm_0, x = attn_output_9)[name = tensor<string, []>("transpose_62")]; |
| tensor<fp32, [1, 1, ?]> var_983 = reshape(shape = var_982, x = var_980)[name = tensor<string, []>("op_983")]; |
| tensor<fp32, [1, 1, 1024]> input_33 = linear(bias = decoder_layers_1_self_attn_out_proj_bias, weight = decoder_layers_1_self_attn_out_proj_weight, x = var_983)[name = tensor<string, []>("linear_11")]; |
| tensor<fp32, [1, 1, 1024]> input_35 = add(x = input_29, y = input_33)[name = tensor<string, []>("input_35")]; |
| tensor<int32, [1]> hidden_states_15_axes_0 = const()[name = tensor<string, []>("hidden_states_15_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_15 = layer_norm(axes = hidden_states_15_axes_0, beta = decoder_layers_1_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_1_encoder_attn_layer_norm_weight, x = input_35)[name = tensor<string, []>("hidden_states_15")]; |
| tensor<fp32, [1, 1, 1024]> var_1004 = linear(bias = decoder_layers_1_encoder_attn_q_proj_bias, weight = decoder_layers_1_encoder_attn_q_proj_weight, x = hidden_states_15)[name = tensor<string, []>("linear_12")]; |
| tensor<int32, [4]> var_1005 = const()[name = tensor<string, []>("op_1005"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1006 = reshape(shape = var_1005, x = var_1004)[name = tensor<string, []>("op_1006")]; |
| tensor<int32, [4]> query_7_perm_0 = const()[name = tensor<string, []>("query_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1008_shape = shape(x = key_7)[name = tensor<string, []>("op_1008_shape")]; |
| tensor<int32, []> gather_8_indices_0 = const()[name = tensor<string, []>("gather_8_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_8_axis_0 = const()[name = tensor<string, []>("gather_8_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_8_batch_dims_0 = const()[name = tensor<string, []>("gather_8_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_8 = gather(axis = gather_8_axis_0, batch_dims = gather_8_batch_dims_0, indices = gather_8_indices_0, x = var_1008_shape)[name = tensor<string, []>("gather_8")]; |
| tensor<int32, []> concat_12_values0_0 = const()[name = tensor<string, []>("concat_12_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_12_values1_0 = const()[name = tensor<string, []>("concat_12_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_12_values2_0 = const()[name = tensor<string, []>("concat_12_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_12_axis_0 = const()[name = tensor<string, []>("concat_12_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_12_interleave_0 = const()[name = tensor<string, []>("concat_12_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_12 = concat(axis = concat_12_axis_0, interleave = concat_12_interleave_0, values = (concat_12_values0_0, concat_12_values1_0, concat_12_values2_0, gather_8))[name = tensor<string, []>("concat_12")]; |
| tensor<int32, [4]> attention_mask_11_begin_0 = const()[name = tensor<string, []>("attention_mask_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_11_end_mask_0 = const()[name = tensor<string, []>("attention_mask_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_11 = slice_by_index(begin = attention_mask_11_begin_0, end = concat_12, end_mask = attention_mask_11_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_11")]; |
| tensor<fp32, [1, 16, 1, 64]> query_7 = transpose(perm = query_7_perm_0, x = var_1006)[name = tensor<string, []>("transpose_61")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_3 = mul(x = query_7, y = var_687)[name = tensor<string, []>("mul_3")]; |
| tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_3 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = mul_3, y = key_7)[name = tensor<string, []>("matmul_3")]; |
| tensor<fp32, [?, 16, ?, ?]> add_3 = add(x = matmul_3, y = attention_mask_11)[name = tensor<string, []>("add_3")]; |
| tensor<int32, []> softmax_3_axis_0 = const()[name = tensor<string, []>("softmax_3_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_3 = softmax(axis = softmax_3_axis_0, x = add_3)[name = tensor<string, []>("softmax_3")]; |
| tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_13 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = softmax_3, y = value_7)[name = tensor<string, []>("attn_output_13")]; |
| tensor<int32, [4]> var_1014_perm_0 = const()[name = tensor<string, []>("op_1014_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1016 = const()[name = tensor<string, []>("op_1016"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1014 = transpose(perm = var_1014_perm_0, x = attn_output_13)[name = tensor<string, []>("transpose_60")]; |
| tensor<fp32, [1, 1, ?]> var_1017 = reshape(shape = var_1016, x = var_1014)[name = tensor<string, []>("op_1017")]; |
| tensor<fp32, [1, 1, 1024]> input_39 = linear(bias = decoder_layers_1_encoder_attn_out_proj_bias, weight = decoder_layers_1_encoder_attn_out_proj_weight, x = var_1017)[name = tensor<string, []>("linear_13")]; |
| tensor<fp32, [1, 1, 1024]> input_41 = add(x = input_35, y = input_39)[name = tensor<string, []>("input_41")]; |
| tensor<int32, [1]> input_43_axes_0 = const()[name = tensor<string, []>("input_43_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_43 = layer_norm(axes = input_43_axes_0, beta = decoder_layers_1_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_1_final_layer_norm_weight, x = input_41)[name = tensor<string, []>("input_43")]; |
| tensor<fp32, [1, 1, 4096]> input_45 = linear(bias = decoder_layers_1_fc1_bias, weight = decoder_layers_1_fc1_weight, x = input_43)[name = tensor<string, []>("linear_14")]; |
| tensor<fp32, [1, 1, 4096]> input_47 = relu(x = input_45)[name = tensor<string, []>("input_47")]; |
| tensor<fp32, [1, 1, 1024]> input_51 = linear(bias = decoder_layers_1_fc2_bias, weight = decoder_layers_1_fc2_weight, x = input_47)[name = tensor<string, []>("linear_15")]; |
| tensor<fp32, [1, 1, 1024]> input_53 = add(x = input_41, y = input_51)[name = tensor<string, []>("input_53")]; |
| tensor<int32, [1]> hidden_states_21_axes_0 = const()[name = tensor<string, []>("hidden_states_21_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_21 = layer_norm(axes = hidden_states_21_axes_0, beta = decoder_layers_2_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_2_self_attn_layer_norm_weight, x = input_53)[name = tensor<string, []>("hidden_states_21")]; |
| tensor<fp32, [1, 1, 1024]> var_1061 = linear(bias = decoder_layers_2_self_attn_q_proj_bias, weight = decoder_layers_2_self_attn_q_proj_weight, x = hidden_states_21)[name = tensor<string, []>("linear_16")]; |
| tensor<int32, [4]> var_1062 = const()[name = tensor<string, []>("op_1062"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1063 = reshape(shape = var_1062, x = var_1061)[name = tensor<string, []>("op_1063")]; |
| tensor<int32, [4]> query_9_perm_0 = const()[name = tensor<string, []>("query_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_57 = linear(bias = decoder_layers_2_self_attn_k_proj_bias, weight = decoder_layers_2_self_attn_k_proj_weight, x = hidden_states_21)[name = tensor<string, []>("linear_17")]; |
| tensor<fp32, [1, 1, 1024]> value_states_57 = linear(bias = decoder_layers_2_self_attn_v_proj_bias, weight = decoder_layers_2_self_attn_v_proj_weight, x = hidden_states_21)[name = tensor<string, []>("linear_18")]; |
| tensor<int32, [4]> var_1071 = const()[name = tensor<string, []>("op_1071"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1072 = reshape(shape = var_1071, x = key_states_57)[name = tensor<string, []>("op_1072")]; |
| tensor<int32, [4]> key_states_59_perm_0 = const()[name = tensor<string, []>("key_states_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1074 = const()[name = tensor<string, []>("op_1074"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1075 = reshape(shape = var_1074, x = value_states_57)[name = tensor<string, []>("op_1075")]; |
| tensor<int32, [4]> value_states_59_perm_0 = const()[name = tensor<string, []>("value_states_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_9_interleave_0 = const()[name = tensor<string, []>("key_9_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_59 = transpose(perm = key_states_59_perm_0, x = var_1072)[name = tensor<string, []>("transpose_58")]; |
| tensor<fp32, [1, 16, ?, 64]> key_9 = concat(axis = var_689, interleave = key_9_interleave_0, values = (var_262, key_states_59))[name = tensor<string, []>("key_9")]; |
| tensor<bool, []> value_9_interleave_0 = const()[name = tensor<string, []>("value_9_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_59 = transpose(perm = value_states_59_perm_0, x = var_1075)[name = tensor<string, []>("transpose_57")]; |
| tensor<fp32, [1, 16, ?, 64]> value_9 = concat(axis = var_689, interleave = value_9_interleave_0, values = (var_265, value_states_59))[name = tensor<string, []>("value_9")]; |
| tensor<int32, [4]> var_1081_shape = shape(x = key_9)[name = tensor<string, []>("op_1081_shape")]; |
| tensor<int32, []> gather_9_indices_0 = const()[name = tensor<string, []>("gather_9_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_9_axis_0 = const()[name = tensor<string, []>("gather_9_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_9_batch_dims_0 = const()[name = tensor<string, []>("gather_9_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_9 = gather(axis = gather_9_axis_0, batch_dims = gather_9_batch_dims_0, indices = gather_9_indices_0, x = var_1081_shape)[name = tensor<string, []>("gather_9")]; |
| tensor<int32, []> concat_13_values0_0 = const()[name = tensor<string, []>("concat_13_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_13_values1_0 = const()[name = tensor<string, []>("concat_13_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_13_values2_0 = const()[name = tensor<string, []>("concat_13_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_13_axis_0 = const()[name = tensor<string, []>("concat_13_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_13_interleave_0 = const()[name = tensor<string, []>("concat_13_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_13 = concat(axis = concat_13_axis_0, interleave = concat_13_interleave_0, values = (concat_13_values0_0, concat_13_values1_0, concat_13_values2_0, gather_9))[name = tensor<string, []>("concat_13")]; |
| tensor<int32, [4]> attention_mask_13_begin_0 = const()[name = tensor<string, []>("attention_mask_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_13_end_mask_0 = const()[name = tensor<string, []>("attention_mask_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_13 = slice_by_index(begin = attention_mask_13_begin_0, end = concat_13, end_mask = attention_mask_13_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_13")]; |
| tensor<fp32, [1, 16, 1, 64]> query_9 = transpose(perm = query_9_perm_0, x = var_1063)[name = tensor<string, []>("transpose_59")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_4 = mul(x = query_9, y = var_687)[name = tensor<string, []>("mul_4")]; |
| tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_4 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = mul_4, y = key_9)[name = tensor<string, []>("matmul_4")]; |
| tensor<fp32, [?, 16, ?, ?]> add_4 = add(x = matmul_4, y = attention_mask_13)[name = tensor<string, []>("add_4")]; |
| tensor<int32, []> softmax_4_axis_0 = const()[name = tensor<string, []>("softmax_4_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_4 = softmax(axis = softmax_4_axis_0, x = add_4)[name = tensor<string, []>("softmax_4")]; |
| tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_17 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = softmax_4, y = value_9)[name = tensor<string, []>("attn_output_17")]; |
| tensor<int32, [4]> var_1087_perm_0 = const()[name = tensor<string, []>("op_1087_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1089 = const()[name = tensor<string, []>("op_1089"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1087 = transpose(perm = var_1087_perm_0, x = attn_output_17)[name = tensor<string, []>("transpose_56")]; |
| tensor<fp32, [1, 1, ?]> var_1090 = reshape(shape = var_1089, x = var_1087)[name = tensor<string, []>("op_1090")]; |
| tensor<fp32, [1, 1, 1024]> input_57 = linear(bias = decoder_layers_2_self_attn_out_proj_bias, weight = decoder_layers_2_self_attn_out_proj_weight, x = var_1090)[name = tensor<string, []>("linear_19")]; |
| tensor<fp32, [1, 1, 1024]> input_59 = add(x = input_53, y = input_57)[name = tensor<string, []>("input_59")]; |
| tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_25 = layer_norm(axes = hidden_states_25_axes_0, beta = decoder_layers_2_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_2_encoder_attn_layer_norm_weight, x = input_59)[name = tensor<string, []>("hidden_states_25")]; |
| tensor<fp32, [1, 1, 1024]> var_1111 = linear(bias = decoder_layers_2_encoder_attn_q_proj_bias, weight = decoder_layers_2_encoder_attn_q_proj_weight, x = hidden_states_25)[name = tensor<string, []>("linear_20")]; |
| tensor<int32, [4]> var_1112 = const()[name = tensor<string, []>("op_1112"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1113 = reshape(shape = var_1112, x = var_1111)[name = tensor<string, []>("op_1113")]; |
| tensor<int32, [4]> query_11_perm_0 = const()[name = tensor<string, []>("query_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1115_shape = shape(x = key_11)[name = tensor<string, []>("op_1115_shape")]; |
| tensor<int32, []> gather_10_indices_0 = const()[name = tensor<string, []>("gather_10_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_10_axis_0 = const()[name = tensor<string, []>("gather_10_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_10_batch_dims_0 = const()[name = tensor<string, []>("gather_10_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_10 = gather(axis = gather_10_axis_0, batch_dims = gather_10_batch_dims_0, indices = gather_10_indices_0, x = var_1115_shape)[name = tensor<string, []>("gather_10")]; |
| tensor<int32, []> concat_14_values0_0 = const()[name = tensor<string, []>("concat_14_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_14_values1_0 = const()[name = tensor<string, []>("concat_14_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_14_values2_0 = const()[name = tensor<string, []>("concat_14_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_14_axis_0 = const()[name = tensor<string, []>("concat_14_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_14_interleave_0 = const()[name = tensor<string, []>("concat_14_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_14 = concat(axis = concat_14_axis_0, interleave = concat_14_interleave_0, values = (concat_14_values0_0, concat_14_values1_0, concat_14_values2_0, gather_10))[name = tensor<string, []>("concat_14")]; |
| tensor<int32, [4]> attention_mask_15_begin_0 = const()[name = tensor<string, []>("attention_mask_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_15_end_mask_0 = const()[name = tensor<string, []>("attention_mask_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_15 = slice_by_index(begin = attention_mask_15_begin_0, end = concat_14, end_mask = attention_mask_15_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_15")]; |
| tensor<fp32, [1, 16, 1, 64]> query_11 = transpose(perm = query_11_perm_0, x = var_1113)[name = tensor<string, []>("transpose_55")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_5 = mul(x = query_11, y = var_687)[name = tensor<string, []>("mul_5")]; |
| tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_5 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = mul_5, y = key_11)[name = tensor<string, []>("matmul_5")]; |
| tensor<fp32, [?, 16, ?, ?]> add_5 = add(x = matmul_5, y = attention_mask_15)[name = tensor<string, []>("add_5")]; |
| tensor<int32, []> softmax_5_axis_0 = const()[name = tensor<string, []>("softmax_5_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_5 = softmax(axis = softmax_5_axis_0, x = add_5)[name = tensor<string, []>("softmax_5")]; |
| tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_21 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = softmax_5, y = value_11)[name = tensor<string, []>("attn_output_21")]; |
| tensor<int32, [4]> var_1121_perm_0 = const()[name = tensor<string, []>("op_1121_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1123 = const()[name = tensor<string, []>("op_1123"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1121 = transpose(perm = var_1121_perm_0, x = attn_output_21)[name = tensor<string, []>("transpose_54")]; |
| tensor<fp32, [1, 1, ?]> var_1124 = reshape(shape = var_1123, x = var_1121)[name = tensor<string, []>("op_1124")]; |
| tensor<fp32, [1, 1, 1024]> input_63 = linear(bias = decoder_layers_2_encoder_attn_out_proj_bias, weight = decoder_layers_2_encoder_attn_out_proj_weight, x = var_1124)[name = tensor<string, []>("linear_21")]; |
| tensor<fp32, [1, 1, 1024]> input_65 = add(x = input_59, y = input_63)[name = tensor<string, []>("input_65")]; |
| tensor<int32, [1]> input_67_axes_0 = const()[name = tensor<string, []>("input_67_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_67 = layer_norm(axes = input_67_axes_0, beta = decoder_layers_2_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_2_final_layer_norm_weight, x = input_65)[name = tensor<string, []>("input_67")]; |
| tensor<fp32, [1, 1, 4096]> input_69 = linear(bias = decoder_layers_2_fc1_bias, weight = decoder_layers_2_fc1_weight, x = input_67)[name = tensor<string, []>("linear_22")]; |
| tensor<fp32, [1, 1, 4096]> input_71 = relu(x = input_69)[name = tensor<string, []>("input_71")]; |
| tensor<fp32, [1, 1, 1024]> input_75 = linear(bias = decoder_layers_2_fc2_bias, weight = decoder_layers_2_fc2_weight, x = input_71)[name = tensor<string, []>("linear_23")]; |
| tensor<fp32, [1, 1, 1024]> input_77 = add(x = input_65, y = input_75)[name = tensor<string, []>("input_77")]; |
| tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_31 = layer_norm(axes = hidden_states_31_axes_0, beta = decoder_layers_3_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_3_self_attn_layer_norm_weight, x = input_77)[name = tensor<string, []>("hidden_states_31")]; |
| tensor<fp32, [1, 1, 1024]> var_1168 = linear(bias = decoder_layers_3_self_attn_q_proj_bias, weight = decoder_layers_3_self_attn_q_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_24")]; |
| tensor<int32, [4]> var_1169 = const()[name = tensor<string, []>("op_1169"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1170 = reshape(shape = var_1169, x = var_1168)[name = tensor<string, []>("op_1170")]; |
| tensor<int32, [4]> query_13_perm_0 = const()[name = tensor<string, []>("query_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_61 = linear(bias = decoder_layers_3_self_attn_k_proj_bias, weight = decoder_layers_3_self_attn_k_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_25")]; |
| tensor<fp32, [1, 1, 1024]> value_states_61 = linear(bias = decoder_layers_3_self_attn_v_proj_bias, weight = decoder_layers_3_self_attn_v_proj_weight, x = hidden_states_31)[name = tensor<string, []>("linear_26")]; |
| tensor<int32, [4]> var_1178 = const()[name = tensor<string, []>("op_1178"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1179 = reshape(shape = var_1178, x = key_states_61)[name = tensor<string, []>("op_1179")]; |
| tensor<int32, [4]> key_states_63_perm_0 = const()[name = tensor<string, []>("key_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1181 = const()[name = tensor<string, []>("op_1181"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1182 = reshape(shape = var_1181, x = value_states_61)[name = tensor<string, []>("op_1182")]; |
| tensor<int32, [4]> value_states_63_perm_0 = const()[name = tensor<string, []>("value_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_13_interleave_0 = const()[name = tensor<string, []>("key_13_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_63 = transpose(perm = key_states_63_perm_0, x = var_1179)[name = tensor<string, []>("transpose_52")]; |
| tensor<fp32, [1, 16, ?, 64]> key_13 = concat(axis = var_689, interleave = key_13_interleave_0, values = (var_306, key_states_63))[name = tensor<string, []>("key_13")]; |
| tensor<bool, []> value_13_interleave_0 = const()[name = tensor<string, []>("value_13_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_63 = transpose(perm = value_states_63_perm_0, x = var_1182)[name = tensor<string, []>("transpose_51")]; |
| tensor<fp32, [1, 16, ?, 64]> value_13 = concat(axis = var_689, interleave = value_13_interleave_0, values = (var_309, value_states_63))[name = tensor<string, []>("value_13")]; |
| tensor<int32, [4]> var_1188_shape = shape(x = key_13)[name = tensor<string, []>("op_1188_shape")]; |
| tensor<int32, []> gather_11_indices_0 = const()[name = tensor<string, []>("gather_11_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_11_axis_0 = const()[name = tensor<string, []>("gather_11_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_11_batch_dims_0 = const()[name = tensor<string, []>("gather_11_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_11 = gather(axis = gather_11_axis_0, batch_dims = gather_11_batch_dims_0, indices = gather_11_indices_0, x = var_1188_shape)[name = tensor<string, []>("gather_11")]; |
| tensor<int32, []> concat_15_values0_0 = const()[name = tensor<string, []>("concat_15_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_15_values1_0 = const()[name = tensor<string, []>("concat_15_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_15_values2_0 = const()[name = tensor<string, []>("concat_15_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_15_axis_0 = const()[name = tensor<string, []>("concat_15_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_15_interleave_0 = const()[name = tensor<string, []>("concat_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_15 = concat(axis = concat_15_axis_0, interleave = concat_15_interleave_0, values = (concat_15_values0_0, concat_15_values1_0, concat_15_values2_0, gather_11))[name = tensor<string, []>("concat_15")]; |
| tensor<int32, [4]> attention_mask_17_begin_0 = const()[name = tensor<string, []>("attention_mask_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_17_end_mask_0 = const()[name = tensor<string, []>("attention_mask_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_17 = slice_by_index(begin = attention_mask_17_begin_0, end = concat_15, end_mask = attention_mask_17_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_17")]; |
| tensor<fp32, [1, 16, 1, 64]> query_13 = transpose(perm = query_13_perm_0, x = var_1170)[name = tensor<string, []>("transpose_53")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_6 = mul(x = query_13, y = var_687)[name = tensor<string, []>("mul_6")]; |
| tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_6 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = mul_6, y = key_13)[name = tensor<string, []>("matmul_6")]; |
| tensor<fp32, [?, 16, ?, ?]> add_6 = add(x = matmul_6, y = attention_mask_17)[name = tensor<string, []>("add_6")]; |
| tensor<int32, []> softmax_6_axis_0 = const()[name = tensor<string, []>("softmax_6_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_6 = softmax(axis = softmax_6_axis_0, x = add_6)[name = tensor<string, []>("softmax_6")]; |
| tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_25 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = softmax_6, y = value_13)[name = tensor<string, []>("attn_output_25")]; |
| tensor<int32, [4]> var_1194_perm_0 = const()[name = tensor<string, []>("op_1194_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1196 = const()[name = tensor<string, []>("op_1196"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1194 = transpose(perm = var_1194_perm_0, x = attn_output_25)[name = tensor<string, []>("transpose_50")]; |
| tensor<fp32, [1, 1, ?]> var_1197 = reshape(shape = var_1196, x = var_1194)[name = tensor<string, []>("op_1197")]; |
| tensor<fp32, [1, 1, 1024]> input_81 = linear(bias = decoder_layers_3_self_attn_out_proj_bias, weight = decoder_layers_3_self_attn_out_proj_weight, x = var_1197)[name = tensor<string, []>("linear_27")]; |
| tensor<fp32, [1, 1, 1024]> input_83 = add(x = input_77, y = input_81)[name = tensor<string, []>("input_83")]; |
| tensor<int32, [1]> hidden_states_35_axes_0 = const()[name = tensor<string, []>("hidden_states_35_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_35 = layer_norm(axes = hidden_states_35_axes_0, beta = decoder_layers_3_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_3_encoder_attn_layer_norm_weight, x = input_83)[name = tensor<string, []>("hidden_states_35")]; |
| tensor<fp32, [1, 1, 1024]> var_1218 = linear(bias = decoder_layers_3_encoder_attn_q_proj_bias, weight = decoder_layers_3_encoder_attn_q_proj_weight, x = hidden_states_35)[name = tensor<string, []>("linear_28")]; |
| tensor<int32, [4]> var_1219 = const()[name = tensor<string, []>("op_1219"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1220 = reshape(shape = var_1219, x = var_1218)[name = tensor<string, []>("op_1220")]; |
| tensor<int32, [4]> query_15_perm_0 = const()[name = tensor<string, []>("query_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1222_shape = shape(x = key_15)[name = tensor<string, []>("op_1222_shape")]; |
| tensor<int32, []> gather_12_indices_0 = const()[name = tensor<string, []>("gather_12_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_12_axis_0 = const()[name = tensor<string, []>("gather_12_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_12_batch_dims_0 = const()[name = tensor<string, []>("gather_12_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_12 = gather(axis = gather_12_axis_0, batch_dims = gather_12_batch_dims_0, indices = gather_12_indices_0, x = var_1222_shape)[name = tensor<string, []>("gather_12")]; |
| tensor<int32, []> concat_16_values0_0 = const()[name = tensor<string, []>("concat_16_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_16_values1_0 = const()[name = tensor<string, []>("concat_16_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_16_values2_0 = const()[name = tensor<string, []>("concat_16_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_16_axis_0 = const()[name = tensor<string, []>("concat_16_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_16_interleave_0 = const()[name = tensor<string, []>("concat_16_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_16 = concat(axis = concat_16_axis_0, interleave = concat_16_interleave_0, values = (concat_16_values0_0, concat_16_values1_0, concat_16_values2_0, gather_12))[name = tensor<string, []>("concat_16")]; |
| tensor<int32, [4]> attention_mask_19_begin_0 = const()[name = tensor<string, []>("attention_mask_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_19_end_mask_0 = const()[name = tensor<string, []>("attention_mask_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_19 = slice_by_index(begin = attention_mask_19_begin_0, end = concat_16, end_mask = attention_mask_19_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_19")]; |
| tensor<fp32, [1, 16, 1, 64]> query_15 = transpose(perm = query_15_perm_0, x = var_1220)[name = tensor<string, []>("transpose_49")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_7 = mul(x = query_15, y = var_687)[name = tensor<string, []>("mul_7")]; |
| tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_7 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = mul_7, y = key_15)[name = tensor<string, []>("matmul_7")]; |
| tensor<fp32, [?, 16, ?, ?]> add_7 = add(x = matmul_7, y = attention_mask_19)[name = tensor<string, []>("add_7")]; |
| tensor<int32, []> softmax_7_axis_0 = const()[name = tensor<string, []>("softmax_7_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_7 = softmax(axis = softmax_7_axis_0, x = add_7)[name = tensor<string, []>("softmax_7")]; |
| tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_29 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = softmax_7, y = value_15)[name = tensor<string, []>("attn_output_29")]; |
| tensor<int32, [4]> var_1228_perm_0 = const()[name = tensor<string, []>("op_1228_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1230 = const()[name = tensor<string, []>("op_1230"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1228 = transpose(perm = var_1228_perm_0, x = attn_output_29)[name = tensor<string, []>("transpose_48")]; |
| tensor<fp32, [1, 1, ?]> var_1231 = reshape(shape = var_1230, x = var_1228)[name = tensor<string, []>("op_1231")]; |
| tensor<fp32, [1, 1, 1024]> input_87 = linear(bias = decoder_layers_3_encoder_attn_out_proj_bias, weight = decoder_layers_3_encoder_attn_out_proj_weight, x = var_1231)[name = tensor<string, []>("linear_29")]; |
| tensor<fp32, [1, 1, 1024]> input_89 = add(x = input_83, y = input_87)[name = tensor<string, []>("input_89")]; |
| tensor<int32, [1]> input_91_axes_0 = const()[name = tensor<string, []>("input_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_91 = layer_norm(axes = input_91_axes_0, beta = decoder_layers_3_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_3_final_layer_norm_weight, x = input_89)[name = tensor<string, []>("input_91")]; |
| tensor<fp32, [1, 1, 4096]> input_93 = linear(bias = decoder_layers_3_fc1_bias, weight = decoder_layers_3_fc1_weight, x = input_91)[name = tensor<string, []>("linear_30")]; |
| tensor<fp32, [1, 1, 4096]> input_95 = relu(x = input_93)[name = tensor<string, []>("input_95")]; |
| tensor<fp32, [1, 1, 1024]> input_99 = linear(bias = decoder_layers_3_fc2_bias, weight = decoder_layers_3_fc2_weight, x = input_95)[name = tensor<string, []>("linear_31")]; |
| tensor<fp32, [1, 1, 1024]> input_101 = add(x = input_89, y = input_99)[name = tensor<string, []>("input_101")]; |
| tensor<int32, [1]> hidden_states_41_axes_0 = const()[name = tensor<string, []>("hidden_states_41_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_41 = layer_norm(axes = hidden_states_41_axes_0, beta = decoder_layers_4_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_4_self_attn_layer_norm_weight, x = input_101)[name = tensor<string, []>("hidden_states_41")]; |
| tensor<fp32, [1, 1, 1024]> var_1275 = linear(bias = decoder_layers_4_self_attn_q_proj_bias, weight = decoder_layers_4_self_attn_q_proj_weight, x = hidden_states_41)[name = tensor<string, []>("linear_32")]; |
| tensor<int32, [4]> var_1276 = const()[name = tensor<string, []>("op_1276"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1277 = reshape(shape = var_1276, x = var_1275)[name = tensor<string, []>("op_1277")]; |
| tensor<int32, [4]> query_17_perm_0 = const()[name = tensor<string, []>("query_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_65 = linear(bias = decoder_layers_4_self_attn_k_proj_bias, weight = decoder_layers_4_self_attn_k_proj_weight, x = hidden_states_41)[name = tensor<string, []>("linear_33")]; |
| tensor<fp32, [1, 1, 1024]> value_states_65 = linear(bias = decoder_layers_4_self_attn_v_proj_bias, weight = decoder_layers_4_self_attn_v_proj_weight, x = hidden_states_41)[name = tensor<string, []>("linear_34")]; |
| tensor<int32, [4]> var_1285 = const()[name = tensor<string, []>("op_1285"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1286 = reshape(shape = var_1285, x = key_states_65)[name = tensor<string, []>("op_1286")]; |
| tensor<int32, [4]> key_states_67_perm_0 = const()[name = tensor<string, []>("key_states_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1288 = const()[name = tensor<string, []>("op_1288"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1289 = reshape(shape = var_1288, x = value_states_65)[name = tensor<string, []>("op_1289")]; |
| tensor<int32, [4]> value_states_67_perm_0 = const()[name = tensor<string, []>("value_states_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_17_interleave_0 = const()[name = tensor<string, []>("key_17_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_67 = transpose(perm = key_states_67_perm_0, x = var_1286)[name = tensor<string, []>("transpose_46")]; |
| tensor<fp32, [1, 16, ?, 64]> key_17 = concat(axis = var_689, interleave = key_17_interleave_0, values = (var_350, key_states_67))[name = tensor<string, []>("key_17")]; |
| tensor<bool, []> value_17_interleave_0 = const()[name = tensor<string, []>("value_17_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_67 = transpose(perm = value_states_67_perm_0, x = var_1289)[name = tensor<string, []>("transpose_45")]; |
| tensor<fp32, [1, 16, ?, 64]> value_17 = concat(axis = var_689, interleave = value_17_interleave_0, values = (var_353, value_states_67))[name = tensor<string, []>("value_17")]; |
| tensor<int32, [4]> var_1295_shape = shape(x = key_17)[name = tensor<string, []>("op_1295_shape")]; |
| tensor<int32, []> gather_13_indices_0 = const()[name = tensor<string, []>("gather_13_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_13_axis_0 = const()[name = tensor<string, []>("gather_13_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_13_batch_dims_0 = const()[name = tensor<string, []>("gather_13_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_13 = gather(axis = gather_13_axis_0, batch_dims = gather_13_batch_dims_0, indices = gather_13_indices_0, x = var_1295_shape)[name = tensor<string, []>("gather_13")]; |
| tensor<int32, []> concat_17_values0_0 = const()[name = tensor<string, []>("concat_17_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_17_values1_0 = const()[name = tensor<string, []>("concat_17_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_17_values2_0 = const()[name = tensor<string, []>("concat_17_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_17_axis_0 = const()[name = tensor<string, []>("concat_17_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_17_interleave_0 = const()[name = tensor<string, []>("concat_17_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_17 = concat(axis = concat_17_axis_0, interleave = concat_17_interleave_0, values = (concat_17_values0_0, concat_17_values1_0, concat_17_values2_0, gather_13))[name = tensor<string, []>("concat_17")]; |
| tensor<int32, [4]> attention_mask_21_begin_0 = const()[name = tensor<string, []>("attention_mask_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_21_end_mask_0 = const()[name = tensor<string, []>("attention_mask_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_21 = slice_by_index(begin = attention_mask_21_begin_0, end = concat_17, end_mask = attention_mask_21_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_21")]; |
| tensor<fp32, [1, 16, 1, 64]> query_17 = transpose(perm = query_17_perm_0, x = var_1277)[name = tensor<string, []>("transpose_47")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_8 = mul(x = query_17, y = var_687)[name = tensor<string, []>("mul_8")]; |
| tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_8 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = mul_8, y = key_17)[name = tensor<string, []>("matmul_8")]; |
| tensor<fp32, [?, 16, ?, ?]> add_8 = add(x = matmul_8, y = attention_mask_21)[name = tensor<string, []>("add_8")]; |
| tensor<int32, []> softmax_8_axis_0 = const()[name = tensor<string, []>("softmax_8_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_8 = softmax(axis = softmax_8_axis_0, x = add_8)[name = tensor<string, []>("softmax_8")]; |
| tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_33 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = softmax_8, y = value_17)[name = tensor<string, []>("attn_output_33")]; |
| tensor<int32, [4]> var_1301_perm_0 = const()[name = tensor<string, []>("op_1301_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1303 = const()[name = tensor<string, []>("op_1303"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1301 = transpose(perm = var_1301_perm_0, x = attn_output_33)[name = tensor<string, []>("transpose_44")]; |
| tensor<fp32, [1, 1, ?]> var_1304 = reshape(shape = var_1303, x = var_1301)[name = tensor<string, []>("op_1304")]; |
| tensor<fp32, [1, 1, 1024]> input_105 = linear(bias = decoder_layers_4_self_attn_out_proj_bias, weight = decoder_layers_4_self_attn_out_proj_weight, x = var_1304)[name = tensor<string, []>("linear_35")]; |
| tensor<fp32, [1, 1, 1024]> input_107 = add(x = input_101, y = input_105)[name = tensor<string, []>("input_107")]; |
| tensor<int32, [1]> hidden_states_45_axes_0 = const()[name = tensor<string, []>("hidden_states_45_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_45 = layer_norm(axes = hidden_states_45_axes_0, beta = decoder_layers_4_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_4_encoder_attn_layer_norm_weight, x = input_107)[name = tensor<string, []>("hidden_states_45")]; |
| tensor<fp32, [1, 1, 1024]> var_1325 = linear(bias = decoder_layers_4_encoder_attn_q_proj_bias, weight = decoder_layers_4_encoder_attn_q_proj_weight, x = hidden_states_45)[name = tensor<string, []>("linear_36")]; |
| tensor<int32, [4]> var_1326 = const()[name = tensor<string, []>("op_1326"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1327 = reshape(shape = var_1326, x = var_1325)[name = tensor<string, []>("op_1327")]; |
| tensor<int32, [4]> query_19_perm_0 = const()[name = tensor<string, []>("query_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1329_shape = shape(x = key_19)[name = tensor<string, []>("op_1329_shape")]; |
| tensor<int32, []> gather_14_indices_0 = const()[name = tensor<string, []>("gather_14_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_14_axis_0 = const()[name = tensor<string, []>("gather_14_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_14_batch_dims_0 = const()[name = tensor<string, []>("gather_14_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_14 = gather(axis = gather_14_axis_0, batch_dims = gather_14_batch_dims_0, indices = gather_14_indices_0, x = var_1329_shape)[name = tensor<string, []>("gather_14")]; |
| tensor<int32, []> concat_18_values0_0 = const()[name = tensor<string, []>("concat_18_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_18_values1_0 = const()[name = tensor<string, []>("concat_18_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_18_values2_0 = const()[name = tensor<string, []>("concat_18_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_18_axis_0 = const()[name = tensor<string, []>("concat_18_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_18_interleave_0 = const()[name = tensor<string, []>("concat_18_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_18 = concat(axis = concat_18_axis_0, interleave = concat_18_interleave_0, values = (concat_18_values0_0, concat_18_values1_0, concat_18_values2_0, gather_14))[name = tensor<string, []>("concat_18")]; |
| tensor<int32, [4]> attention_mask_23_begin_0 = const()[name = tensor<string, []>("attention_mask_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_23_end_mask_0 = const()[name = tensor<string, []>("attention_mask_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_23 = slice_by_index(begin = attention_mask_23_begin_0, end = concat_18, end_mask = attention_mask_23_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_23")]; |
| tensor<fp32, [1, 16, 1, 64]> query_19 = transpose(perm = query_19_perm_0, x = var_1327)[name = tensor<string, []>("transpose_43")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_9 = mul(x = query_19, y = var_687)[name = tensor<string, []>("mul_9")]; |
| tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_9 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = mul_9, y = key_19)[name = tensor<string, []>("matmul_9")]; |
| tensor<fp32, [?, 16, ?, ?]> add_9 = add(x = matmul_9, y = attention_mask_23)[name = tensor<string, []>("add_9")]; |
| tensor<int32, []> softmax_9_axis_0 = const()[name = tensor<string, []>("softmax_9_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_9 = softmax(axis = softmax_9_axis_0, x = add_9)[name = tensor<string, []>("softmax_9")]; |
| tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_37 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = softmax_9, y = value_19)[name = tensor<string, []>("attn_output_37")]; |
| tensor<int32, [4]> var_1335_perm_0 = const()[name = tensor<string, []>("op_1335_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1337 = const()[name = tensor<string, []>("op_1337"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1335 = transpose(perm = var_1335_perm_0, x = attn_output_37)[name = tensor<string, []>("transpose_42")]; |
| tensor<fp32, [1, 1, ?]> var_1338 = reshape(shape = var_1337, x = var_1335)[name = tensor<string, []>("op_1338")]; |
| tensor<fp32, [1, 1, 1024]> input_111 = linear(bias = decoder_layers_4_encoder_attn_out_proj_bias, weight = decoder_layers_4_encoder_attn_out_proj_weight, x = var_1338)[name = tensor<string, []>("linear_37")]; |
| tensor<fp32, [1, 1, 1024]> input_113 = add(x = input_107, y = input_111)[name = tensor<string, []>("input_113")]; |
| tensor<int32, [1]> input_115_axes_0 = const()[name = tensor<string, []>("input_115_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_115 = layer_norm(axes = input_115_axes_0, beta = decoder_layers_4_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_4_final_layer_norm_weight, x = input_113)[name = tensor<string, []>("input_115")]; |
| tensor<fp32, [1, 1, 4096]> input_117 = linear(bias = decoder_layers_4_fc1_bias, weight = decoder_layers_4_fc1_weight, x = input_115)[name = tensor<string, []>("linear_38")]; |
| tensor<fp32, [1, 1, 4096]> input_119 = relu(x = input_117)[name = tensor<string, []>("input_119")]; |
| tensor<fp32, [1, 1, 1024]> input_123 = linear(bias = decoder_layers_4_fc2_bias, weight = decoder_layers_4_fc2_weight, x = input_119)[name = tensor<string, []>("linear_39")]; |
| tensor<fp32, [1, 1, 1024]> input_125 = add(x = input_113, y = input_123)[name = tensor<string, []>("input_125")]; |
| tensor<int32, [1]> hidden_states_51_axes_0 = const()[name = tensor<string, []>("hidden_states_51_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_51 = layer_norm(axes = hidden_states_51_axes_0, beta = decoder_layers_5_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_5_self_attn_layer_norm_weight, x = input_125)[name = tensor<string, []>("hidden_states_51")]; |
| tensor<fp32, [1, 1, 1024]> var_1382 = linear(bias = decoder_layers_5_self_attn_q_proj_bias, weight = decoder_layers_5_self_attn_q_proj_weight, x = hidden_states_51)[name = tensor<string, []>("linear_40")]; |
| tensor<int32, [4]> var_1383 = const()[name = tensor<string, []>("op_1383"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1384 = reshape(shape = var_1383, x = var_1382)[name = tensor<string, []>("op_1384")]; |
| tensor<int32, [4]> query_21_perm_0 = const()[name = tensor<string, []>("query_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_69 = linear(bias = decoder_layers_5_self_attn_k_proj_bias, weight = decoder_layers_5_self_attn_k_proj_weight, x = hidden_states_51)[name = tensor<string, []>("linear_41")]; |
| tensor<fp32, [1, 1, 1024]> value_states_69 = linear(bias = decoder_layers_5_self_attn_v_proj_bias, weight = decoder_layers_5_self_attn_v_proj_weight, x = hidden_states_51)[name = tensor<string, []>("linear_42")]; |
| tensor<int32, [4]> var_1392 = const()[name = tensor<string, []>("op_1392"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1393 = reshape(shape = var_1392, x = key_states_69)[name = tensor<string, []>("op_1393")]; |
| tensor<int32, [4]> key_states_71_perm_0 = const()[name = tensor<string, []>("key_states_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1395 = const()[name = tensor<string, []>("op_1395"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1396 = reshape(shape = var_1395, x = value_states_69)[name = tensor<string, []>("op_1396")]; |
| tensor<int32, [4]> value_states_71_perm_0 = const()[name = tensor<string, []>("value_states_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_21_interleave_0 = const()[name = tensor<string, []>("key_21_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_71 = transpose(perm = key_states_71_perm_0, x = var_1393)[name = tensor<string, []>("transpose_40")]; |
| tensor<fp32, [1, 16, ?, 64]> key_21 = concat(axis = var_689, interleave = key_21_interleave_0, values = (var_394, key_states_71))[name = tensor<string, []>("key_21")]; |
| tensor<bool, []> value_21_interleave_0 = const()[name = tensor<string, []>("value_21_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_71 = transpose(perm = value_states_71_perm_0, x = var_1396)[name = tensor<string, []>("transpose_39")]; |
| tensor<fp32, [1, 16, ?, 64]> value_21 = concat(axis = var_689, interleave = value_21_interleave_0, values = (var_397, value_states_71))[name = tensor<string, []>("value_21")]; |
| tensor<int32, [4]> var_1402_shape = shape(x = key_21)[name = tensor<string, []>("op_1402_shape")]; |
| tensor<int32, []> gather_15_indices_0 = const()[name = tensor<string, []>("gather_15_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_15_axis_0 = const()[name = tensor<string, []>("gather_15_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_15_batch_dims_0 = const()[name = tensor<string, []>("gather_15_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_15 = gather(axis = gather_15_axis_0, batch_dims = gather_15_batch_dims_0, indices = gather_15_indices_0, x = var_1402_shape)[name = tensor<string, []>("gather_15")]; |
| tensor<int32, []> concat_19_values0_0 = const()[name = tensor<string, []>("concat_19_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_19_values1_0 = const()[name = tensor<string, []>("concat_19_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_19_values2_0 = const()[name = tensor<string, []>("concat_19_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_19_axis_0 = const()[name = tensor<string, []>("concat_19_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_19_interleave_0 = const()[name = tensor<string, []>("concat_19_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_19 = concat(axis = concat_19_axis_0, interleave = concat_19_interleave_0, values = (concat_19_values0_0, concat_19_values1_0, concat_19_values2_0, gather_15))[name = tensor<string, []>("concat_19")]; |
| tensor<int32, [4]> attention_mask_25_begin_0 = const()[name = tensor<string, []>("attention_mask_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_25_end_mask_0 = const()[name = tensor<string, []>("attention_mask_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_25 = slice_by_index(begin = attention_mask_25_begin_0, end = concat_19, end_mask = attention_mask_25_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_25")]; |
| tensor<fp32, [1, 16, 1, 64]> query_21 = transpose(perm = query_21_perm_0, x = var_1384)[name = tensor<string, []>("transpose_41")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_10 = mul(x = query_21, y = var_687)[name = tensor<string, []>("mul_10")]; |
| tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_10 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = mul_10, y = key_21)[name = tensor<string, []>("matmul_10")]; |
| tensor<fp32, [?, 16, ?, ?]> add_10 = add(x = matmul_10, y = attention_mask_25)[name = tensor<string, []>("add_10")]; |
| tensor<int32, []> softmax_10_axis_0 = const()[name = tensor<string, []>("softmax_10_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_10 = softmax(axis = softmax_10_axis_0, x = add_10)[name = tensor<string, []>("softmax_10")]; |
| tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_41 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = softmax_10, y = value_21)[name = tensor<string, []>("attn_output_41")]; |
| tensor<int32, [4]> var_1408_perm_0 = const()[name = tensor<string, []>("op_1408_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1410 = const()[name = tensor<string, []>("op_1410"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1408 = transpose(perm = var_1408_perm_0, x = attn_output_41)[name = tensor<string, []>("transpose_38")]; |
| tensor<fp32, [1, 1, ?]> var_1411 = reshape(shape = var_1410, x = var_1408)[name = tensor<string, []>("op_1411")]; |
| tensor<fp32, [1, 1, 1024]> input_129 = linear(bias = decoder_layers_5_self_attn_out_proj_bias, weight = decoder_layers_5_self_attn_out_proj_weight, x = var_1411)[name = tensor<string, []>("linear_43")]; |
| tensor<fp32, [1, 1, 1024]> input_131 = add(x = input_125, y = input_129)[name = tensor<string, []>("input_131")]; |
| tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_55 = layer_norm(axes = hidden_states_55_axes_0, beta = decoder_layers_5_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_5_encoder_attn_layer_norm_weight, x = input_131)[name = tensor<string, []>("hidden_states_55")]; |
| tensor<fp32, [1, 1, 1024]> var_1432 = linear(bias = decoder_layers_5_encoder_attn_q_proj_bias, weight = decoder_layers_5_encoder_attn_q_proj_weight, x = hidden_states_55)[name = tensor<string, []>("linear_44")]; |
| tensor<int32, [4]> var_1433 = const()[name = tensor<string, []>("op_1433"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1434 = reshape(shape = var_1433, x = var_1432)[name = tensor<string, []>("op_1434")]; |
| tensor<int32, [4]> query_23_perm_0 = const()[name = tensor<string, []>("query_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1436_shape = shape(x = key_23)[name = tensor<string, []>("op_1436_shape")]; |
| tensor<int32, []> gather_16_indices_0 = const()[name = tensor<string, []>("gather_16_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_16_axis_0 = const()[name = tensor<string, []>("gather_16_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_16_batch_dims_0 = const()[name = tensor<string, []>("gather_16_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_16 = gather(axis = gather_16_axis_0, batch_dims = gather_16_batch_dims_0, indices = gather_16_indices_0, x = var_1436_shape)[name = tensor<string, []>("gather_16")]; |
| tensor<int32, []> concat_20_values0_0 = const()[name = tensor<string, []>("concat_20_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_20_values1_0 = const()[name = tensor<string, []>("concat_20_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_20_values2_0 = const()[name = tensor<string, []>("concat_20_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_20_axis_0 = const()[name = tensor<string, []>("concat_20_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_20_interleave_0 = const()[name = tensor<string, []>("concat_20_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_20 = concat(axis = concat_20_axis_0, interleave = concat_20_interleave_0, values = (concat_20_values0_0, concat_20_values1_0, concat_20_values2_0, gather_16))[name = tensor<string, []>("concat_20")]; |
| tensor<int32, [4]> attention_mask_27_begin_0 = const()[name = tensor<string, []>("attention_mask_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_27_end_mask_0 = const()[name = tensor<string, []>("attention_mask_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_27 = slice_by_index(begin = attention_mask_27_begin_0, end = concat_20, end_mask = attention_mask_27_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_27")]; |
| tensor<fp32, [1, 16, 1, 64]> query_23 = transpose(perm = query_23_perm_0, x = var_1434)[name = tensor<string, []>("transpose_37")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_11 = mul(x = query_23, y = var_687)[name = tensor<string, []>("mul_11")]; |
| tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_11 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = mul_11, y = key_23)[name = tensor<string, []>("matmul_11")]; |
| tensor<fp32, [?, 16, ?, ?]> add_11 = add(x = matmul_11, y = attention_mask_27)[name = tensor<string, []>("add_11")]; |
| tensor<int32, []> softmax_11_axis_0 = const()[name = tensor<string, []>("softmax_11_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_11 = softmax(axis = softmax_11_axis_0, x = add_11)[name = tensor<string, []>("softmax_11")]; |
| tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_45 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = softmax_11, y = value_23)[name = tensor<string, []>("attn_output_45")]; |
| tensor<int32, [4]> var_1442_perm_0 = const()[name = tensor<string, []>("op_1442_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1444 = const()[name = tensor<string, []>("op_1444"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1442 = transpose(perm = var_1442_perm_0, x = attn_output_45)[name = tensor<string, []>("transpose_36")]; |
| tensor<fp32, [1, 1, ?]> var_1445 = reshape(shape = var_1444, x = var_1442)[name = tensor<string, []>("op_1445")]; |
| tensor<fp32, [1, 1, 1024]> input_135 = linear(bias = decoder_layers_5_encoder_attn_out_proj_bias, weight = decoder_layers_5_encoder_attn_out_proj_weight, x = var_1445)[name = tensor<string, []>("linear_45")]; |
| tensor<fp32, [1, 1, 1024]> input_137 = add(x = input_131, y = input_135)[name = tensor<string, []>("input_137")]; |
| tensor<int32, [1]> input_139_axes_0 = const()[name = tensor<string, []>("input_139_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_139 = layer_norm(axes = input_139_axes_0, beta = decoder_layers_5_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_5_final_layer_norm_weight, x = input_137)[name = tensor<string, []>("input_139")]; |
| tensor<fp32, [1, 1, 4096]> input_141 = linear(bias = decoder_layers_5_fc1_bias, weight = decoder_layers_5_fc1_weight, x = input_139)[name = tensor<string, []>("linear_46")]; |
| tensor<fp32, [1, 1, 4096]> input_143 = relu(x = input_141)[name = tensor<string, []>("input_143")]; |
| tensor<fp32, [1, 1, 1024]> input_147 = linear(bias = decoder_layers_5_fc2_bias, weight = decoder_layers_5_fc2_weight, x = input_143)[name = tensor<string, []>("linear_47")]; |
| tensor<fp32, [1, 1, 1024]> input_149 = add(x = input_137, y = input_147)[name = tensor<string, []>("input_149")]; |
| tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_61 = layer_norm(axes = hidden_states_61_axes_0, beta = decoder_layers_6_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_6_self_attn_layer_norm_weight, x = input_149)[name = tensor<string, []>("hidden_states_61")]; |
| tensor<fp32, [1, 1, 1024]> var_1489 = linear(bias = decoder_layers_6_self_attn_q_proj_bias, weight = decoder_layers_6_self_attn_q_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_48")]; |
| tensor<int32, [4]> var_1490 = const()[name = tensor<string, []>("op_1490"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1491 = reshape(shape = var_1490, x = var_1489)[name = tensor<string, []>("op_1491")]; |
| tensor<int32, [4]> query_25_perm_0 = const()[name = tensor<string, []>("query_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_73 = linear(bias = decoder_layers_6_self_attn_k_proj_bias, weight = decoder_layers_6_self_attn_k_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_49")]; |
| tensor<fp32, [1, 1, 1024]> value_states_73 = linear(bias = decoder_layers_6_self_attn_v_proj_bias, weight = decoder_layers_6_self_attn_v_proj_weight, x = hidden_states_61)[name = tensor<string, []>("linear_50")]; |
| tensor<int32, [4]> var_1499 = const()[name = tensor<string, []>("op_1499"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1500 = reshape(shape = var_1499, x = key_states_73)[name = tensor<string, []>("op_1500")]; |
| tensor<int32, [4]> key_states_75_perm_0 = const()[name = tensor<string, []>("key_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1502 = const()[name = tensor<string, []>("op_1502"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1503 = reshape(shape = var_1502, x = value_states_73)[name = tensor<string, []>("op_1503")]; |
| tensor<int32, [4]> value_states_75_perm_0 = const()[name = tensor<string, []>("value_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_25_interleave_0 = const()[name = tensor<string, []>("key_25_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_75 = transpose(perm = key_states_75_perm_0, x = var_1500)[name = tensor<string, []>("transpose_34")]; |
| tensor<fp32, [1, 16, ?, 64]> key_25 = concat(axis = var_689, interleave = key_25_interleave_0, values = (var_438, key_states_75))[name = tensor<string, []>("key_25")]; |
| tensor<bool, []> value_25_interleave_0 = const()[name = tensor<string, []>("value_25_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_75 = transpose(perm = value_states_75_perm_0, x = var_1503)[name = tensor<string, []>("transpose_33")]; |
| tensor<fp32, [1, 16, ?, 64]> value_25 = concat(axis = var_689, interleave = value_25_interleave_0, values = (var_441, value_states_75))[name = tensor<string, []>("value_25")]; |
| tensor<int32, [4]> var_1509_shape = shape(x = key_25)[name = tensor<string, []>("op_1509_shape")]; |
| tensor<int32, []> gather_17_indices_0 = const()[name = tensor<string, []>("gather_17_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_17_axis_0 = const()[name = tensor<string, []>("gather_17_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_17_batch_dims_0 = const()[name = tensor<string, []>("gather_17_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_17 = gather(axis = gather_17_axis_0, batch_dims = gather_17_batch_dims_0, indices = gather_17_indices_0, x = var_1509_shape)[name = tensor<string, []>("gather_17")]; |
| tensor<int32, []> concat_21_values0_0 = const()[name = tensor<string, []>("concat_21_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_21_values1_0 = const()[name = tensor<string, []>("concat_21_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_21_values2_0 = const()[name = tensor<string, []>("concat_21_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_21_axis_0 = const()[name = tensor<string, []>("concat_21_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_21_interleave_0 = const()[name = tensor<string, []>("concat_21_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_21 = concat(axis = concat_21_axis_0, interleave = concat_21_interleave_0, values = (concat_21_values0_0, concat_21_values1_0, concat_21_values2_0, gather_17))[name = tensor<string, []>("concat_21")]; |
| tensor<int32, [4]> attention_mask_29_begin_0 = const()[name = tensor<string, []>("attention_mask_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_29_end_mask_0 = const()[name = tensor<string, []>("attention_mask_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_29 = slice_by_index(begin = attention_mask_29_begin_0, end = concat_21, end_mask = attention_mask_29_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_29")]; |
| tensor<fp32, [1, 16, 1, 64]> query_25 = transpose(perm = query_25_perm_0, x = var_1491)[name = tensor<string, []>("transpose_35")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_12 = mul(x = query_25, y = var_687)[name = tensor<string, []>("mul_12")]; |
| tensor<bool, []> matmul_12_transpose_y_0 = const()[name = tensor<string, []>("matmul_12_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_12_transpose_x_0 = const()[name = tensor<string, []>("matmul_12_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_12 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = mul_12, y = key_25)[name = tensor<string, []>("matmul_12")]; |
| tensor<fp32, [?, 16, ?, ?]> add_12 = add(x = matmul_12, y = attention_mask_29)[name = tensor<string, []>("add_12")]; |
| tensor<int32, []> softmax_12_axis_0 = const()[name = tensor<string, []>("softmax_12_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_12 = softmax(axis = softmax_12_axis_0, x = add_12)[name = tensor<string, []>("softmax_12")]; |
| tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_49 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = softmax_12, y = value_25)[name = tensor<string, []>("attn_output_49")]; |
| tensor<int32, [4]> var_1515_perm_0 = const()[name = tensor<string, []>("op_1515_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1517 = const()[name = tensor<string, []>("op_1517"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1515 = transpose(perm = var_1515_perm_0, x = attn_output_49)[name = tensor<string, []>("transpose_32")]; |
| tensor<fp32, [1, 1, ?]> var_1518 = reshape(shape = var_1517, x = var_1515)[name = tensor<string, []>("op_1518")]; |
| tensor<fp32, [1, 1, 1024]> input_153 = linear(bias = decoder_layers_6_self_attn_out_proj_bias, weight = decoder_layers_6_self_attn_out_proj_weight, x = var_1518)[name = tensor<string, []>("linear_51")]; |
| tensor<fp32, [1, 1, 1024]> input_155 = add(x = input_149, y = input_153)[name = tensor<string, []>("input_155")]; |
| tensor<int32, [1]> hidden_states_65_axes_0 = const()[name = tensor<string, []>("hidden_states_65_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_65 = layer_norm(axes = hidden_states_65_axes_0, beta = decoder_layers_6_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_6_encoder_attn_layer_norm_weight, x = input_155)[name = tensor<string, []>("hidden_states_65")]; |
| tensor<fp32, [1, 1, 1024]> var_1539 = linear(bias = decoder_layers_6_encoder_attn_q_proj_bias, weight = decoder_layers_6_encoder_attn_q_proj_weight, x = hidden_states_65)[name = tensor<string, []>("linear_52")]; |
| tensor<int32, [4]> var_1540 = const()[name = tensor<string, []>("op_1540"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1541 = reshape(shape = var_1540, x = var_1539)[name = tensor<string, []>("op_1541")]; |
| tensor<int32, [4]> query_27_perm_0 = const()[name = tensor<string, []>("query_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1543_shape = shape(x = key_27)[name = tensor<string, []>("op_1543_shape")]; |
| tensor<int32, []> gather_18_indices_0 = const()[name = tensor<string, []>("gather_18_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_18_axis_0 = const()[name = tensor<string, []>("gather_18_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_18_batch_dims_0 = const()[name = tensor<string, []>("gather_18_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_18 = gather(axis = gather_18_axis_0, batch_dims = gather_18_batch_dims_0, indices = gather_18_indices_0, x = var_1543_shape)[name = tensor<string, []>("gather_18")]; |
| tensor<int32, []> concat_22_values0_0 = const()[name = tensor<string, []>("concat_22_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_22_values1_0 = const()[name = tensor<string, []>("concat_22_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_22_values2_0 = const()[name = tensor<string, []>("concat_22_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_22_axis_0 = const()[name = tensor<string, []>("concat_22_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_22_interleave_0 = const()[name = tensor<string, []>("concat_22_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_22 = concat(axis = concat_22_axis_0, interleave = concat_22_interleave_0, values = (concat_22_values0_0, concat_22_values1_0, concat_22_values2_0, gather_18))[name = tensor<string, []>("concat_22")]; |
| tensor<int32, [4]> attention_mask_31_begin_0 = const()[name = tensor<string, []>("attention_mask_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_31_end_mask_0 = const()[name = tensor<string, []>("attention_mask_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_31 = slice_by_index(begin = attention_mask_31_begin_0, end = concat_22, end_mask = attention_mask_31_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_31")]; |
| tensor<fp32, [1, 16, 1, 64]> query_27 = transpose(perm = query_27_perm_0, x = var_1541)[name = tensor<string, []>("transpose_31")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_13 = mul(x = query_27, y = var_687)[name = tensor<string, []>("mul_13")]; |
| tensor<bool, []> matmul_13_transpose_y_0 = const()[name = tensor<string, []>("matmul_13_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_13_transpose_x_0 = const()[name = tensor<string, []>("matmul_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_13 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = mul_13, y = key_27)[name = tensor<string, []>("matmul_13")]; |
| tensor<fp32, [?, 16, ?, ?]> add_13 = add(x = matmul_13, y = attention_mask_31)[name = tensor<string, []>("add_13")]; |
| tensor<int32, []> softmax_13_axis_0 = const()[name = tensor<string, []>("softmax_13_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_13 = softmax(axis = softmax_13_axis_0, x = add_13)[name = tensor<string, []>("softmax_13")]; |
| tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_53 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = softmax_13, y = value_27)[name = tensor<string, []>("attn_output_53")]; |
| tensor<int32, [4]> var_1549_perm_0 = const()[name = tensor<string, []>("op_1549_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1551 = const()[name = tensor<string, []>("op_1551"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1549 = transpose(perm = var_1549_perm_0, x = attn_output_53)[name = tensor<string, []>("transpose_30")]; |
| tensor<fp32, [1, 1, ?]> var_1552 = reshape(shape = var_1551, x = var_1549)[name = tensor<string, []>("op_1552")]; |
| tensor<fp32, [1, 1, 1024]> input_159 = linear(bias = decoder_layers_6_encoder_attn_out_proj_bias, weight = decoder_layers_6_encoder_attn_out_proj_weight, x = var_1552)[name = tensor<string, []>("linear_53")]; |
| tensor<fp32, [1, 1, 1024]> input_161 = add(x = input_155, y = input_159)[name = tensor<string, []>("input_161")]; |
| tensor<int32, [1]> input_163_axes_0 = const()[name = tensor<string, []>("input_163_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_163 = layer_norm(axes = input_163_axes_0, beta = decoder_layers_6_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_6_final_layer_norm_weight, x = input_161)[name = tensor<string, []>("input_163")]; |
| tensor<fp32, [1, 1, 4096]> input_165 = linear(bias = decoder_layers_6_fc1_bias, weight = decoder_layers_6_fc1_weight, x = input_163)[name = tensor<string, []>("linear_54")]; |
| tensor<fp32, [1, 1, 4096]> input_167 = relu(x = input_165)[name = tensor<string, []>("input_167")]; |
| tensor<fp32, [1, 1, 1024]> input_171 = linear(bias = decoder_layers_6_fc2_bias, weight = decoder_layers_6_fc2_weight, x = input_167)[name = tensor<string, []>("linear_55")]; |
| tensor<fp32, [1, 1, 1024]> input_173 = add(x = input_161, y = input_171)[name = tensor<string, []>("input_173")]; |
| tensor<int32, [1]> hidden_states_71_axes_0 = const()[name = tensor<string, []>("hidden_states_71_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_71 = layer_norm(axes = hidden_states_71_axes_0, beta = decoder_layers_7_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_7_self_attn_layer_norm_weight, x = input_173)[name = tensor<string, []>("hidden_states_71")]; |
| tensor<fp32, [1, 1, 1024]> var_1596 = linear(bias = decoder_layers_7_self_attn_q_proj_bias, weight = decoder_layers_7_self_attn_q_proj_weight, x = hidden_states_71)[name = tensor<string, []>("linear_56")]; |
| tensor<int32, [4]> var_1597 = const()[name = tensor<string, []>("op_1597"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1598 = reshape(shape = var_1597, x = var_1596)[name = tensor<string, []>("op_1598")]; |
| tensor<int32, [4]> query_29_perm_0 = const()[name = tensor<string, []>("query_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_77 = linear(bias = decoder_layers_7_self_attn_k_proj_bias, weight = decoder_layers_7_self_attn_k_proj_weight, x = hidden_states_71)[name = tensor<string, []>("linear_57")]; |
| tensor<fp32, [1, 1, 1024]> value_states_77 = linear(bias = decoder_layers_7_self_attn_v_proj_bias, weight = decoder_layers_7_self_attn_v_proj_weight, x = hidden_states_71)[name = tensor<string, []>("linear_58")]; |
| tensor<int32, [4]> var_1606 = const()[name = tensor<string, []>("op_1606"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1607 = reshape(shape = var_1606, x = key_states_77)[name = tensor<string, []>("op_1607")]; |
| tensor<int32, [4]> key_states_79_perm_0 = const()[name = tensor<string, []>("key_states_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1609 = const()[name = tensor<string, []>("op_1609"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1610 = reshape(shape = var_1609, x = value_states_77)[name = tensor<string, []>("op_1610")]; |
| tensor<int32, [4]> value_states_79_perm_0 = const()[name = tensor<string, []>("value_states_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_29_interleave_0 = const()[name = tensor<string, []>("key_29_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_79 = transpose(perm = key_states_79_perm_0, x = var_1607)[name = tensor<string, []>("transpose_28")]; |
| tensor<fp32, [1, 16, ?, 64]> key_29 = concat(axis = var_689, interleave = key_29_interleave_0, values = (var_482, key_states_79))[name = tensor<string, []>("key_29")]; |
| tensor<bool, []> value_29_interleave_0 = const()[name = tensor<string, []>("value_29_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_79 = transpose(perm = value_states_79_perm_0, x = var_1610)[name = tensor<string, []>("transpose_27")]; |
| tensor<fp32, [1, 16, ?, 64]> value_29 = concat(axis = var_689, interleave = value_29_interleave_0, values = (var_485, value_states_79))[name = tensor<string, []>("value_29")]; |
| tensor<int32, [4]> var_1616_shape = shape(x = key_29)[name = tensor<string, []>("op_1616_shape")]; |
| tensor<int32, []> gather_19_indices_0 = const()[name = tensor<string, []>("gather_19_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_19_axis_0 = const()[name = tensor<string, []>("gather_19_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_19_batch_dims_0 = const()[name = tensor<string, []>("gather_19_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_19 = gather(axis = gather_19_axis_0, batch_dims = gather_19_batch_dims_0, indices = gather_19_indices_0, x = var_1616_shape)[name = tensor<string, []>("gather_19")]; |
| tensor<int32, []> concat_23_values0_0 = const()[name = tensor<string, []>("concat_23_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_23_values1_0 = const()[name = tensor<string, []>("concat_23_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_23_values2_0 = const()[name = tensor<string, []>("concat_23_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_23_axis_0 = const()[name = tensor<string, []>("concat_23_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_23_interleave_0 = const()[name = tensor<string, []>("concat_23_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_23 = concat(axis = concat_23_axis_0, interleave = concat_23_interleave_0, values = (concat_23_values0_0, concat_23_values1_0, concat_23_values2_0, gather_19))[name = tensor<string, []>("concat_23")]; |
| tensor<int32, [4]> attention_mask_33_begin_0 = const()[name = tensor<string, []>("attention_mask_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_33_end_mask_0 = const()[name = tensor<string, []>("attention_mask_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_33 = slice_by_index(begin = attention_mask_33_begin_0, end = concat_23, end_mask = attention_mask_33_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_33")]; |
| tensor<fp32, [1, 16, 1, 64]> query_29 = transpose(perm = query_29_perm_0, x = var_1598)[name = tensor<string, []>("transpose_29")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_14 = mul(x = query_29, y = var_687)[name = tensor<string, []>("mul_14")]; |
| tensor<bool, []> matmul_14_transpose_y_0 = const()[name = tensor<string, []>("matmul_14_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_14_transpose_x_0 = const()[name = tensor<string, []>("matmul_14_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_14 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = mul_14, y = key_29)[name = tensor<string, []>("matmul_14")]; |
| tensor<fp32, [?, 16, ?, ?]> add_14 = add(x = matmul_14, y = attention_mask_33)[name = tensor<string, []>("add_14")]; |
| tensor<int32, []> softmax_14_axis_0 = const()[name = tensor<string, []>("softmax_14_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_14 = softmax(axis = softmax_14_axis_0, x = add_14)[name = tensor<string, []>("softmax_14")]; |
| tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_57 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = softmax_14, y = value_29)[name = tensor<string, []>("attn_output_57")]; |
| tensor<int32, [4]> var_1622_perm_0 = const()[name = tensor<string, []>("op_1622_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1624 = const()[name = tensor<string, []>("op_1624"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1622 = transpose(perm = var_1622_perm_0, x = attn_output_57)[name = tensor<string, []>("transpose_26")]; |
| tensor<fp32, [1, 1, ?]> var_1625 = reshape(shape = var_1624, x = var_1622)[name = tensor<string, []>("op_1625")]; |
| tensor<fp32, [1, 1, 1024]> input_177 = linear(bias = decoder_layers_7_self_attn_out_proj_bias, weight = decoder_layers_7_self_attn_out_proj_weight, x = var_1625)[name = tensor<string, []>("linear_59")]; |
| tensor<fp32, [1, 1, 1024]> input_179 = add(x = input_173, y = input_177)[name = tensor<string, []>("input_179")]; |
| tensor<int32, [1]> hidden_states_75_axes_0 = const()[name = tensor<string, []>("hidden_states_75_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_75 = layer_norm(axes = hidden_states_75_axes_0, beta = decoder_layers_7_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_7_encoder_attn_layer_norm_weight, x = input_179)[name = tensor<string, []>("hidden_states_75")]; |
| tensor<fp32, [1, 1, 1024]> var_1646 = linear(bias = decoder_layers_7_encoder_attn_q_proj_bias, weight = decoder_layers_7_encoder_attn_q_proj_weight, x = hidden_states_75)[name = tensor<string, []>("linear_60")]; |
| tensor<int32, [4]> var_1647 = const()[name = tensor<string, []>("op_1647"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1648 = reshape(shape = var_1647, x = var_1646)[name = tensor<string, []>("op_1648")]; |
| tensor<int32, [4]> query_31_perm_0 = const()[name = tensor<string, []>("query_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1650_shape = shape(x = key_31)[name = tensor<string, []>("op_1650_shape")]; |
| tensor<int32, []> gather_20_indices_0 = const()[name = tensor<string, []>("gather_20_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_20_axis_0 = const()[name = tensor<string, []>("gather_20_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_20_batch_dims_0 = const()[name = tensor<string, []>("gather_20_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_20 = gather(axis = gather_20_axis_0, batch_dims = gather_20_batch_dims_0, indices = gather_20_indices_0, x = var_1650_shape)[name = tensor<string, []>("gather_20")]; |
| tensor<int32, []> concat_24_values0_0 = const()[name = tensor<string, []>("concat_24_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_24_values1_0 = const()[name = tensor<string, []>("concat_24_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_24_values2_0 = const()[name = tensor<string, []>("concat_24_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_24_axis_0 = const()[name = tensor<string, []>("concat_24_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_24_interleave_0 = const()[name = tensor<string, []>("concat_24_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_24 = concat(axis = concat_24_axis_0, interleave = concat_24_interleave_0, values = (concat_24_values0_0, concat_24_values1_0, concat_24_values2_0, gather_20))[name = tensor<string, []>("concat_24")]; |
| tensor<int32, [4]> attention_mask_35_begin_0 = const()[name = tensor<string, []>("attention_mask_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_35_end_mask_0 = const()[name = tensor<string, []>("attention_mask_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_35 = slice_by_index(begin = attention_mask_35_begin_0, end = concat_24, end_mask = attention_mask_35_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_35")]; |
| tensor<fp32, [1, 16, 1, 64]> query_31 = transpose(perm = query_31_perm_0, x = var_1648)[name = tensor<string, []>("transpose_25")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_15 = mul(x = query_31, y = var_687)[name = tensor<string, []>("mul_15")]; |
| tensor<bool, []> matmul_15_transpose_y_0 = const()[name = tensor<string, []>("matmul_15_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_15_transpose_x_0 = const()[name = tensor<string, []>("matmul_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_15 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = mul_15, y = key_31)[name = tensor<string, []>("matmul_15")]; |
| tensor<fp32, [?, 16, ?, ?]> add_15 = add(x = matmul_15, y = attention_mask_35)[name = tensor<string, []>("add_15")]; |
| tensor<int32, []> softmax_15_axis_0 = const()[name = tensor<string, []>("softmax_15_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_15 = softmax(axis = softmax_15_axis_0, x = add_15)[name = tensor<string, []>("softmax_15")]; |
| tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_61 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = softmax_15, y = value_31)[name = tensor<string, []>("attn_output_61")]; |
| tensor<int32, [4]> var_1656_perm_0 = const()[name = tensor<string, []>("op_1656_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1658 = const()[name = tensor<string, []>("op_1658"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1656 = transpose(perm = var_1656_perm_0, x = attn_output_61)[name = tensor<string, []>("transpose_24")]; |
| tensor<fp32, [1, 1, ?]> var_1659 = reshape(shape = var_1658, x = var_1656)[name = tensor<string, []>("op_1659")]; |
| tensor<fp32, [1, 1, 1024]> input_183 = linear(bias = decoder_layers_7_encoder_attn_out_proj_bias, weight = decoder_layers_7_encoder_attn_out_proj_weight, x = var_1659)[name = tensor<string, []>("linear_61")]; |
| tensor<fp32, [1, 1, 1024]> input_185 = add(x = input_179, y = input_183)[name = tensor<string, []>("input_185")]; |
| tensor<int32, [1]> input_187_axes_0 = const()[name = tensor<string, []>("input_187_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_187 = layer_norm(axes = input_187_axes_0, beta = decoder_layers_7_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_7_final_layer_norm_weight, x = input_185)[name = tensor<string, []>("input_187")]; |
| tensor<fp32, [1, 1, 4096]> input_189 = linear(bias = decoder_layers_7_fc1_bias, weight = decoder_layers_7_fc1_weight, x = input_187)[name = tensor<string, []>("linear_62")]; |
| tensor<fp32, [1, 1, 4096]> input_191 = relu(x = input_189)[name = tensor<string, []>("input_191")]; |
| tensor<fp32, [1, 1, 1024]> input_195 = linear(bias = decoder_layers_7_fc2_bias, weight = decoder_layers_7_fc2_weight, x = input_191)[name = tensor<string, []>("linear_63")]; |
| tensor<fp32, [1, 1, 1024]> input_197 = add(x = input_185, y = input_195)[name = tensor<string, []>("input_197")]; |
| tensor<int32, [1]> hidden_states_81_axes_0 = const()[name = tensor<string, []>("hidden_states_81_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_81 = layer_norm(axes = hidden_states_81_axes_0, beta = decoder_layers_8_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_8_self_attn_layer_norm_weight, x = input_197)[name = tensor<string, []>("hidden_states_81")]; |
| tensor<fp32, [1, 1, 1024]> var_1703 = linear(bias = decoder_layers_8_self_attn_q_proj_bias, weight = decoder_layers_8_self_attn_q_proj_weight, x = hidden_states_81)[name = tensor<string, []>("linear_64")]; |
| tensor<int32, [4]> var_1704 = const()[name = tensor<string, []>("op_1704"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1705 = reshape(shape = var_1704, x = var_1703)[name = tensor<string, []>("op_1705")]; |
| tensor<int32, [4]> query_33_perm_0 = const()[name = tensor<string, []>("query_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_81 = linear(bias = decoder_layers_8_self_attn_k_proj_bias, weight = decoder_layers_8_self_attn_k_proj_weight, x = hidden_states_81)[name = tensor<string, []>("linear_65")]; |
| tensor<fp32, [1, 1, 1024]> value_states_81 = linear(bias = decoder_layers_8_self_attn_v_proj_bias, weight = decoder_layers_8_self_attn_v_proj_weight, x = hidden_states_81)[name = tensor<string, []>("linear_66")]; |
| tensor<int32, [4]> var_1713 = const()[name = tensor<string, []>("op_1713"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1714 = reshape(shape = var_1713, x = key_states_81)[name = tensor<string, []>("op_1714")]; |
| tensor<int32, [4]> key_states_83_perm_0 = const()[name = tensor<string, []>("key_states_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1716 = const()[name = tensor<string, []>("op_1716"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1717 = reshape(shape = var_1716, x = value_states_81)[name = tensor<string, []>("op_1717")]; |
| tensor<int32, [4]> value_states_83_perm_0 = const()[name = tensor<string, []>("value_states_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_33_interleave_0 = const()[name = tensor<string, []>("key_33_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_83 = transpose(perm = key_states_83_perm_0, x = var_1714)[name = tensor<string, []>("transpose_22")]; |
| tensor<fp32, [1, 16, ?, 64]> key_33 = concat(axis = var_689, interleave = key_33_interleave_0, values = (var_526, key_states_83))[name = tensor<string, []>("key_33")]; |
| tensor<bool, []> value_33_interleave_0 = const()[name = tensor<string, []>("value_33_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_83 = transpose(perm = value_states_83_perm_0, x = var_1717)[name = tensor<string, []>("transpose_21")]; |
| tensor<fp32, [1, 16, ?, 64]> value_33 = concat(axis = var_689, interleave = value_33_interleave_0, values = (var_529, value_states_83))[name = tensor<string, []>("value_33")]; |
| tensor<int32, [4]> var_1723_shape = shape(x = key_33)[name = tensor<string, []>("op_1723_shape")]; |
| tensor<int32, []> gather_21_indices_0 = const()[name = tensor<string, []>("gather_21_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_21_axis_0 = const()[name = tensor<string, []>("gather_21_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_21_batch_dims_0 = const()[name = tensor<string, []>("gather_21_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_21 = gather(axis = gather_21_axis_0, batch_dims = gather_21_batch_dims_0, indices = gather_21_indices_0, x = var_1723_shape)[name = tensor<string, []>("gather_21")]; |
| tensor<int32, []> concat_25_values0_0 = const()[name = tensor<string, []>("concat_25_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_25_values1_0 = const()[name = tensor<string, []>("concat_25_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_25_values2_0 = const()[name = tensor<string, []>("concat_25_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_25_axis_0 = const()[name = tensor<string, []>("concat_25_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_25_interleave_0 = const()[name = tensor<string, []>("concat_25_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_25 = concat(axis = concat_25_axis_0, interleave = concat_25_interleave_0, values = (concat_25_values0_0, concat_25_values1_0, concat_25_values2_0, gather_21))[name = tensor<string, []>("concat_25")]; |
| tensor<int32, [4]> attention_mask_37_begin_0 = const()[name = tensor<string, []>("attention_mask_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_37_end_mask_0 = const()[name = tensor<string, []>("attention_mask_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_37 = slice_by_index(begin = attention_mask_37_begin_0, end = concat_25, end_mask = attention_mask_37_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_37")]; |
| tensor<fp32, [1, 16, 1, 64]> query_33 = transpose(perm = query_33_perm_0, x = var_1705)[name = tensor<string, []>("transpose_23")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_16 = mul(x = query_33, y = var_687)[name = tensor<string, []>("mul_16")]; |
| tensor<bool, []> matmul_16_transpose_y_0 = const()[name = tensor<string, []>("matmul_16_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_16_transpose_x_0 = const()[name = tensor<string, []>("matmul_16_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = mul_16, y = key_33)[name = tensor<string, []>("matmul_16")]; |
| tensor<fp32, [?, 16, ?, ?]> add_16 = add(x = matmul_16, y = attention_mask_37)[name = tensor<string, []>("add_16")]; |
| tensor<int32, []> softmax_16_axis_0 = const()[name = tensor<string, []>("softmax_16_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_16 = softmax(axis = softmax_16_axis_0, x = add_16)[name = tensor<string, []>("softmax_16")]; |
| tensor<bool, []> attn_output_65_transpose_x_0 = const()[name = tensor<string, []>("attn_output_65_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_65_transpose_y_0 = const()[name = tensor<string, []>("attn_output_65_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_65 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = softmax_16, y = value_33)[name = tensor<string, []>("attn_output_65")]; |
| tensor<int32, [4]> var_1729_perm_0 = const()[name = tensor<string, []>("op_1729_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1731 = const()[name = tensor<string, []>("op_1731"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1729 = transpose(perm = var_1729_perm_0, x = attn_output_65)[name = tensor<string, []>("transpose_20")]; |
| tensor<fp32, [1, 1, ?]> var_1732 = reshape(shape = var_1731, x = var_1729)[name = tensor<string, []>("op_1732")]; |
| tensor<fp32, [1, 1, 1024]> input_201 = linear(bias = decoder_layers_8_self_attn_out_proj_bias, weight = decoder_layers_8_self_attn_out_proj_weight, x = var_1732)[name = tensor<string, []>("linear_67")]; |
| tensor<fp32, [1, 1, 1024]> input_203 = add(x = input_197, y = input_201)[name = tensor<string, []>("input_203")]; |
| tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_85 = layer_norm(axes = hidden_states_85_axes_0, beta = decoder_layers_8_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_8_encoder_attn_layer_norm_weight, x = input_203)[name = tensor<string, []>("hidden_states_85")]; |
| tensor<fp32, [1, 1, 1024]> var_1753 = linear(bias = decoder_layers_8_encoder_attn_q_proj_bias, weight = decoder_layers_8_encoder_attn_q_proj_weight, x = hidden_states_85)[name = tensor<string, []>("linear_68")]; |
| tensor<int32, [4]> var_1754 = const()[name = tensor<string, []>("op_1754"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1755 = reshape(shape = var_1754, x = var_1753)[name = tensor<string, []>("op_1755")]; |
| tensor<int32, [4]> query_35_perm_0 = const()[name = tensor<string, []>("query_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1757_shape = shape(x = key_35)[name = tensor<string, []>("op_1757_shape")]; |
| tensor<int32, []> gather_22_indices_0 = const()[name = tensor<string, []>("gather_22_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_22_axis_0 = const()[name = tensor<string, []>("gather_22_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_22_batch_dims_0 = const()[name = tensor<string, []>("gather_22_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_22 = gather(axis = gather_22_axis_0, batch_dims = gather_22_batch_dims_0, indices = gather_22_indices_0, x = var_1757_shape)[name = tensor<string, []>("gather_22")]; |
| tensor<int32, []> concat_26_values0_0 = const()[name = tensor<string, []>("concat_26_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_26_values1_0 = const()[name = tensor<string, []>("concat_26_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_26_values2_0 = const()[name = tensor<string, []>("concat_26_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_26_axis_0 = const()[name = tensor<string, []>("concat_26_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_26_interleave_0 = const()[name = tensor<string, []>("concat_26_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_26 = concat(axis = concat_26_axis_0, interleave = concat_26_interleave_0, values = (concat_26_values0_0, concat_26_values1_0, concat_26_values2_0, gather_22))[name = tensor<string, []>("concat_26")]; |
| tensor<int32, [4]> attention_mask_39_begin_0 = const()[name = tensor<string, []>("attention_mask_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_39_end_mask_0 = const()[name = tensor<string, []>("attention_mask_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_39 = slice_by_index(begin = attention_mask_39_begin_0, end = concat_26, end_mask = attention_mask_39_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_39")]; |
| tensor<fp32, [1, 16, 1, 64]> query_35 = transpose(perm = query_35_perm_0, x = var_1755)[name = tensor<string, []>("transpose_19")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_17 = mul(x = query_35, y = var_687)[name = tensor<string, []>("mul_17")]; |
| tensor<bool, []> matmul_17_transpose_y_0 = const()[name = tensor<string, []>("matmul_17_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_17_transpose_x_0 = const()[name = tensor<string, []>("matmul_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_17 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = mul_17, y = key_35)[name = tensor<string, []>("matmul_17")]; |
| tensor<fp32, [?, 16, ?, ?]> add_17 = add(x = matmul_17, y = attention_mask_39)[name = tensor<string, []>("add_17")]; |
| tensor<int32, []> softmax_17_axis_0 = const()[name = tensor<string, []>("softmax_17_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_17 = softmax(axis = softmax_17_axis_0, x = add_17)[name = tensor<string, []>("softmax_17")]; |
| tensor<bool, []> attn_output_69_transpose_x_0 = const()[name = tensor<string, []>("attn_output_69_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_69_transpose_y_0 = const()[name = tensor<string, []>("attn_output_69_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_69 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = softmax_17, y = value_35)[name = tensor<string, []>("attn_output_69")]; |
| tensor<int32, [4]> var_1763_perm_0 = const()[name = tensor<string, []>("op_1763_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1765 = const()[name = tensor<string, []>("op_1765"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1763 = transpose(perm = var_1763_perm_0, x = attn_output_69)[name = tensor<string, []>("transpose_18")]; |
| tensor<fp32, [1, 1, ?]> var_1766 = reshape(shape = var_1765, x = var_1763)[name = tensor<string, []>("op_1766")]; |
| tensor<fp32, [1, 1, 1024]> input_207 = linear(bias = decoder_layers_8_encoder_attn_out_proj_bias, weight = decoder_layers_8_encoder_attn_out_proj_weight, x = var_1766)[name = tensor<string, []>("linear_69")]; |
| tensor<fp32, [1, 1, 1024]> input_209 = add(x = input_203, y = input_207)[name = tensor<string, []>("input_209")]; |
| tensor<int32, [1]> input_211_axes_0 = const()[name = tensor<string, []>("input_211_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_211 = layer_norm(axes = input_211_axes_0, beta = decoder_layers_8_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_8_final_layer_norm_weight, x = input_209)[name = tensor<string, []>("input_211")]; |
| tensor<fp32, [1, 1, 4096]> input_213 = linear(bias = decoder_layers_8_fc1_bias, weight = decoder_layers_8_fc1_weight, x = input_211)[name = tensor<string, []>("linear_70")]; |
| tensor<fp32, [1, 1, 4096]> input_215 = relu(x = input_213)[name = tensor<string, []>("input_215")]; |
| tensor<fp32, [1, 1, 1024]> input_219 = linear(bias = decoder_layers_8_fc2_bias, weight = decoder_layers_8_fc2_weight, x = input_215)[name = tensor<string, []>("linear_71")]; |
| tensor<fp32, [1, 1, 1024]> input_221 = add(x = input_209, y = input_219)[name = tensor<string, []>("input_221")]; |
| tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = tensor<string, []>("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_91 = layer_norm(axes = hidden_states_91_axes_0, beta = decoder_layers_9_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_9_self_attn_layer_norm_weight, x = input_221)[name = tensor<string, []>("hidden_states_91")]; |
| tensor<fp32, [1, 1, 1024]> var_1810 = linear(bias = decoder_layers_9_self_attn_q_proj_bias, weight = decoder_layers_9_self_attn_q_proj_weight, x = hidden_states_91)[name = tensor<string, []>("linear_72")]; |
| tensor<int32, [4]> var_1811 = const()[name = tensor<string, []>("op_1811"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1812 = reshape(shape = var_1811, x = var_1810)[name = tensor<string, []>("op_1812")]; |
| tensor<int32, [4]> query_37_perm_0 = const()[name = tensor<string, []>("query_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_85 = linear(bias = decoder_layers_9_self_attn_k_proj_bias, weight = decoder_layers_9_self_attn_k_proj_weight, x = hidden_states_91)[name = tensor<string, []>("linear_73")]; |
| tensor<fp32, [1, 1, 1024]> value_states_85 = linear(bias = decoder_layers_9_self_attn_v_proj_bias, weight = decoder_layers_9_self_attn_v_proj_weight, x = hidden_states_91)[name = tensor<string, []>("linear_74")]; |
| tensor<int32, [4]> var_1820 = const()[name = tensor<string, []>("op_1820"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1821 = reshape(shape = var_1820, x = key_states_85)[name = tensor<string, []>("op_1821")]; |
| tensor<int32, [4]> key_states_87_perm_0 = const()[name = tensor<string, []>("key_states_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1823 = const()[name = tensor<string, []>("op_1823"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1824 = reshape(shape = var_1823, x = value_states_85)[name = tensor<string, []>("op_1824")]; |
| tensor<int32, [4]> value_states_87_perm_0 = const()[name = tensor<string, []>("value_states_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_37_interleave_0 = const()[name = tensor<string, []>("key_37_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_87 = transpose(perm = key_states_87_perm_0, x = var_1821)[name = tensor<string, []>("transpose_16")]; |
| tensor<fp32, [1, 16, ?, 64]> key_37 = concat(axis = var_689, interleave = key_37_interleave_0, values = (var_570, key_states_87))[name = tensor<string, []>("key_37")]; |
| tensor<bool, []> value_37_interleave_0 = const()[name = tensor<string, []>("value_37_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_87 = transpose(perm = value_states_87_perm_0, x = var_1824)[name = tensor<string, []>("transpose_15")]; |
| tensor<fp32, [1, 16, ?, 64]> value_37 = concat(axis = var_689, interleave = value_37_interleave_0, values = (var_573, value_states_87))[name = tensor<string, []>("value_37")]; |
| tensor<int32, [4]> var_1830_shape = shape(x = key_37)[name = tensor<string, []>("op_1830_shape")]; |
| tensor<int32, []> gather_23_indices_0 = const()[name = tensor<string, []>("gather_23_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_23_axis_0 = const()[name = tensor<string, []>("gather_23_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_23_batch_dims_0 = const()[name = tensor<string, []>("gather_23_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_23 = gather(axis = gather_23_axis_0, batch_dims = gather_23_batch_dims_0, indices = gather_23_indices_0, x = var_1830_shape)[name = tensor<string, []>("gather_23")]; |
| tensor<int32, []> concat_27_values0_0 = const()[name = tensor<string, []>("concat_27_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_27_values1_0 = const()[name = tensor<string, []>("concat_27_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_27_values2_0 = const()[name = tensor<string, []>("concat_27_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_27_axis_0 = const()[name = tensor<string, []>("concat_27_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_27_interleave_0 = const()[name = tensor<string, []>("concat_27_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_27 = concat(axis = concat_27_axis_0, interleave = concat_27_interleave_0, values = (concat_27_values0_0, concat_27_values1_0, concat_27_values2_0, gather_23))[name = tensor<string, []>("concat_27")]; |
| tensor<int32, [4]> attention_mask_41_begin_0 = const()[name = tensor<string, []>("attention_mask_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_41_end_mask_0 = const()[name = tensor<string, []>("attention_mask_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_41 = slice_by_index(begin = attention_mask_41_begin_0, end = concat_27, end_mask = attention_mask_41_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_41")]; |
| tensor<fp32, [1, 16, 1, 64]> query_37 = transpose(perm = query_37_perm_0, x = var_1812)[name = tensor<string, []>("transpose_17")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_18 = mul(x = query_37, y = var_687)[name = tensor<string, []>("mul_18")]; |
| tensor<bool, []> matmul_18_transpose_y_0 = const()[name = tensor<string, []>("matmul_18_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_18_transpose_x_0 = const()[name = tensor<string, []>("matmul_18_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_18 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = mul_18, y = key_37)[name = tensor<string, []>("matmul_18")]; |
| tensor<fp32, [?, 16, ?, ?]> add_18 = add(x = matmul_18, y = attention_mask_41)[name = tensor<string, []>("add_18")]; |
| tensor<int32, []> softmax_18_axis_0 = const()[name = tensor<string, []>("softmax_18_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_18 = softmax(axis = softmax_18_axis_0, x = add_18)[name = tensor<string, []>("softmax_18")]; |
| tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_73 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = softmax_18, y = value_37)[name = tensor<string, []>("attn_output_73")]; |
| tensor<int32, [4]> var_1836_perm_0 = const()[name = tensor<string, []>("op_1836_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1838 = const()[name = tensor<string, []>("op_1838"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1836 = transpose(perm = var_1836_perm_0, x = attn_output_73)[name = tensor<string, []>("transpose_14")]; |
| tensor<fp32, [1, 1, ?]> var_1839 = reshape(shape = var_1838, x = var_1836)[name = tensor<string, []>("op_1839")]; |
| tensor<fp32, [1, 1, 1024]> input_225 = linear(bias = decoder_layers_9_self_attn_out_proj_bias, weight = decoder_layers_9_self_attn_out_proj_weight, x = var_1839)[name = tensor<string, []>("linear_75")]; |
| tensor<fp32, [1, 1, 1024]> input_227 = add(x = input_221, y = input_225)[name = tensor<string, []>("input_227")]; |
| tensor<int32, [1]> hidden_states_95_axes_0 = const()[name = tensor<string, []>("hidden_states_95_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_95 = layer_norm(axes = hidden_states_95_axes_0, beta = decoder_layers_9_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_9_encoder_attn_layer_norm_weight, x = input_227)[name = tensor<string, []>("hidden_states_95")]; |
| tensor<fp32, [1, 1, 1024]> var_1860 = linear(bias = decoder_layers_9_encoder_attn_q_proj_bias, weight = decoder_layers_9_encoder_attn_q_proj_weight, x = hidden_states_95)[name = tensor<string, []>("linear_76")]; |
| tensor<int32, [4]> var_1861 = const()[name = tensor<string, []>("op_1861"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1862 = reshape(shape = var_1861, x = var_1860)[name = tensor<string, []>("op_1862")]; |
| tensor<int32, [4]> query_39_perm_0 = const()[name = tensor<string, []>("query_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1864_shape = shape(x = key_39)[name = tensor<string, []>("op_1864_shape")]; |
| tensor<int32, []> gather_24_indices_0 = const()[name = tensor<string, []>("gather_24_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_24_axis_0 = const()[name = tensor<string, []>("gather_24_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_24_batch_dims_0 = const()[name = tensor<string, []>("gather_24_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_24 = gather(axis = gather_24_axis_0, batch_dims = gather_24_batch_dims_0, indices = gather_24_indices_0, x = var_1864_shape)[name = tensor<string, []>("gather_24")]; |
| tensor<int32, []> concat_28_values0_0 = const()[name = tensor<string, []>("concat_28_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_28_values1_0 = const()[name = tensor<string, []>("concat_28_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_28_values2_0 = const()[name = tensor<string, []>("concat_28_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_28_axis_0 = const()[name = tensor<string, []>("concat_28_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_28_interleave_0 = const()[name = tensor<string, []>("concat_28_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_28 = concat(axis = concat_28_axis_0, interleave = concat_28_interleave_0, values = (concat_28_values0_0, concat_28_values1_0, concat_28_values2_0, gather_24))[name = tensor<string, []>("concat_28")]; |
| tensor<int32, [4]> attention_mask_43_begin_0 = const()[name = tensor<string, []>("attention_mask_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_43_end_mask_0 = const()[name = tensor<string, []>("attention_mask_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_43 = slice_by_index(begin = attention_mask_43_begin_0, end = concat_28, end_mask = attention_mask_43_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_43")]; |
| tensor<fp32, [1, 16, 1, 64]> query_39 = transpose(perm = query_39_perm_0, x = var_1862)[name = tensor<string, []>("transpose_13")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_19 = mul(x = query_39, y = var_687)[name = tensor<string, []>("mul_19")]; |
| tensor<bool, []> matmul_19_transpose_y_0 = const()[name = tensor<string, []>("matmul_19_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_19_transpose_x_0 = const()[name = tensor<string, []>("matmul_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_19 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = mul_19, y = key_39)[name = tensor<string, []>("matmul_19")]; |
| tensor<fp32, [?, 16, ?, ?]> add_19 = add(x = matmul_19, y = attention_mask_43)[name = tensor<string, []>("add_19")]; |
| tensor<int32, []> softmax_19_axis_0 = const()[name = tensor<string, []>("softmax_19_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_19 = softmax(axis = softmax_19_axis_0, x = add_19)[name = tensor<string, []>("softmax_19")]; |
| tensor<bool, []> attn_output_77_transpose_x_0 = const()[name = tensor<string, []>("attn_output_77_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_77_transpose_y_0 = const()[name = tensor<string, []>("attn_output_77_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_77 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = softmax_19, y = value_39)[name = tensor<string, []>("attn_output_77")]; |
| tensor<int32, [4]> var_1870_perm_0 = const()[name = tensor<string, []>("op_1870_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1872 = const()[name = tensor<string, []>("op_1872"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1870 = transpose(perm = var_1870_perm_0, x = attn_output_77)[name = tensor<string, []>("transpose_12")]; |
| tensor<fp32, [1, 1, ?]> var_1873 = reshape(shape = var_1872, x = var_1870)[name = tensor<string, []>("op_1873")]; |
| tensor<fp32, [1, 1, 1024]> input_231 = linear(bias = decoder_layers_9_encoder_attn_out_proj_bias, weight = decoder_layers_9_encoder_attn_out_proj_weight, x = var_1873)[name = tensor<string, []>("linear_77")]; |
| tensor<fp32, [1, 1, 1024]> input_233 = add(x = input_227, y = input_231)[name = tensor<string, []>("input_233")]; |
| tensor<int32, [1]> input_235_axes_0 = const()[name = tensor<string, []>("input_235_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_235 = layer_norm(axes = input_235_axes_0, beta = decoder_layers_9_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_9_final_layer_norm_weight, x = input_233)[name = tensor<string, []>("input_235")]; |
| tensor<fp32, [1, 1, 4096]> input_237 = linear(bias = decoder_layers_9_fc1_bias, weight = decoder_layers_9_fc1_weight, x = input_235)[name = tensor<string, []>("linear_78")]; |
| tensor<fp32, [1, 1, 4096]> input_239 = relu(x = input_237)[name = tensor<string, []>("input_239")]; |
| tensor<fp32, [1, 1, 1024]> input_243 = linear(bias = decoder_layers_9_fc2_bias, weight = decoder_layers_9_fc2_weight, x = input_239)[name = tensor<string, []>("linear_79")]; |
| tensor<fp32, [1, 1, 1024]> input_245 = add(x = input_233, y = input_243)[name = tensor<string, []>("input_245")]; |
| tensor<int32, [1]> hidden_states_101_axes_0 = const()[name = tensor<string, []>("hidden_states_101_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_101 = layer_norm(axes = hidden_states_101_axes_0, beta = decoder_layers_10_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_10_self_attn_layer_norm_weight, x = input_245)[name = tensor<string, []>("hidden_states_101")]; |
| tensor<fp32, [1, 1, 1024]> var_1917 = linear(bias = decoder_layers_10_self_attn_q_proj_bias, weight = decoder_layers_10_self_attn_q_proj_weight, x = hidden_states_101)[name = tensor<string, []>("linear_80")]; |
| tensor<int32, [4]> var_1918 = const()[name = tensor<string, []>("op_1918"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1919 = reshape(shape = var_1918, x = var_1917)[name = tensor<string, []>("op_1919")]; |
| tensor<int32, [4]> query_41_perm_0 = const()[name = tensor<string, []>("query_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_89 = linear(bias = decoder_layers_10_self_attn_k_proj_bias, weight = decoder_layers_10_self_attn_k_proj_weight, x = hidden_states_101)[name = tensor<string, []>("linear_81")]; |
| tensor<fp32, [1, 1, 1024]> value_states_89 = linear(bias = decoder_layers_10_self_attn_v_proj_bias, weight = decoder_layers_10_self_attn_v_proj_weight, x = hidden_states_101)[name = tensor<string, []>("linear_82")]; |
| tensor<int32, [4]> var_1927 = const()[name = tensor<string, []>("op_1927"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1928 = reshape(shape = var_1927, x = key_states_89)[name = tensor<string, []>("op_1928")]; |
| tensor<int32, [4]> key_states_91_perm_0 = const()[name = tensor<string, []>("key_states_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1930 = const()[name = tensor<string, []>("op_1930"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1931 = reshape(shape = var_1930, x = value_states_89)[name = tensor<string, []>("op_1931")]; |
| tensor<int32, [4]> value_states_91_perm_0 = const()[name = tensor<string, []>("value_states_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_41_interleave_0 = const()[name = tensor<string, []>("key_41_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states_91 = transpose(perm = key_states_91_perm_0, x = var_1928)[name = tensor<string, []>("transpose_10")]; |
| tensor<fp32, [1, 16, ?, 64]> key_41 = concat(axis = var_689, interleave = key_41_interleave_0, values = (var_614, key_states_91))[name = tensor<string, []>("key_41")]; |
| tensor<bool, []> value_41_interleave_0 = const()[name = tensor<string, []>("value_41_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states_91 = transpose(perm = value_states_91_perm_0, x = var_1931)[name = tensor<string, []>("transpose_9")]; |
| tensor<fp32, [1, 16, ?, 64]> value_41 = concat(axis = var_689, interleave = value_41_interleave_0, values = (var_617, value_states_91))[name = tensor<string, []>("value_41")]; |
| tensor<int32, [4]> var_1937_shape = shape(x = key_41)[name = tensor<string, []>("op_1937_shape")]; |
| tensor<int32, []> gather_25_indices_0 = const()[name = tensor<string, []>("gather_25_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_25_axis_0 = const()[name = tensor<string, []>("gather_25_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_25_batch_dims_0 = const()[name = tensor<string, []>("gather_25_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_25 = gather(axis = gather_25_axis_0, batch_dims = gather_25_batch_dims_0, indices = gather_25_indices_0, x = var_1937_shape)[name = tensor<string, []>("gather_25")]; |
| tensor<int32, []> concat_29_values0_0 = const()[name = tensor<string, []>("concat_29_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_29_values1_0 = const()[name = tensor<string, []>("concat_29_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_29_values2_0 = const()[name = tensor<string, []>("concat_29_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_29_axis_0 = const()[name = tensor<string, []>("concat_29_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_29_interleave_0 = const()[name = tensor<string, []>("concat_29_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_29 = concat(axis = concat_29_axis_0, interleave = concat_29_interleave_0, values = (concat_29_values0_0, concat_29_values1_0, concat_29_values2_0, gather_25))[name = tensor<string, []>("concat_29")]; |
| tensor<int32, [4]> attention_mask_45_begin_0 = const()[name = tensor<string, []>("attention_mask_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_45_end_mask_0 = const()[name = tensor<string, []>("attention_mask_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_45 = slice_by_index(begin = attention_mask_45_begin_0, end = concat_29, end_mask = attention_mask_45_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_45")]; |
| tensor<fp32, [1, 16, 1, 64]> query_41 = transpose(perm = query_41_perm_0, x = var_1919)[name = tensor<string, []>("transpose_11")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_20 = mul(x = query_41, y = var_687)[name = tensor<string, []>("mul_20")]; |
| tensor<bool, []> matmul_20_transpose_y_0 = const()[name = tensor<string, []>("matmul_20_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_20_transpose_x_0 = const()[name = tensor<string, []>("matmul_20_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_20 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = mul_20, y = key_41)[name = tensor<string, []>("matmul_20")]; |
| tensor<fp32, [?, 16, ?, ?]> add_20 = add(x = matmul_20, y = attention_mask_45)[name = tensor<string, []>("add_20")]; |
| tensor<int32, []> softmax_20_axis_0 = const()[name = tensor<string, []>("softmax_20_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_20 = softmax(axis = softmax_20_axis_0, x = add_20)[name = tensor<string, []>("softmax_20")]; |
| tensor<bool, []> attn_output_81_transpose_x_0 = const()[name = tensor<string, []>("attn_output_81_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_81_transpose_y_0 = const()[name = tensor<string, []>("attn_output_81_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_81 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = softmax_20, y = value_41)[name = tensor<string, []>("attn_output_81")]; |
| tensor<int32, [4]> var_1943_perm_0 = const()[name = tensor<string, []>("op_1943_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1945 = const()[name = tensor<string, []>("op_1945"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1943 = transpose(perm = var_1943_perm_0, x = attn_output_81)[name = tensor<string, []>("transpose_8")]; |
| tensor<fp32, [1, 1, ?]> var_1946 = reshape(shape = var_1945, x = var_1943)[name = tensor<string, []>("op_1946")]; |
| tensor<fp32, [1, 1, 1024]> input_249 = linear(bias = decoder_layers_10_self_attn_out_proj_bias, weight = decoder_layers_10_self_attn_out_proj_weight, x = var_1946)[name = tensor<string, []>("linear_83")]; |
| tensor<fp32, [1, 1, 1024]> input_251 = add(x = input_245, y = input_249)[name = tensor<string, []>("input_251")]; |
| tensor<int32, [1]> hidden_states_105_axes_0 = const()[name = tensor<string, []>("hidden_states_105_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_105 = layer_norm(axes = hidden_states_105_axes_0, beta = decoder_layers_10_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_10_encoder_attn_layer_norm_weight, x = input_251)[name = tensor<string, []>("hidden_states_105")]; |
| tensor<fp32, [1, 1, 1024]> var_1967 = linear(bias = decoder_layers_10_encoder_attn_q_proj_bias, weight = decoder_layers_10_encoder_attn_q_proj_weight, x = hidden_states_105)[name = tensor<string, []>("linear_84")]; |
| tensor<int32, [4]> var_1968 = const()[name = tensor<string, []>("op_1968"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_1969 = reshape(shape = var_1968, x = var_1967)[name = tensor<string, []>("op_1969")]; |
| tensor<int32, [4]> query_43_perm_0 = const()[name = tensor<string, []>("query_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_1971_shape = shape(x = key_43)[name = tensor<string, []>("op_1971_shape")]; |
| tensor<int32, []> gather_26_indices_0 = const()[name = tensor<string, []>("gather_26_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_26_axis_0 = const()[name = tensor<string, []>("gather_26_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_26_batch_dims_0 = const()[name = tensor<string, []>("gather_26_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_26 = gather(axis = gather_26_axis_0, batch_dims = gather_26_batch_dims_0, indices = gather_26_indices_0, x = var_1971_shape)[name = tensor<string, []>("gather_26")]; |
| tensor<int32, []> concat_30_values0_0 = const()[name = tensor<string, []>("concat_30_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_30_values1_0 = const()[name = tensor<string, []>("concat_30_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_30_values2_0 = const()[name = tensor<string, []>("concat_30_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_30_axis_0 = const()[name = tensor<string, []>("concat_30_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_30_interleave_0 = const()[name = tensor<string, []>("concat_30_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_30 = concat(axis = concat_30_axis_0, interleave = concat_30_interleave_0, values = (concat_30_values0_0, concat_30_values1_0, concat_30_values2_0, gather_26))[name = tensor<string, []>("concat_30")]; |
| tensor<int32, [4]> attention_mask_47_begin_0 = const()[name = tensor<string, []>("attention_mask_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_47_end_mask_0 = const()[name = tensor<string, []>("attention_mask_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_47 = slice_by_index(begin = attention_mask_47_begin_0, end = concat_30, end_mask = attention_mask_47_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_47")]; |
| tensor<fp32, [1, 16, 1, 64]> query_43 = transpose(perm = query_43_perm_0, x = var_1969)[name = tensor<string, []>("transpose_7")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_21 = mul(x = query_43, y = var_687)[name = tensor<string, []>("mul_21")]; |
| tensor<bool, []> matmul_21_transpose_y_0 = const()[name = tensor<string, []>("matmul_21_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_21_transpose_x_0 = const()[name = tensor<string, []>("matmul_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_21 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = mul_21, y = key_43)[name = tensor<string, []>("matmul_21")]; |
| tensor<fp32, [?, 16, ?, ?]> add_21 = add(x = matmul_21, y = attention_mask_47)[name = tensor<string, []>("add_21")]; |
| tensor<int32, []> softmax_21_axis_0 = const()[name = tensor<string, []>("softmax_21_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_21 = softmax(axis = softmax_21_axis_0, x = add_21)[name = tensor<string, []>("softmax_21")]; |
| tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_85 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = softmax_21, y = value_43)[name = tensor<string, []>("attn_output_85")]; |
| tensor<int32, [4]> var_1977_perm_0 = const()[name = tensor<string, []>("op_1977_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1979 = const()[name = tensor<string, []>("op_1979"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_1977 = transpose(perm = var_1977_perm_0, x = attn_output_85)[name = tensor<string, []>("transpose_6")]; |
| tensor<fp32, [1, 1, ?]> var_1980 = reshape(shape = var_1979, x = var_1977)[name = tensor<string, []>("op_1980")]; |
| tensor<fp32, [1, 1, 1024]> input_255 = linear(bias = decoder_layers_10_encoder_attn_out_proj_bias, weight = decoder_layers_10_encoder_attn_out_proj_weight, x = var_1980)[name = tensor<string, []>("linear_85")]; |
| tensor<fp32, [1, 1, 1024]> input_257 = add(x = input_251, y = input_255)[name = tensor<string, []>("input_257")]; |
| tensor<int32, [1]> input_259_axes_0 = const()[name = tensor<string, []>("input_259_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_259 = layer_norm(axes = input_259_axes_0, beta = decoder_layers_10_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_10_final_layer_norm_weight, x = input_257)[name = tensor<string, []>("input_259")]; |
| tensor<fp32, [1, 1, 4096]> input_261 = linear(bias = decoder_layers_10_fc1_bias, weight = decoder_layers_10_fc1_weight, x = input_259)[name = tensor<string, []>("linear_86")]; |
| tensor<fp32, [1, 1, 4096]> input_263 = relu(x = input_261)[name = tensor<string, []>("input_263")]; |
| tensor<fp32, [1, 1, 1024]> input_267 = linear(bias = decoder_layers_10_fc2_bias, weight = decoder_layers_10_fc2_weight, x = input_263)[name = tensor<string, []>("linear_87")]; |
| tensor<fp32, [1, 1, 1024]> input_269 = add(x = input_257, y = input_267)[name = tensor<string, []>("input_269")]; |
| tensor<int32, [1]> hidden_states_111_axes_0 = const()[name = tensor<string, []>("hidden_states_111_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_111 = layer_norm(axes = hidden_states_111_axes_0, beta = decoder_layers_11_self_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_11_self_attn_layer_norm_weight, x = input_269)[name = tensor<string, []>("hidden_states_111")]; |
| tensor<fp32, [1, 1, 1024]> var_2024 = linear(bias = decoder_layers_11_self_attn_q_proj_bias, weight = decoder_layers_11_self_attn_q_proj_weight, x = hidden_states_111)[name = tensor<string, []>("linear_88")]; |
| tensor<int32, [4]> var_2025 = const()[name = tensor<string, []>("op_2025"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_2026 = reshape(shape = var_2025, x = var_2024)[name = tensor<string, []>("op_2026")]; |
| tensor<int32, [4]> query_45_perm_0 = const()[name = tensor<string, []>("query_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 1, 1024]> key_states_93 = linear(bias = decoder_layers_11_self_attn_k_proj_bias, weight = decoder_layers_11_self_attn_k_proj_weight, x = hidden_states_111)[name = tensor<string, []>("linear_89")]; |
| tensor<fp32, [1, 1, 1024]> value_states_93 = linear(bias = decoder_layers_11_self_attn_v_proj_bias, weight = decoder_layers_11_self_attn_v_proj_weight, x = hidden_states_111)[name = tensor<string, []>("linear_90")]; |
| tensor<int32, [4]> var_2034 = const()[name = tensor<string, []>("op_2034"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_2035 = reshape(shape = var_2034, x = key_states_93)[name = tensor<string, []>("op_2035")]; |
| tensor<int32, [4]> key_states_perm_0 = const()[name = tensor<string, []>("key_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2037 = const()[name = tensor<string, []>("op_2037"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_2038 = reshape(shape = var_2037, x = value_states_93)[name = tensor<string, []>("op_2038")]; |
| tensor<int32, [4]> value_states_perm_0 = const()[name = tensor<string, []>("value_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_45_interleave_0 = const()[name = tensor<string, []>("key_45_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> key_states = transpose(perm = key_states_perm_0, x = var_2035)[name = tensor<string, []>("transpose_4")]; |
| tensor<fp32, [1, 16, ?, 64]> key_45 = concat(axis = var_689, interleave = key_45_interleave_0, values = (var_658, key_states))[name = tensor<string, []>("key_45")]; |
| tensor<bool, []> value_45_interleave_0 = const()[name = tensor<string, []>("value_45_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, 64]> value_states = transpose(perm = value_states_perm_0, x = var_2038)[name = tensor<string, []>("transpose_3")]; |
| tensor<fp32, [1, 16, ?, 64]> value_45 = concat(axis = var_689, interleave = value_45_interleave_0, values = (var_661, value_states))[name = tensor<string, []>("value_45")]; |
| tensor<int32, [4]> var_2044_shape = shape(x = key_45)[name = tensor<string, []>("op_2044_shape")]; |
| tensor<int32, []> gather_27_indices_0 = const()[name = tensor<string, []>("gather_27_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_27_axis_0 = const()[name = tensor<string, []>("gather_27_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_27_batch_dims_0 = const()[name = tensor<string, []>("gather_27_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_27 = gather(axis = gather_27_axis_0, batch_dims = gather_27_batch_dims_0, indices = gather_27_indices_0, x = var_2044_shape)[name = tensor<string, []>("gather_27")]; |
| tensor<int32, []> concat_31_values0_0 = const()[name = tensor<string, []>("concat_31_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_31_values1_0 = const()[name = tensor<string, []>("concat_31_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_31_values2_0 = const()[name = tensor<string, []>("concat_31_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_31_axis_0 = const()[name = tensor<string, []>("concat_31_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_31_interleave_0 = const()[name = tensor<string, []>("concat_31_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_31 = concat(axis = concat_31_axis_0, interleave = concat_31_interleave_0, values = (concat_31_values0_0, concat_31_values1_0, concat_31_values2_0, gather_27))[name = tensor<string, []>("concat_31")]; |
| tensor<int32, [4]> attention_mask_49_begin_0 = const()[name = tensor<string, []>("attention_mask_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_49_end_mask_0 = const()[name = tensor<string, []>("attention_mask_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_49 = slice_by_index(begin = attention_mask_49_begin_0, end = concat_31, end_mask = attention_mask_49_end_mask_0, x = reshape_4)[name = tensor<string, []>("attention_mask_49")]; |
| tensor<fp32, [1, 16, 1, 64]> query_45 = transpose(perm = query_45_perm_0, x = var_2026)[name = tensor<string, []>("transpose_5")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_22 = mul(x = query_45, y = var_687)[name = tensor<string, []>("mul_22")]; |
| tensor<bool, []> matmul_22_transpose_y_0 = const()[name = tensor<string, []>("matmul_22_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_22_transpose_x_0 = const()[name = tensor<string, []>("matmul_22_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_22 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = mul_22, y = key_45)[name = tensor<string, []>("matmul_22")]; |
| tensor<fp32, [?, 16, ?, ?]> add_22 = add(x = matmul_22, y = attention_mask_49)[name = tensor<string, []>("add_22")]; |
| tensor<int32, []> softmax_22_axis_0 = const()[name = tensor<string, []>("softmax_22_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_22 = softmax(axis = softmax_22_axis_0, x = add_22)[name = tensor<string, []>("softmax_22")]; |
| tensor<bool, []> attn_output_89_transpose_x_0 = const()[name = tensor<string, []>("attn_output_89_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_89_transpose_y_0 = const()[name = tensor<string, []>("attn_output_89_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_89 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = softmax_22, y = value_45)[name = tensor<string, []>("attn_output_89")]; |
| tensor<int32, [4]> var_2050_perm_0 = const()[name = tensor<string, []>("op_2050_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2052 = const()[name = tensor<string, []>("op_2052"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_2050 = transpose(perm = var_2050_perm_0, x = attn_output_89)[name = tensor<string, []>("transpose_2")]; |
| tensor<fp32, [1, 1, ?]> var_2053 = reshape(shape = var_2052, x = var_2050)[name = tensor<string, []>("op_2053")]; |
| tensor<fp32, [1, 1, 1024]> input_273 = linear(bias = decoder_layers_11_self_attn_out_proj_bias, weight = decoder_layers_11_self_attn_out_proj_weight, x = var_2053)[name = tensor<string, []>("linear_91")]; |
| tensor<fp32, [1, 1, 1024]> input_275 = add(x = input_269, y = input_273)[name = tensor<string, []>("input_275")]; |
| tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = tensor<string, []>("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> hidden_states_115 = layer_norm(axes = hidden_states_115_axes_0, beta = decoder_layers_11_encoder_attn_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_11_encoder_attn_layer_norm_weight, x = input_275)[name = tensor<string, []>("hidden_states_115")]; |
| tensor<fp32, [1, 1, 1024]> var_2074 = linear(bias = decoder_layers_11_encoder_attn_q_proj_bias, weight = decoder_layers_11_encoder_attn_q_proj_weight, x = hidden_states_115)[name = tensor<string, []>("linear_92")]; |
| tensor<int32, [4]> var_2075 = const()[name = tensor<string, []>("op_2075"), val = tensor<int32, [4]>([1, 1, -1, 64])]; |
| tensor<fp32, [1, 1, 16, 64]> var_2076 = reshape(shape = var_2075, x = var_2074)[name = tensor<string, []>("op_2076")]; |
| tensor<int32, [4]> query_perm_0 = const()[name = tensor<string, []>("query_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> var_2078_shape = shape(x = key)[name = tensor<string, []>("op_2078_shape")]; |
| tensor<int32, []> gather_28_indices_0 = const()[name = tensor<string, []>("gather_28_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_28_axis_0 = const()[name = tensor<string, []>("gather_28_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_28_batch_dims_0 = const()[name = tensor<string, []>("gather_28_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_28 = gather(axis = gather_28_axis_0, batch_dims = gather_28_batch_dims_0, indices = gather_28_indices_0, x = var_2078_shape)[name = tensor<string, []>("gather_28")]; |
| tensor<int32, []> concat_32_values0_0 = const()[name = tensor<string, []>("concat_32_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_32_values1_0 = const()[name = tensor<string, []>("concat_32_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_32_values2_0 = const()[name = tensor<string, []>("concat_32_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_32_axis_0 = const()[name = tensor<string, []>("concat_32_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_32_interleave_0 = const()[name = tensor<string, []>("concat_32_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_32 = concat(axis = concat_32_axis_0, interleave = concat_32_interleave_0, values = (concat_32_values0_0, concat_32_values1_0, concat_32_values2_0, gather_28))[name = tensor<string, []>("concat_32")]; |
| tensor<int32, [4]> attention_mask_begin_0 = const()[name = tensor<string, []>("attention_mask_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<bool, [4]> attention_mask_end_mask_0 = const()[name = tensor<string, []>("attention_mask_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask = slice_by_index(begin = attention_mask_begin_0, end = concat_32, end_mask = attention_mask_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask")]; |
| tensor<fp32, [1, 16, 1, 64]> query = transpose(perm = query_perm_0, x = var_2076)[name = tensor<string, []>("transpose_1")]; |
| tensor<fp32, [1, 16, 1, 64]> mul_23 = mul(x = query, y = var_687)[name = tensor<string, []>("mul_23")]; |
| tensor<bool, []> matmul_23_transpose_y_0 = const()[name = tensor<string, []>("matmul_23_transpose_y_0"), val = tensor<bool, []>(true)]; |
| tensor<bool, []> matmul_23_transpose_x_0 = const()[name = tensor<string, []>("matmul_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, 1, ?]> matmul_23 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = mul_23, y = key)[name = tensor<string, []>("matmul_23")]; |
| tensor<fp32, [?, 16, ?, ?]> add_23 = add(x = matmul_23, y = attention_mask)[name = tensor<string, []>("add_23")]; |
| tensor<int32, []> softmax_23_axis_0 = const()[name = tensor<string, []>("softmax_23_axis_0"), val = tensor<int32, []>(-1)]; |
| tensor<fp32, [?, 16, ?, ?]> softmax_23 = softmax(axis = softmax_23_axis_0, x = add_23)[name = tensor<string, []>("softmax_23")]; |
| tensor<bool, []> attn_output_93_transpose_x_0 = const()[name = tensor<string, []>("attn_output_93_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> attn_output_93_transpose_y_0 = const()[name = tensor<string, []>("attn_output_93_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [?, 16, ?, 64]> attn_output_93 = matmul(transpose_x = attn_output_93_transpose_x_0, transpose_y = attn_output_93_transpose_y_0, x = softmax_23, y = value)[name = tensor<string, []>("attn_output_93")]; |
| tensor<int32, [4]> var_2084_perm_0 = const()[name = tensor<string, []>("op_2084_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_2086 = const()[name = tensor<string, []>("op_2086"), val = tensor<int32, [3]>([1, 1, -1])]; |
| tensor<fp32, [?, ?, 16, 64]> var_2084 = transpose(perm = var_2084_perm_0, x = attn_output_93)[name = tensor<string, []>("transpose_0")]; |
| tensor<fp32, [1, 1, ?]> var_2087 = reshape(shape = var_2086, x = var_2084)[name = tensor<string, []>("op_2087")]; |
| tensor<fp32, [1, 1, 1024]> input_279 = linear(bias = decoder_layers_11_encoder_attn_out_proj_bias, weight = decoder_layers_11_encoder_attn_out_proj_weight, x = var_2087)[name = tensor<string, []>("linear_93")]; |
| tensor<fp32, [1, 1, 1024]> input_281 = add(x = input_275, y = input_279)[name = tensor<string, []>("input_281")]; |
| tensor<int32, [1]> input_283_axes_0 = const()[name = tensor<string, []>("input_283_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input_283 = layer_norm(axes = input_283_axes_0, beta = decoder_layers_11_final_layer_norm_bias, epsilon = var_685, gamma = decoder_layers_11_final_layer_norm_weight, x = input_281)[name = tensor<string, []>("input_283")]; |
| tensor<fp32, [1, 1, 4096]> input_285 = linear(bias = decoder_layers_11_fc1_bias, weight = decoder_layers_11_fc1_weight, x = input_283)[name = tensor<string, []>("linear_94")]; |
| tensor<fp32, [1, 1, 4096]> input_287 = relu(x = input_285)[name = tensor<string, []>("input_287")]; |
| tensor<fp32, [1, 1, 1024]> input_291 = linear(bias = decoder_layers_11_fc2_bias, weight = decoder_layers_11_fc2_weight, x = input_287)[name = tensor<string, []>("linear_95")]; |
| tensor<fp32, [1, 1, 1024]> input_293 = add(x = input_281, y = input_291)[name = tensor<string, []>("input_293")]; |
| tensor<int32, [1]> input_axes_0 = const()[name = tensor<string, []>("input_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 1, 1024]> input = layer_norm(axes = input_axes_0, beta = decoder_layer_norm_bias, epsilon = var_685, gamma = decoder_layer_norm_weight, x = input_293)[name = tensor<string, []>("input")]; |
| tensor<fp32, [256206]> linear_96_bias_0 = const()[name = tensor<string, []>("linear_96_bias_0"), val = tensor<fp32, [256206]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1759126336)))]; |
| tensor<fp32, [1, 1, 256206]> logits = linear(bias = linear_96_bias_0, weight = decoder_embed_tokens_weight, x = input)[name = tensor<string, []>("linear_96")]; |
| tensor<int32, []> var_2146_axis_0 = const()[name = tensor<string, []>("op_2146_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> new_past_self_key = stack(axis = var_2146_axis_0, values = (key_1, key_5, key_9, key_13, key_17, key_21, key_25, key_29, key_33, key_37, key_41, key_45))[name = tensor<string, []>("op_2146")]; |
| tensor<int32, []> var_2149_axis_0 = const()[name = tensor<string, []>("op_2149_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> new_past_self_value = stack(axis = var_2149_axis_0, values = (value_1, value_5, value_9, value_13, value_17, value_21, value_25, value_29, value_33, value_37, value_41, value_45))[name = tensor<string, []>("op_2149")]; |
| tensor<int32, []> var_2152_axis_0 = const()[name = tensor<string, []>("op_2152_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> new_past_cross_key = stack(axis = var_2152_axis_0, values = (key_3, key_7, key_11, key_15, key_19, key_23, key_27, key_31, key_35, key_39, key_43, key))[name = tensor<string, []>("op_2152")]; |
| tensor<int32, []> var_2155_axis_0 = const()[name = tensor<string, []>("op_2155_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> new_past_cross_value = stack(axis = var_2155_axis_0, values = (value_3, value_7, value_11, value_15, value_19, value_23, value_27, value_31, value_35, value_39, value_43, value))[name = tensor<string, []>("op_2155")]; |
| tensor<fp32, [1, ?, 1024]> encoder_hidden_states_tmp = identity(x = encoder_hidden_states)[name = tensor<string, []>("encoder_hidden_states_tmp")]; |
| } -> (logits, new_past_self_key, new_past_self_value, new_past_cross_key, new_past_cross_value); |
| } |
