| 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, 2]> input_ids) [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]}}), ("RangeDims", {{"encoder_attention_mask", [[1, 1], [1, 1024]]}, {"encoder_hidden_states", [[1, 1], [1, 1024], [1024, 1024]]}})))] { |
| 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_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_k_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_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_k_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_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078830272)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_0_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_0_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1078834432)))]; |
| 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, []>(1083028800)))]; |
| 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, []>(1083032960)))]; |
| 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, []>(1087227328)))]; |
| 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, []>(1087231488)))]; |
| 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, []>(1087235648)))]; |
| 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, []>(1087252096)))]; |
| 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, []>(1104029376)))]; |
| 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, []>(1104033536)))]; |
| 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, []>(1120810816)))]; |
| 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, []>(1120814976)))]; |
| 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, []>(1120819136)))]; |
| 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, []>(1120823296)))]; |
| 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, []>(1125017664)))]; |
| 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, []>(1125021824)))]; |
| 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, []>(1129216192)))]; |
| 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, []>(1129220352)))]; |
| 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, []>(1133414720)))]; |
| 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, []>(1133418880)))]; |
| 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, []>(1137613248)))]; |
| 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, []>(1137617408)))]; |
| 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, []>(1137621568)))]; |
| 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, []>(1137625728)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1141820096)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1141824256)))]; |
| tensor<fp32, [1024]> decoder_layers_1_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1146018624)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_1_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_1_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1146022784)))]; |
| 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, []>(1150217152)))]; |
| 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, []>(1150221312)))]; |
| 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, []>(1154415680)))]; |
| 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, []>(1154419840)))]; |
| 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, []>(1154424000)))]; |
| 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, []>(1154440448)))]; |
| 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, []>(1171217728)))]; |
| 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, []>(1171221888)))]; |
| 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, []>(1187999168)))]; |
| 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, []>(1188003328)))]; |
| 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, []>(1188007488)))]; |
| 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, []>(1188011648)))]; |
| 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, []>(1192206016)))]; |
| 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, []>(1192210176)))]; |
| 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, []>(1196404544)))]; |
| 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, []>(1196408704)))]; |
| 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, []>(1200603072)))]; |
| 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, []>(1200607232)))]; |
| 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, []>(1204801600)))]; |
| 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, []>(1204805760)))]; |
| 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, []>(1204809920)))]; |
| 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, []>(1204814080)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1209008448)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1209012608)))]; |
| tensor<fp32, [1024]> decoder_layers_2_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213206976)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_2_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_2_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1213211136)))]; |
| 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, []>(1217405504)))]; |
| 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, []>(1217409664)))]; |
| 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, []>(1221604032)))]; |
| 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, []>(1221608192)))]; |
| 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, []>(1221612352)))]; |
| 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, []>(1221628800)))]; |
| 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, []>(1238406080)))]; |
| 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, []>(1238410240)))]; |
| 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, []>(1255187520)))]; |
| 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, []>(1255191680)))]; |
| 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, []>(1255195840)))]; |
| 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, []>(1255200000)))]; |
| 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, []>(1259394368)))]; |
| 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, []>(1259398528)))]; |
| 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, []>(1263592896)))]; |
| 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, []>(1263597056)))]; |
| 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, []>(1267791424)))]; |
| 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, []>(1267795584)))]; |
| 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, []>(1271989952)))]; |
| 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, []>(1271994112)))]; |
| 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, []>(1271998272)))]; |
| 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, []>(1272002432)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1276196800)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1276200960)))]; |
| tensor<fp32, [1024]> decoder_layers_3_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1280395328)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_3_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_3_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1280399488)))]; |
| 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, []>(1284593856)))]; |
| 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, []>(1284598016)))]; |
| 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, []>(1288792384)))]; |
| 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, []>(1288796544)))]; |
| 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, []>(1288800704)))]; |
| 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, []>(1288817152)))]; |
| 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, []>(1305594432)))]; |
| 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, []>(1305598592)))]; |
| 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, []>(1322375872)))]; |
| 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, []>(1322380032)))]; |
| 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, []>(1322384192)))]; |
| 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, []>(1322388352)))]; |
| 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, []>(1326582720)))]; |
| 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, []>(1326586880)))]; |
| 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, []>(1330781248)))]; |
| 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, []>(1330785408)))]; |
| 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, []>(1334979776)))]; |
| 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, []>(1334983936)))]; |
| 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, []>(1339178304)))]; |
| 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, []>(1339182464)))]; |
| 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, []>(1339186624)))]; |
| 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, []>(1339190784)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1343385152)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1343389312)))]; |
| tensor<fp32, [1024]> decoder_layers_4_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347583680)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_4_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_4_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1347587840)))]; |
| 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, []>(1351782208)))]; |
| 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, []>(1351786368)))]; |
| 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, []>(1355980736)))]; |
| 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, []>(1355984896)))]; |
| 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, []>(1355989056)))]; |
| 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, []>(1356005504)))]; |
| 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, []>(1372782784)))]; |
| 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, []>(1372786944)))]; |
| 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, []>(1389564224)))]; |
| 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, []>(1389568384)))]; |
| 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, []>(1389572544)))]; |
| 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, []>(1389576704)))]; |
| 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, []>(1393771072)))]; |
| 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, []>(1393775232)))]; |
| 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, []>(1397969600)))]; |
| 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, []>(1397973760)))]; |
| 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, []>(1402168128)))]; |
| 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, []>(1402172288)))]; |
| 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, []>(1406366656)))]; |
| 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, []>(1406370816)))]; |
| 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, []>(1406374976)))]; |
| 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, []>(1406379136)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410573504)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1410577664)))]; |
| tensor<fp32, [1024]> decoder_layers_5_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414772032)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_5_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_5_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1414776192)))]; |
| 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, []>(1418970560)))]; |
| 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, []>(1418974720)))]; |
| 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, []>(1423169088)))]; |
| 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, []>(1423173248)))]; |
| 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, []>(1423177408)))]; |
| 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, []>(1423193856)))]; |
| 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, []>(1439971136)))]; |
| 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, []>(1439975296)))]; |
| 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, []>(1456752576)))]; |
| 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, []>(1456756736)))]; |
| 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, []>(1456760896)))]; |
| 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, []>(1456765056)))]; |
| 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, []>(1460959424)))]; |
| 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, []>(1460963584)))]; |
| 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, []>(1465157952)))]; |
| 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, []>(1465162112)))]; |
| 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, []>(1469356480)))]; |
| 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, []>(1469360640)))]; |
| 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, []>(1473555008)))]; |
| 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, []>(1473559168)))]; |
| 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, []>(1473563328)))]; |
| 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, []>(1473567488)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1477761856)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1477766016)))]; |
| tensor<fp32, [1024]> decoder_layers_6_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481960384)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_6_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_6_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1481964544)))]; |
| 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, []>(1486158912)))]; |
| 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, []>(1486163072)))]; |
| 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, []>(1490357440)))]; |
| 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, []>(1490361600)))]; |
| 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, []>(1490365760)))]; |
| 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, []>(1490382208)))]; |
| 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, []>(1507159488)))]; |
| 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, []>(1507163648)))]; |
| 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, []>(1523940928)))]; |
| 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, []>(1523945088)))]; |
| 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, []>(1523949248)))]; |
| 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, []>(1523953408)))]; |
| 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, []>(1528147776)))]; |
| 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, []>(1528151936)))]; |
| 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, []>(1532346304)))]; |
| 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, []>(1532350464)))]; |
| 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, []>(1536544832)))]; |
| 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, []>(1536548992)))]; |
| 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, []>(1540743360)))]; |
| 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, []>(1540747520)))]; |
| 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, []>(1540751680)))]; |
| 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, []>(1540755840)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1544950208)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1544954368)))]; |
| tensor<fp32, [1024]> decoder_layers_7_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549148736)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_7_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_7_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1549152896)))]; |
| 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, []>(1553347264)))]; |
| 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, []>(1553351424)))]; |
| 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, []>(1557545792)))]; |
| 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, []>(1557549952)))]; |
| 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, []>(1557554112)))]; |
| 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, []>(1557570560)))]; |
| 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, []>(1574347840)))]; |
| 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, []>(1574352000)))]; |
| 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, []>(1591129280)))]; |
| 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, []>(1591133440)))]; |
| 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, []>(1591137600)))]; |
| 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, []>(1591141760)))]; |
| 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, []>(1595336128)))]; |
| 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, []>(1595340288)))]; |
| 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, []>(1599534656)))]; |
| 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, []>(1599538816)))]; |
| 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, []>(1603733184)))]; |
| 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, []>(1603737344)))]; |
| 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, []>(1607931712)))]; |
| 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, []>(1607935872)))]; |
| 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, []>(1607940032)))]; |
| 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, []>(1607944192)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612138560)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1612142720)))]; |
| tensor<fp32, [1024]> decoder_layers_8_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616337088)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_8_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_8_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1616341248)))]; |
| 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, []>(1620535616)))]; |
| 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, []>(1620539776)))]; |
| 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, []>(1624734144)))]; |
| 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, []>(1624738304)))]; |
| 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, []>(1624742464)))]; |
| 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, []>(1624758912)))]; |
| 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, []>(1641536192)))]; |
| 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, []>(1641540352)))]; |
| 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, []>(1658317632)))]; |
| 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, []>(1658321792)))]; |
| 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, []>(1658325952)))]; |
| 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, []>(1658330112)))]; |
| 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, []>(1662524480)))]; |
| 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, []>(1662528640)))]; |
| 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, []>(1666723008)))]; |
| 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, []>(1666727168)))]; |
| 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, []>(1670921536)))]; |
| 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, []>(1670925696)))]; |
| 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, []>(1675120064)))]; |
| 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, []>(1675124224)))]; |
| 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, []>(1675128384)))]; |
| 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, []>(1675132544)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1679326912)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1679331072)))]; |
| tensor<fp32, [1024]> decoder_layers_9_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1683525440)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_9_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_9_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1683529600)))]; |
| 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, []>(1687723968)))]; |
| 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, []>(1687728128)))]; |
| 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, []>(1691922496)))]; |
| 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, []>(1691926656)))]; |
| 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, []>(1691930816)))]; |
| 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, []>(1691947264)))]; |
| 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, []>(1708724544)))]; |
| 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, []>(1708728704)))]; |
| 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, []>(1725505984)))]; |
| 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, []>(1725510144)))]; |
| 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, []>(1725514304)))]; |
| 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, []>(1725518464)))]; |
| 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, []>(1729712832)))]; |
| 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, []>(1729716992)))]; |
| 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, []>(1733911360)))]; |
| 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, []>(1733915520)))]; |
| 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, []>(1738109888)))]; |
| 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, []>(1738114048)))]; |
| 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, []>(1742308416)))]; |
| 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, []>(1742312576)))]; |
| 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, []>(1742316736)))]; |
| 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, []>(1742320896)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1746515264)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1746519424)))]; |
| tensor<fp32, [1024]> decoder_layers_10_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1750713792)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_10_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_10_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1750717952)))]; |
| 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, []>(1754912320)))]; |
| 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, []>(1754916480)))]; |
| 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, []>(1759110848)))]; |
| 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, []>(1759115008)))]; |
| 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, []>(1759119168)))]; |
| 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, []>(1759135616)))]; |
| 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, []>(1775912896)))]; |
| 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, []>(1775917056)))]; |
| 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, []>(1792694336)))]; |
| 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, []>(1792698496)))]; |
| 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, []>(1792702656)))]; |
| 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, []>(1792706816)))]; |
| 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, []>(1796901184)))]; |
| 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, []>(1796905344)))]; |
| 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, []>(1801099712)))]; |
| 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, []>(1801103872)))]; |
| 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, []>(1805298240)))]; |
| 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, []>(1805302400)))]; |
| 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, []>(1809496768)))]; |
| 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, []>(1809500928)))]; |
| 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, []>(1809505088)))]; |
| 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, []>(1809509248)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_k_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_k_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1813703616)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_encoder_attn_k_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_k_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1813707776)))]; |
| tensor<fp32, [1024]> decoder_layers_11_encoder_attn_v_proj_bias = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_v_proj_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1817902144)))]; |
| tensor<fp32, [1024, 1024]> decoder_layers_11_encoder_attn_v_proj_weight = const()[name = tensor<string, []>("decoder_layers_11_encoder_attn_v_proj_weight"), val = tensor<fp32, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1817906304)))]; |
| 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, []>(1822100672)))]; |
| 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, []>(1822104832)))]; |
| 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, []>(1826299200)))]; |
| 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, []>(1826303360)))]; |
| 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, []>(1826307520)))]; |
| 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, []>(1826323968)))]; |
| 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, []>(1843101248)))]; |
| 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, []>(1843105408)))]; |
| 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, []>(1859882688)))]; |
| 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, []>(1859886848)))]; |
| tensor<fp32, []> var_9 = const()[name = tensor<string, []>("op_9"), val = tensor<fp32, []>(0x1.4f8b58p-17)]; |
| tensor<fp32, []> var_11 = const()[name = tensor<string, []>("op_11"), val = tensor<fp32, []>(0x1p-3)]; |
| tensor<int32, []> var_13 = const()[name = tensor<string, []>("op_13"), val = tensor<int32, []>(-2)]; |
| tensor<fp32, []> var_24 = const()[name = tensor<string, []>("op_24"), val = tensor<fp32, []>(-0x1.fffffep+127)]; |
| tensor<int32, []> var_27 = const()[name = tensor<string, []>("op_27"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_30 = const()[name = tensor<string, []>("op_30"), val = tensor<int32, []>(1)]; |
| tensor<int32, []> const_0 = const()[name = tensor<string, []>("const_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> var_62_axis_0 = const()[name = tensor<string, []>("op_62_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> var_62_batch_dims_0 = const()[name = tensor<string, []>("op_62_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [1, 2, 1024]> var_62 = gather(axis = var_62_axis_0, batch_dims = var_62_batch_dims_0, indices = input_ids, x = decoder_embed_tokens_weight)[name = tensor<string, []>("op_62")]; |
| tensor<fp32, []> var_63 = const()[name = tensor<string, []>("op_63"), val = tensor<fp32, []>(0x1p+5)]; |
| tensor<fp32, [1, 2, 1024]> inputs_embeds = mul(x = var_62, y = var_63)[name = tensor<string, []>("inputs_embeds")]; |
| tensor<int32, [4]> shape_1 = const()[name = tensor<string, []>("shape_1"), val = tensor<int32, [4]>([1, 1, 2, 2])]; |
| tensor<int32, [4]> reshape_1 = const()[name = tensor<string, []>("reshape_1"), val = tensor<int32, [4]>([0, 1, 2, 3])]; |
| tensor<fp32, [4]> reshape_2 = const()[name = tensor<string, []>("reshape_2"), val = tensor<fp32, [4]>([0x0p+0, -0x1.fffffep+127, 0x0p+0, 0x0p+0])]; |
| tensor<fp32, [4]> reshape_3 = const()[name = tensor<string, []>("reshape_3"), val = tensor<fp32, [4]>([0x0p+0, -0x1.fffffep+127, 0x0p+0, 0x0p+0])]; |
| 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, [4]> 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, [1, 1, 2, 2]> reshape_4 = reshape(shape = shape_1, x = scatter_0)[name = tensor<string, []>("reshape_4")]; |
| tensor<int32, [2]> var_117_shape = shape(x = encoder_attention_mask)[name = tensor<string, []>("op_117_shape")]; |
| tensor<int32, []> gather_0 = const()[name = tensor<string, []>("gather_0"), val = tensor<int32, []>(1)]; |
| 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_117_shape)[name = tensor<string, []>("gather_1")]; |
| tensor<int32, [1]> var_120_axes_0 = const()[name = tensor<string, []>("op_120_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1, 1, ?]> var_120 = expand_dims(axes = var_120_axes_0, x = encoder_attention_mask)[name = tensor<string, []>("op_120")]; |
| tensor<int32, [1]> var_121_axes_0 = const()[name = tensor<string, []>("op_121_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1, 1, 1, ?]> var_121 = expand_dims(axes = var_121_axes_0, x = var_120)[name = tensor<string, []>("op_121")]; |
| 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 = (gather_0, var_30, const_0, gather_1))[name = tensor<string, []>("concat_3")]; |
| tensor<int32, [4]> shape_0 = shape(x = var_121)[name = tensor<string, []>("shape_0")]; |
| tensor<int32, []> equal_0_y_0 = const()[name = tensor<string, []>("equal_0_y_0"), val = tensor<int32, []>(-1)]; |
| tensor<bool, [4]> equal_0 = equal(x = concat_3, y = equal_0_y_0)[name = tensor<string, []>("equal_0")]; |
| tensor<int32, [4]> select_0 = select(a = shape_0, b = concat_3, 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<int32, [?, ?, ?, ?]> var_124 = tile(reps = real_div_0, x = var_121)[name = tensor<string, []>("op_124")]; |
| tensor<string, []> expanded_mask_dtype_0 = const()[name = tensor<string, []>("expanded_mask_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, []> const_11 = const()[name = tensor<string, []>("const_11"), val = tensor<fp32, []>(0x1p+0)]; |
| tensor<fp32, [?, ?, ?, ?]> expanded_mask = cast(dtype = expanded_mask_dtype_0, x = var_124)[name = tensor<string, []>("cast_104")]; |
| tensor<fp32, [?, ?, ?, ?]> inverted_mask = sub(x = const_11, y = expanded_mask)[name = tensor<string, []>("inverted_mask")]; |
| tensor<string, []> var_129_dtype_0 = const()[name = tensor<string, []>("op_129_dtype_0"), val = tensor<string, []>("bool")]; |
| tensor<bool, [?, ?, ?, ?]> var_129 = cast(dtype = var_129_dtype_0, x = inverted_mask)[name = tensor<string, []>("cast_103")]; |
| tensor<fp32, [?, ?, ?, ?]> attention_mask_5 = select(a = var_24, b = inverted_mask, cond = var_129)[name = tensor<string, []>("attention_mask_5")]; |
| tensor<bool, [1, 2]> var_134 = not_equal(x = input_ids, y = var_30)[name = tensor<string, []>("op_134")]; |
| tensor<string, []> mask_dtype_0 = const()[name = tensor<string, []>("mask_dtype_0"), val = tensor<string, []>("int32")]; |
| tensor<bool, []> var_136_exclusive_0 = const()[name = tensor<string, []>("op_136_exclusive_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_136_reverse_0 = const()[name = tensor<string, []>("op_136_reverse_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [1, 2]> mask = cast(dtype = mask_dtype_0, x = var_134)[name = tensor<string, []>("cast_102")]; |
| tensor<int32, [1, 2]> var_136 = cumsum(axis = var_30, exclusive = var_136_exclusive_0, reverse = var_136_reverse_0, x = mask)[name = tensor<string, []>("op_136")]; |
| tensor<int32, [1, 2]> incremental_indices = mul(x = var_136, y = mask)[name = tensor<string, []>("incremental_indices")]; |
| tensor<int32, []> var_142 = const()[name = tensor<string, []>("op_142"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1, 2]> var_143 = add(x = incremental_indices, y = var_142)[name = tensor<string, []>("op_143")]; |
| tensor<int32, [1]> var_145 = const()[name = tensor<string, []>("op_145"), val = tensor<int32, [1]>([-1])]; |
| tensor<int32, [2]> var_146 = reshape(shape = var_145, x = var_143)[name = tensor<string, []>("op_146")]; |
| tensor<int32, []> var_147_batch_dims_0 = const()[name = tensor<string, []>("op_147_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [2, 1024]> var_147 = gather(axis = var_27, batch_dims = var_147_batch_dims_0, indices = var_146, x = decoder_embed_positions_weights)[name = tensor<string, []>("op_147")]; |
| tensor<int32, [3]> var_149 = const()[name = tensor<string, []>("op_149"), val = tensor<int32, [3]>([1, 2, 1024])]; |
| tensor<fp32, [1, 2, 1024]> var_150 = reshape(shape = var_149, x = var_147)[name = tensor<string, []>("op_150")]; |
| tensor<fp32, [1, 2, 1024]> input_3 = add(x = inputs_embeds, y = var_150)[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, 2, 1024]> hidden_states_1 = layer_norm(axes = hidden_states_1_axes_0, beta = decoder_layers_0_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_0_self_attn_layer_norm_weight, x = input_3)[name = tensor<string, []>("hidden_states_1")]; |
| tensor<fp32, [1, 2, 1024]> var_174 = 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_175 = const()[name = tensor<string, []>("op_175"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_176 = reshape(shape = var_175, x = var_174)[name = tensor<string, []>("op_176")]; |
| tensor<fp32, [1, 2, 1024]> key_states_1 = 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, 2, 1024]> value_states_1 = 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_184 = const()[name = tensor<string, []>("op_184"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_185 = reshape(shape = var_184, x = key_states_1)[name = tensor<string, []>("op_185")]; |
| tensor<int32, [4]> var_187 = const()[name = tensor<string, []>("op_187"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_188 = reshape(shape = var_187, x = value_states_1)[name = tensor<string, []>("op_188")]; |
| tensor<int32, [4]> value_states_3_perm_0 = const()[name = tensor<string, []>("value_states_3_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<int32, []> const_123 = const()[name = tensor<string, []>("const_123"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_1 = concat(axis = const_123, interleave = key_1_interleave_0, values = var_185)[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, 2, 64]> value_states_3 = transpose(perm = value_states_3_perm_0, x = var_188)[name = tensor<string, []>("transpose_215")]; |
| tensor<fp32, [1, 16, 2, 64]> value_1 = concat(axis = var_13, interleave = value_1_interleave_0, values = value_states_3)[name = tensor<string, []>("value_1")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_0 = mul(x = var_176, y = var_11)[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<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_73 = transpose(perm = transpose_73_perm_0, x = key_1)[name = tensor<string, []>("transpose_213")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = mul_0)[name = tensor<string, []>("transpose_214")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_0 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = transpose_72, y = transpose_73)[name = tensor<string, []>("matmul_0")]; |
| tensor<fp32, [1, 16, 2, 2]> add_0 = add(x = matmul_0, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_204_perm_0 = const()[name = tensor<string, []>("op_204_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_206 = const()[name = tensor<string, []>("op_206"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_204 = transpose(perm = var_204_perm_0, x = attn_output_1)[name = tensor<string, []>("transpose_212")]; |
| tensor<fp32, [1, 2, 1024]> var_207 = reshape(shape = var_206, x = var_204)[name = tensor<string, []>("op_207")]; |
| tensor<fp32, [1, 2, 1024]> input_9 = linear(bias = decoder_layers_0_self_attn_out_proj_bias, weight = decoder_layers_0_self_attn_out_proj_weight, x = var_207)[name = tensor<string, []>("linear_3")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_5 = layer_norm(axes = hidden_states_5_axes_0, beta = decoder_layers_0_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_0_encoder_attn_layer_norm_weight, x = input_11)[name = tensor<string, []>("hidden_states_5")]; |
| tensor<fp32, [1, 2, 1024]> var_231 = 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_232 = const()[name = tensor<string, []>("op_232"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_233 = reshape(shape = var_232, x = var_231)[name = tensor<string, []>("op_233")]; |
| tensor<int32, [4]> query_3_perm_0 = const()[name = tensor<string, []>("query_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_5 = linear(bias = decoder_layers_0_encoder_attn_k_proj_bias, weight = decoder_layers_0_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_5")]; |
| tensor<fp32, [1, ?, 1024]> value_states_5 = linear(bias = decoder_layers_0_encoder_attn_v_proj_bias, weight = decoder_layers_0_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_6")]; |
| tensor<int32, [4]> concat_4x = const()[name = tensor<string, []>("concat_4x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_242 = reshape(shape = concat_4x, x = key_states_5)[name = tensor<string, []>("op_242")]; |
| tensor<int32, [4]> key_states_7_perm_0 = const()[name = tensor<string, []>("key_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_5x = const()[name = tensor<string, []>("concat_5x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_245 = reshape(shape = concat_5x, x = value_states_5)[name = tensor<string, []>("op_245")]; |
| tensor<int32, [4]> value_states_7_perm_0 = const()[name = tensor<string, []>("value_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_3_interleave_0 = const()[name = tensor<string, []>("key_3_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_7 = transpose(perm = key_states_7_perm_0, x = var_242)[name = tensor<string, []>("transpose_210")]; |
| tensor<fp32, [1, 16, ?, 64]> key_3 = concat(axis = var_13, interleave = key_3_interleave_0, values = key_states_7)[name = tensor<string, []>("key_3")]; |
| tensor<bool, []> value_3_interleave_0 = const()[name = tensor<string, []>("value_3_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_7 = transpose(perm = value_states_7_perm_0, x = var_245)[name = tensor<string, []>("transpose_209")]; |
| tensor<fp32, [1, 16, ?, 64]> value_3 = concat(axis = var_13, interleave = value_3_interleave_0, values = value_states_7)[name = tensor<string, []>("value_3")]; |
| tensor<int32, [4]> var_255_shape = shape(x = key_3)[name = tensor<string, []>("op_255_shape")]; |
| tensor<int32, []> gather_3_indices_0 = const()[name = tensor<string, []>("gather_3_indices_0"), val = tensor<int32, []>(2)]; |
| tensor<int32, []> gather_3_axis_0 = const()[name = tensor<string, []>("gather_3_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_3_batch_dims_0 = const()[name = tensor<string, []>("gather_3_batch_dims_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> gather_3 = gather(axis = gather_3_axis_0, batch_dims = gather_3_batch_dims_0, indices = gather_3_indices_0, x = var_255_shape)[name = tensor<string, []>("gather_3")]; |
| tensor<int32, []> concat_6_values0_0 = const()[name = tensor<string, []>("concat_6_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_6_values1_0 = const()[name = tensor<string, []>("concat_6_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_6_values2_0 = const()[name = tensor<string, []>("concat_6_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_6_axis_0 = const()[name = tensor<string, []>("concat_6_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_6_interleave_0 = const()[name = tensor<string, []>("concat_6_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_6 = concat(axis = concat_6_axis_0, interleave = concat_6_interleave_0, values = (concat_6_values0_0, concat_6_values1_0, concat_6_values2_0, gather_3))[name = tensor<string, []>("concat_6")]; |
| 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_6, end_mask = attention_mask_7_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_7")]; |
| tensor<fp32, [1, 16, 2, 64]> query_3 = transpose(perm = query_3_perm_0, x = var_233)[name = tensor<string, []>("transpose_211")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_1 = mul(x = query_3, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_261_perm_0 = const()[name = tensor<string, []>("op_261_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_263 = const()[name = tensor<string, []>("op_263"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_261 = transpose(perm = var_261_perm_0, x = attn_output_5)[name = tensor<string, []>("transpose_208")]; |
| tensor<fp32, [1, 2, ?]> var_264 = reshape(shape = var_263, x = var_261)[name = tensor<string, []>("op_264")]; |
| tensor<fp32, [1, 2, 1024]> input_15 = linear(bias = decoder_layers_0_encoder_attn_out_proj_bias, weight = decoder_layers_0_encoder_attn_out_proj_weight, x = var_264)[name = tensor<string, []>("linear_7")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_19 = layer_norm(axes = input_19_axes_0, beta = decoder_layers_0_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_0_final_layer_norm_weight, x = input_17)[name = tensor<string, []>("input_19")]; |
| tensor<fp32, [1, 2, 4096]> input_21 = linear(bias = decoder_layers_0_fc1_bias, weight = decoder_layers_0_fc1_weight, x = input_19)[name = tensor<string, []>("linear_8")]; |
| tensor<fp32, [1, 2, 4096]> input_23 = relu(x = input_21)[name = tensor<string, []>("input_23")]; |
| tensor<fp32, [1, 2, 1024]> input_27 = linear(bias = decoder_layers_0_fc2_bias, weight = decoder_layers_0_fc2_weight, x = input_23)[name = tensor<string, []>("linear_9")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_11 = layer_norm(axes = hidden_states_11_axes_0, beta = decoder_layers_1_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_1_self_attn_layer_norm_weight, x = input_29)[name = tensor<string, []>("hidden_states_11")]; |
| tensor<fp32, [1, 2, 1024]> var_314 = 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_10")]; |
| tensor<int32, [4]> var_315 = const()[name = tensor<string, []>("op_315"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_316 = reshape(shape = var_315, x = var_314)[name = tensor<string, []>("op_316")]; |
| tensor<fp32, [1, 2, 1024]> key_states_9 = 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_11")]; |
| tensor<fp32, [1, 2, 1024]> value_states_9 = 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_12")]; |
| tensor<int32, [4]> var_324 = const()[name = tensor<string, []>("op_324"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_325 = reshape(shape = var_324, x = key_states_9)[name = tensor<string, []>("op_325")]; |
| tensor<int32, [4]> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_328 = reshape(shape = var_327, x = value_states_9)[name = tensor<string, []>("op_328")]; |
| tensor<int32, [4]> value_states_11_perm_0 = const()[name = tensor<string, []>("value_states_11_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<int32, []> const_124 = const()[name = tensor<string, []>("const_124"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_5 = concat(axis = const_124, interleave = key_5_interleave_0, values = var_325)[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, 2, 64]> value_states_11 = transpose(perm = value_states_11_perm_0, x = var_328)[name = tensor<string, []>("transpose_207")]; |
| tensor<fp32, [1, 16, 2, 64]> value_5 = concat(axis = var_13, interleave = value_5_interleave_0, values = value_states_11)[name = tensor<string, []>("value_5")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_2 = mul(x = var_316, y = var_11)[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<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_75 = transpose(perm = transpose_75_perm_0, x = key_5)[name = tensor<string, []>("transpose_205")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = mul_2)[name = tensor<string, []>("transpose_206")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_2 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_74, y = transpose_75)[name = tensor<string, []>("matmul_2")]; |
| tensor<fp32, [1, 16, 2, 2]> add_2 = add(x = matmul_2, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_344_perm_0 = const()[name = tensor<string, []>("op_344_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_346 = const()[name = tensor<string, []>("op_346"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_344 = transpose(perm = var_344_perm_0, x = attn_output_9)[name = tensor<string, []>("transpose_204")]; |
| tensor<fp32, [1, 2, 1024]> var_347 = reshape(shape = var_346, x = var_344)[name = tensor<string, []>("op_347")]; |
| tensor<fp32, [1, 2, 1024]> input_33 = linear(bias = decoder_layers_1_self_attn_out_proj_bias, weight = decoder_layers_1_self_attn_out_proj_weight, x = var_347)[name = tensor<string, []>("linear_13")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_15 = layer_norm(axes = hidden_states_15_axes_0, beta = decoder_layers_1_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_1_encoder_attn_layer_norm_weight, x = input_35)[name = tensor<string, []>("hidden_states_15")]; |
| tensor<fp32, [1, 2, 1024]> var_371 = 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_14")]; |
| tensor<int32, [4]> var_372 = const()[name = tensor<string, []>("op_372"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_373 = reshape(shape = var_372, x = var_371)[name = tensor<string, []>("op_373")]; |
| tensor<int32, [4]> query_7_perm_0 = const()[name = tensor<string, []>("query_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_13 = linear(bias = decoder_layers_1_encoder_attn_k_proj_bias, weight = decoder_layers_1_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_15")]; |
| tensor<fp32, [1, ?, 1024]> value_states_13 = linear(bias = decoder_layers_1_encoder_attn_v_proj_bias, weight = decoder_layers_1_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_16")]; |
| tensor<int32, [4]> concat_7x = const()[name = tensor<string, []>("concat_7x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_382 = reshape(shape = concat_7x, x = key_states_13)[name = tensor<string, []>("op_382")]; |
| tensor<int32, [4]> key_states_15_perm_0 = const()[name = tensor<string, []>("key_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_8x = const()[name = tensor<string, []>("concat_8x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_385 = reshape(shape = concat_8x, x = value_states_13)[name = tensor<string, []>("op_385")]; |
| tensor<int32, [4]> value_states_15_perm_0 = const()[name = tensor<string, []>("value_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_7_interleave_0 = const()[name = tensor<string, []>("key_7_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_15 = transpose(perm = key_states_15_perm_0, x = var_382)[name = tensor<string, []>("transpose_202")]; |
| tensor<fp32, [1, 16, ?, 64]> key_7 = concat(axis = var_13, interleave = key_7_interleave_0, values = key_states_15)[name = tensor<string, []>("key_7")]; |
| tensor<bool, []> value_7_interleave_0 = const()[name = tensor<string, []>("value_7_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_15 = transpose(perm = value_states_15_perm_0, x = var_385)[name = tensor<string, []>("transpose_201")]; |
| tensor<fp32, [1, 16, ?, 64]> value_7 = concat(axis = var_13, interleave = value_7_interleave_0, values = value_states_15)[name = tensor<string, []>("value_7")]; |
| tensor<int32, [4]> var_395_shape = shape(x = key_7)[name = tensor<string, []>("op_395_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_395_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_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_9, end_mask = attention_mask_11_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_11")]; |
| tensor<fp32, [1, 16, 2, 64]> query_7 = transpose(perm = query_7_perm_0, x = var_373)[name = tensor<string, []>("transpose_203")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_3 = mul(x = query_7, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_401_perm_0 = const()[name = tensor<string, []>("op_401_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_403 = const()[name = tensor<string, []>("op_403"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_401 = transpose(perm = var_401_perm_0, x = attn_output_13)[name = tensor<string, []>("transpose_200")]; |
| tensor<fp32, [1, 2, ?]> var_404 = reshape(shape = var_403, x = var_401)[name = tensor<string, []>("op_404")]; |
| tensor<fp32, [1, 2, 1024]> input_39 = linear(bias = decoder_layers_1_encoder_attn_out_proj_bias, weight = decoder_layers_1_encoder_attn_out_proj_weight, x = var_404)[name = tensor<string, []>("linear_17")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_43 = layer_norm(axes = input_43_axes_0, beta = decoder_layers_1_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_1_final_layer_norm_weight, x = input_41)[name = tensor<string, []>("input_43")]; |
| tensor<fp32, [1, 2, 4096]> input_45 = linear(bias = decoder_layers_1_fc1_bias, weight = decoder_layers_1_fc1_weight, x = input_43)[name = tensor<string, []>("linear_18")]; |
| tensor<fp32, [1, 2, 4096]> input_47 = relu(x = input_45)[name = tensor<string, []>("input_47")]; |
| tensor<fp32, [1, 2, 1024]> input_51 = linear(bias = decoder_layers_1_fc2_bias, weight = decoder_layers_1_fc2_weight, x = input_47)[name = tensor<string, []>("linear_19")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_21 = layer_norm(axes = hidden_states_21_axes_0, beta = decoder_layers_2_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_2_self_attn_layer_norm_weight, x = input_53)[name = tensor<string, []>("hidden_states_21")]; |
| tensor<fp32, [1, 2, 1024]> var_454 = 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_20")]; |
| tensor<int32, [4]> var_455 = const()[name = tensor<string, []>("op_455"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_456 = reshape(shape = var_455, x = var_454)[name = tensor<string, []>("op_456")]; |
| tensor<fp32, [1, 2, 1024]> key_states_17 = 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_21")]; |
| tensor<fp32, [1, 2, 1024]> value_states_17 = 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_22")]; |
| tensor<int32, [4]> var_464 = const()[name = tensor<string, []>("op_464"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_465 = reshape(shape = var_464, x = key_states_17)[name = tensor<string, []>("op_465")]; |
| tensor<int32, [4]> var_467 = const()[name = tensor<string, []>("op_467"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_468 = reshape(shape = var_467, x = value_states_17)[name = tensor<string, []>("op_468")]; |
| tensor<int32, [4]> value_states_19_perm_0 = const()[name = tensor<string, []>("value_states_19_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<int32, []> const_125 = const()[name = tensor<string, []>("const_125"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_9 = concat(axis = const_125, interleave = key_9_interleave_0, values = var_465)[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, 2, 64]> value_states_19 = transpose(perm = value_states_19_perm_0, x = var_468)[name = tensor<string, []>("transpose_199")]; |
| tensor<fp32, [1, 16, 2, 64]> value_9 = concat(axis = var_13, interleave = value_9_interleave_0, values = value_states_19)[name = tensor<string, []>("value_9")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_4 = mul(x = var_456, y = var_11)[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<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_77 = transpose(perm = transpose_77_perm_0, x = key_9)[name = tensor<string, []>("transpose_197")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = mul_4)[name = tensor<string, []>("transpose_198")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_4 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_76, y = transpose_77)[name = tensor<string, []>("matmul_4")]; |
| tensor<fp32, [1, 16, 2, 2]> add_4 = add(x = matmul_4, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_484_perm_0 = const()[name = tensor<string, []>("op_484_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_486 = const()[name = tensor<string, []>("op_486"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_484 = transpose(perm = var_484_perm_0, x = attn_output_17)[name = tensor<string, []>("transpose_196")]; |
| tensor<fp32, [1, 2, 1024]> var_487 = reshape(shape = var_486, x = var_484)[name = tensor<string, []>("op_487")]; |
| tensor<fp32, [1, 2, 1024]> input_57 = linear(bias = decoder_layers_2_self_attn_out_proj_bias, weight = decoder_layers_2_self_attn_out_proj_weight, x = var_487)[name = tensor<string, []>("linear_23")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_25 = layer_norm(axes = hidden_states_25_axes_0, beta = decoder_layers_2_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_2_encoder_attn_layer_norm_weight, x = input_59)[name = tensor<string, []>("hidden_states_25")]; |
| tensor<fp32, [1, 2, 1024]> var_511 = 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_24")]; |
| tensor<int32, [4]> var_512 = const()[name = tensor<string, []>("op_512"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_513 = reshape(shape = var_512, x = var_511)[name = tensor<string, []>("op_513")]; |
| tensor<int32, [4]> query_11_perm_0 = const()[name = tensor<string, []>("query_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_21 = linear(bias = decoder_layers_2_encoder_attn_k_proj_bias, weight = decoder_layers_2_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_25")]; |
| tensor<fp32, [1, ?, 1024]> value_states_21 = linear(bias = decoder_layers_2_encoder_attn_v_proj_bias, weight = decoder_layers_2_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_26")]; |
| tensor<int32, [4]> concat_10x = const()[name = tensor<string, []>("concat_10x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_522 = reshape(shape = concat_10x, x = key_states_21)[name = tensor<string, []>("op_522")]; |
| tensor<int32, [4]> key_states_23_perm_0 = const()[name = tensor<string, []>("key_states_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_11x = const()[name = tensor<string, []>("concat_11x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_525 = reshape(shape = concat_11x, x = value_states_21)[name = tensor<string, []>("op_525")]; |
| tensor<int32, [4]> value_states_23_perm_0 = const()[name = tensor<string, []>("value_states_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_11_interleave_0 = const()[name = tensor<string, []>("key_11_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_23 = transpose(perm = key_states_23_perm_0, x = var_522)[name = tensor<string, []>("transpose_194")]; |
| tensor<fp32, [1, 16, ?, 64]> key_11 = concat(axis = var_13, interleave = key_11_interleave_0, values = key_states_23)[name = tensor<string, []>("key_11")]; |
| tensor<bool, []> value_11_interleave_0 = const()[name = tensor<string, []>("value_11_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_23 = transpose(perm = value_states_23_perm_0, x = var_525)[name = tensor<string, []>("transpose_193")]; |
| tensor<fp32, [1, 16, ?, 64]> value_11 = concat(axis = var_13, interleave = value_11_interleave_0, values = value_states_23)[name = tensor<string, []>("value_11")]; |
| tensor<int32, [4]> var_535_shape = shape(x = key_11)[name = tensor<string, []>("op_535_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_535_shape)[name = tensor<string, []>("gather_7")]; |
| 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_7))[name = tensor<string, []>("concat_12")]; |
| 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_12, end_mask = attention_mask_15_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_15")]; |
| tensor<fp32, [1, 16, 2, 64]> query_11 = transpose(perm = query_11_perm_0, x = var_513)[name = tensor<string, []>("transpose_195")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_5 = mul(x = query_11, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_541_perm_0 = const()[name = tensor<string, []>("op_541_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_543 = const()[name = tensor<string, []>("op_543"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_541 = transpose(perm = var_541_perm_0, x = attn_output_21)[name = tensor<string, []>("transpose_192")]; |
| tensor<fp32, [1, 2, ?]> var_544 = reshape(shape = var_543, x = var_541)[name = tensor<string, []>("op_544")]; |
| tensor<fp32, [1, 2, 1024]> input_63 = linear(bias = decoder_layers_2_encoder_attn_out_proj_bias, weight = decoder_layers_2_encoder_attn_out_proj_weight, x = var_544)[name = tensor<string, []>("linear_27")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_67 = layer_norm(axes = input_67_axes_0, beta = decoder_layers_2_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_2_final_layer_norm_weight, x = input_65)[name = tensor<string, []>("input_67")]; |
| tensor<fp32, [1, 2, 4096]> input_69 = linear(bias = decoder_layers_2_fc1_bias, weight = decoder_layers_2_fc1_weight, x = input_67)[name = tensor<string, []>("linear_28")]; |
| tensor<fp32, [1, 2, 4096]> input_71 = relu(x = input_69)[name = tensor<string, []>("input_71")]; |
| tensor<fp32, [1, 2, 1024]> input_75 = linear(bias = decoder_layers_2_fc2_bias, weight = decoder_layers_2_fc2_weight, x = input_71)[name = tensor<string, []>("linear_29")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_31 = layer_norm(axes = hidden_states_31_axes_0, beta = decoder_layers_3_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_3_self_attn_layer_norm_weight, x = input_77)[name = tensor<string, []>("hidden_states_31")]; |
| tensor<fp32, [1, 2, 1024]> var_594 = 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_30")]; |
| tensor<int32, [4]> var_595 = const()[name = tensor<string, []>("op_595"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_596 = reshape(shape = var_595, x = var_594)[name = tensor<string, []>("op_596")]; |
| tensor<fp32, [1, 2, 1024]> key_states_25 = 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_31")]; |
| tensor<fp32, [1, 2, 1024]> value_states_25 = 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_32")]; |
| tensor<int32, [4]> var_604 = const()[name = tensor<string, []>("op_604"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_605 = reshape(shape = var_604, x = key_states_25)[name = tensor<string, []>("op_605")]; |
| tensor<int32, [4]> var_607 = const()[name = tensor<string, []>("op_607"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_608 = reshape(shape = var_607, x = value_states_25)[name = tensor<string, []>("op_608")]; |
| tensor<int32, [4]> value_states_27_perm_0 = const()[name = tensor<string, []>("value_states_27_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<int32, []> const_126 = const()[name = tensor<string, []>("const_126"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_13 = concat(axis = const_126, interleave = key_13_interleave_0, values = var_605)[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, 2, 64]> value_states_27 = transpose(perm = value_states_27_perm_0, x = var_608)[name = tensor<string, []>("transpose_191")]; |
| tensor<fp32, [1, 16, 2, 64]> value_13 = concat(axis = var_13, interleave = value_13_interleave_0, values = value_states_27)[name = tensor<string, []>("value_13")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_6 = mul(x = var_596, y = var_11)[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<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_79 = transpose(perm = transpose_79_perm_0, x = key_13)[name = tensor<string, []>("transpose_189")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = mul_6)[name = tensor<string, []>("transpose_190")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_6 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_78, y = transpose_79)[name = tensor<string, []>("matmul_6")]; |
| tensor<fp32, [1, 16, 2, 2]> add_6 = add(x = matmul_6, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_624_perm_0 = const()[name = tensor<string, []>("op_624_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_626 = const()[name = tensor<string, []>("op_626"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_624 = transpose(perm = var_624_perm_0, x = attn_output_25)[name = tensor<string, []>("transpose_188")]; |
| tensor<fp32, [1, 2, 1024]> var_627 = reshape(shape = var_626, x = var_624)[name = tensor<string, []>("op_627")]; |
| tensor<fp32, [1, 2, 1024]> input_81 = linear(bias = decoder_layers_3_self_attn_out_proj_bias, weight = decoder_layers_3_self_attn_out_proj_weight, x = var_627)[name = tensor<string, []>("linear_33")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_35 = layer_norm(axes = hidden_states_35_axes_0, beta = decoder_layers_3_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_3_encoder_attn_layer_norm_weight, x = input_83)[name = tensor<string, []>("hidden_states_35")]; |
| tensor<fp32, [1, 2, 1024]> var_651 = 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_34")]; |
| tensor<int32, [4]> var_652 = const()[name = tensor<string, []>("op_652"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_653 = reshape(shape = var_652, x = var_651)[name = tensor<string, []>("op_653")]; |
| tensor<int32, [4]> query_15_perm_0 = const()[name = tensor<string, []>("query_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_29 = linear(bias = decoder_layers_3_encoder_attn_k_proj_bias, weight = decoder_layers_3_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_35")]; |
| tensor<fp32, [1, ?, 1024]> value_states_29 = linear(bias = decoder_layers_3_encoder_attn_v_proj_bias, weight = decoder_layers_3_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_36")]; |
| tensor<int32, [4]> concat_13x = const()[name = tensor<string, []>("concat_13x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_662 = reshape(shape = concat_13x, x = key_states_29)[name = tensor<string, []>("op_662")]; |
| tensor<int32, [4]> key_states_31_perm_0 = const()[name = tensor<string, []>("key_states_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_14x = const()[name = tensor<string, []>("concat_14x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_665 = reshape(shape = concat_14x, x = value_states_29)[name = tensor<string, []>("op_665")]; |
| tensor<int32, [4]> value_states_31_perm_0 = const()[name = tensor<string, []>("value_states_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_15_interleave_0 = const()[name = tensor<string, []>("key_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_31 = transpose(perm = key_states_31_perm_0, x = var_662)[name = tensor<string, []>("transpose_186")]; |
| tensor<fp32, [1, 16, ?, 64]> key_15 = concat(axis = var_13, interleave = key_15_interleave_0, values = key_states_31)[name = tensor<string, []>("key_15")]; |
| tensor<bool, []> value_15_interleave_0 = const()[name = tensor<string, []>("value_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_31 = transpose(perm = value_states_31_perm_0, x = var_665)[name = tensor<string, []>("transpose_185")]; |
| tensor<fp32, [1, 16, ?, 64]> value_15 = concat(axis = var_13, interleave = value_15_interleave_0, values = value_states_31)[name = tensor<string, []>("value_15")]; |
| tensor<int32, [4]> var_675_shape = shape(x = key_15)[name = tensor<string, []>("op_675_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_675_shape)[name = tensor<string, []>("gather_9")]; |
| 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_9))[name = tensor<string, []>("concat_15")]; |
| 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_15, end_mask = attention_mask_19_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_19")]; |
| tensor<fp32, [1, 16, 2, 64]> query_15 = transpose(perm = query_15_perm_0, x = var_653)[name = tensor<string, []>("transpose_187")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_7 = mul(x = query_15, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_681_perm_0 = const()[name = tensor<string, []>("op_681_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_683 = const()[name = tensor<string, []>("op_683"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_681 = transpose(perm = var_681_perm_0, x = attn_output_29)[name = tensor<string, []>("transpose_184")]; |
| tensor<fp32, [1, 2, ?]> var_684 = reshape(shape = var_683, x = var_681)[name = tensor<string, []>("op_684")]; |
| tensor<fp32, [1, 2, 1024]> input_87 = linear(bias = decoder_layers_3_encoder_attn_out_proj_bias, weight = decoder_layers_3_encoder_attn_out_proj_weight, x = var_684)[name = tensor<string, []>("linear_37")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_91 = layer_norm(axes = input_91_axes_0, beta = decoder_layers_3_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_3_final_layer_norm_weight, x = input_89)[name = tensor<string, []>("input_91")]; |
| tensor<fp32, [1, 2, 4096]> input_93 = linear(bias = decoder_layers_3_fc1_bias, weight = decoder_layers_3_fc1_weight, x = input_91)[name = tensor<string, []>("linear_38")]; |
| tensor<fp32, [1, 2, 4096]> input_95 = relu(x = input_93)[name = tensor<string, []>("input_95")]; |
| tensor<fp32, [1, 2, 1024]> input_99 = linear(bias = decoder_layers_3_fc2_bias, weight = decoder_layers_3_fc2_weight, x = input_95)[name = tensor<string, []>("linear_39")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_41 = layer_norm(axes = hidden_states_41_axes_0, beta = decoder_layers_4_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_4_self_attn_layer_norm_weight, x = input_101)[name = tensor<string, []>("hidden_states_41")]; |
| tensor<fp32, [1, 2, 1024]> var_734 = 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_40")]; |
| tensor<int32, [4]> var_735 = const()[name = tensor<string, []>("op_735"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_736 = reshape(shape = var_735, x = var_734)[name = tensor<string, []>("op_736")]; |
| tensor<fp32, [1, 2, 1024]> key_states_33 = 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_41")]; |
| tensor<fp32, [1, 2, 1024]> value_states_33 = 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_42")]; |
| tensor<int32, [4]> var_744 = const()[name = tensor<string, []>("op_744"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_745 = reshape(shape = var_744, x = key_states_33)[name = tensor<string, []>("op_745")]; |
| tensor<int32, [4]> var_747 = const()[name = tensor<string, []>("op_747"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_748 = reshape(shape = var_747, x = value_states_33)[name = tensor<string, []>("op_748")]; |
| tensor<int32, [4]> value_states_35_perm_0 = const()[name = tensor<string, []>("value_states_35_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<int32, []> const_127 = const()[name = tensor<string, []>("const_127"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_17 = concat(axis = const_127, interleave = key_17_interleave_0, values = var_745)[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, 2, 64]> value_states_35 = transpose(perm = value_states_35_perm_0, x = var_748)[name = tensor<string, []>("transpose_183")]; |
| tensor<fp32, [1, 16, 2, 64]> value_17 = concat(axis = var_13, interleave = value_17_interleave_0, values = value_states_35)[name = tensor<string, []>("value_17")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_8 = mul(x = var_736, y = var_11)[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<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_81 = transpose(perm = transpose_81_perm_0, x = key_17)[name = tensor<string, []>("transpose_181")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = mul_8)[name = tensor<string, []>("transpose_182")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_8 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_80, y = transpose_81)[name = tensor<string, []>("matmul_8")]; |
| tensor<fp32, [1, 16, 2, 2]> add_8 = add(x = matmul_8, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_764_perm_0 = const()[name = tensor<string, []>("op_764_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_766 = const()[name = tensor<string, []>("op_766"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_764 = transpose(perm = var_764_perm_0, x = attn_output_33)[name = tensor<string, []>("transpose_180")]; |
| tensor<fp32, [1, 2, 1024]> var_767 = reshape(shape = var_766, x = var_764)[name = tensor<string, []>("op_767")]; |
| tensor<fp32, [1, 2, 1024]> input_105 = linear(bias = decoder_layers_4_self_attn_out_proj_bias, weight = decoder_layers_4_self_attn_out_proj_weight, x = var_767)[name = tensor<string, []>("linear_43")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_45 = layer_norm(axes = hidden_states_45_axes_0, beta = decoder_layers_4_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_4_encoder_attn_layer_norm_weight, x = input_107)[name = tensor<string, []>("hidden_states_45")]; |
| tensor<fp32, [1, 2, 1024]> var_791 = 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_44")]; |
| tensor<int32, [4]> var_792 = const()[name = tensor<string, []>("op_792"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_793 = reshape(shape = var_792, x = var_791)[name = tensor<string, []>("op_793")]; |
| tensor<int32, [4]> query_19_perm_0 = const()[name = tensor<string, []>("query_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_37 = linear(bias = decoder_layers_4_encoder_attn_k_proj_bias, weight = decoder_layers_4_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_45")]; |
| tensor<fp32, [1, ?, 1024]> value_states_37 = linear(bias = decoder_layers_4_encoder_attn_v_proj_bias, weight = decoder_layers_4_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_46")]; |
| tensor<int32, [4]> concat_16x = const()[name = tensor<string, []>("concat_16x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_802 = reshape(shape = concat_16x, x = key_states_37)[name = tensor<string, []>("op_802")]; |
| tensor<int32, [4]> key_states_39_perm_0 = const()[name = tensor<string, []>("key_states_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_17x = const()[name = tensor<string, []>("concat_17x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_805 = reshape(shape = concat_17x, x = value_states_37)[name = tensor<string, []>("op_805")]; |
| tensor<int32, [4]> value_states_39_perm_0 = const()[name = tensor<string, []>("value_states_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_19_interleave_0 = const()[name = tensor<string, []>("key_19_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_39 = transpose(perm = key_states_39_perm_0, x = var_802)[name = tensor<string, []>("transpose_178")]; |
| tensor<fp32, [1, 16, ?, 64]> key_19 = concat(axis = var_13, interleave = key_19_interleave_0, values = key_states_39)[name = tensor<string, []>("key_19")]; |
| tensor<bool, []> value_19_interleave_0 = const()[name = tensor<string, []>("value_19_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_39 = transpose(perm = value_states_39_perm_0, x = var_805)[name = tensor<string, []>("transpose_177")]; |
| tensor<fp32, [1, 16, ?, 64]> value_19 = concat(axis = var_13, interleave = value_19_interleave_0, values = value_states_39)[name = tensor<string, []>("value_19")]; |
| tensor<int32, [4]> var_815_shape = shape(x = key_19)[name = tensor<string, []>("op_815_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_815_shape)[name = tensor<string, []>("gather_11")]; |
| 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_11))[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, 2, 64]> query_19 = transpose(perm = query_19_perm_0, x = var_793)[name = tensor<string, []>("transpose_179")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_9 = mul(x = query_19, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_821_perm_0 = const()[name = tensor<string, []>("op_821_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_823 = const()[name = tensor<string, []>("op_823"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_821 = transpose(perm = var_821_perm_0, x = attn_output_37)[name = tensor<string, []>("transpose_176")]; |
| tensor<fp32, [1, 2, ?]> var_824 = reshape(shape = var_823, x = var_821)[name = tensor<string, []>("op_824")]; |
| tensor<fp32, [1, 2, 1024]> input_111 = linear(bias = decoder_layers_4_encoder_attn_out_proj_bias, weight = decoder_layers_4_encoder_attn_out_proj_weight, x = var_824)[name = tensor<string, []>("linear_47")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_115 = layer_norm(axes = input_115_axes_0, beta = decoder_layers_4_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_4_final_layer_norm_weight, x = input_113)[name = tensor<string, []>("input_115")]; |
| tensor<fp32, [1, 2, 4096]> input_117 = linear(bias = decoder_layers_4_fc1_bias, weight = decoder_layers_4_fc1_weight, x = input_115)[name = tensor<string, []>("linear_48")]; |
| tensor<fp32, [1, 2, 4096]> input_119 = relu(x = input_117)[name = tensor<string, []>("input_119")]; |
| tensor<fp32, [1, 2, 1024]> input_123 = linear(bias = decoder_layers_4_fc2_bias, weight = decoder_layers_4_fc2_weight, x = input_119)[name = tensor<string, []>("linear_49")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_51 = layer_norm(axes = hidden_states_51_axes_0, beta = decoder_layers_5_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_5_self_attn_layer_norm_weight, x = input_125)[name = tensor<string, []>("hidden_states_51")]; |
| tensor<fp32, [1, 2, 1024]> var_874 = 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_50")]; |
| tensor<int32, [4]> var_875 = const()[name = tensor<string, []>("op_875"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_876 = reshape(shape = var_875, x = var_874)[name = tensor<string, []>("op_876")]; |
| tensor<fp32, [1, 2, 1024]> key_states_41 = 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_51")]; |
| tensor<fp32, [1, 2, 1024]> value_states_41 = 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_52")]; |
| tensor<int32, [4]> var_884 = const()[name = tensor<string, []>("op_884"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_885 = reshape(shape = var_884, x = key_states_41)[name = tensor<string, []>("op_885")]; |
| tensor<int32, [4]> var_887 = const()[name = tensor<string, []>("op_887"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_888 = reshape(shape = var_887, x = value_states_41)[name = tensor<string, []>("op_888")]; |
| tensor<int32, [4]> value_states_43_perm_0 = const()[name = tensor<string, []>("value_states_43_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<int32, []> const_128 = const()[name = tensor<string, []>("const_128"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_21 = concat(axis = const_128, interleave = key_21_interleave_0, values = var_885)[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, 2, 64]> value_states_43 = transpose(perm = value_states_43_perm_0, x = var_888)[name = tensor<string, []>("transpose_175")]; |
| tensor<fp32, [1, 16, 2, 64]> value_21 = concat(axis = var_13, interleave = value_21_interleave_0, values = value_states_43)[name = tensor<string, []>("value_21")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_10 = mul(x = var_876, y = var_11)[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<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_83 = transpose(perm = transpose_83_perm_0, x = key_21)[name = tensor<string, []>("transpose_173")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = mul_10)[name = tensor<string, []>("transpose_174")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_10 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_82, y = transpose_83)[name = tensor<string, []>("matmul_10")]; |
| tensor<fp32, [1, 16, 2, 2]> add_10 = add(x = matmul_10, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_904_perm_0 = const()[name = tensor<string, []>("op_904_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_906 = const()[name = tensor<string, []>("op_906"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_904 = transpose(perm = var_904_perm_0, x = attn_output_41)[name = tensor<string, []>("transpose_172")]; |
| tensor<fp32, [1, 2, 1024]> var_907 = reshape(shape = var_906, x = var_904)[name = tensor<string, []>("op_907")]; |
| tensor<fp32, [1, 2, 1024]> input_129 = linear(bias = decoder_layers_5_self_attn_out_proj_bias, weight = decoder_layers_5_self_attn_out_proj_weight, x = var_907)[name = tensor<string, []>("linear_53")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_55 = layer_norm(axes = hidden_states_55_axes_0, beta = decoder_layers_5_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_5_encoder_attn_layer_norm_weight, x = input_131)[name = tensor<string, []>("hidden_states_55")]; |
| tensor<fp32, [1, 2, 1024]> var_931 = 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_54")]; |
| tensor<int32, [4]> var_932 = const()[name = tensor<string, []>("op_932"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_933 = reshape(shape = var_932, x = var_931)[name = tensor<string, []>("op_933")]; |
| tensor<int32, [4]> query_23_perm_0 = const()[name = tensor<string, []>("query_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_45 = linear(bias = decoder_layers_5_encoder_attn_k_proj_bias, weight = decoder_layers_5_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_55")]; |
| tensor<fp32, [1, ?, 1024]> value_states_45 = linear(bias = decoder_layers_5_encoder_attn_v_proj_bias, weight = decoder_layers_5_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_56")]; |
| tensor<int32, [4]> concat_19x = const()[name = tensor<string, []>("concat_19x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_942 = reshape(shape = concat_19x, x = key_states_45)[name = tensor<string, []>("op_942")]; |
| tensor<int32, [4]> key_states_47_perm_0 = const()[name = tensor<string, []>("key_states_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> concat_20x = const()[name = tensor<string, []>("concat_20x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_945 = reshape(shape = concat_20x, x = value_states_45)[name = tensor<string, []>("op_945")]; |
| tensor<int32, [4]> value_states_47_perm_0 = const()[name = tensor<string, []>("value_states_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> key_23_interleave_0 = const()[name = tensor<string, []>("key_23_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_47 = transpose(perm = key_states_47_perm_0, x = var_942)[name = tensor<string, []>("transpose_170")]; |
| tensor<fp32, [1, 16, ?, 64]> key_23 = concat(axis = var_13, interleave = key_23_interleave_0, values = key_states_47)[name = tensor<string, []>("key_23")]; |
| tensor<bool, []> value_23_interleave_0 = const()[name = tensor<string, []>("value_23_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_47 = transpose(perm = value_states_47_perm_0, x = var_945)[name = tensor<string, []>("transpose_169")]; |
| tensor<fp32, [1, 16, ?, 64]> value_23 = concat(axis = var_13, interleave = value_23_interleave_0, values = value_states_47)[name = tensor<string, []>("value_23")]; |
| tensor<int32, [4]> var_955_shape = shape(x = key_23)[name = tensor<string, []>("op_955_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_955_shape)[name = tensor<string, []>("gather_13")]; |
| 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_13))[name = tensor<string, []>("concat_21")]; |
| 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_21, end_mask = attention_mask_27_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_27")]; |
| tensor<fp32, [1, 16, 2, 64]> query_23 = transpose(perm = query_23_perm_0, x = var_933)[name = tensor<string, []>("transpose_171")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_11 = mul(x = query_23, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_961_perm_0 = const()[name = tensor<string, []>("op_961_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_963 = const()[name = tensor<string, []>("op_963"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_961 = transpose(perm = var_961_perm_0, x = attn_output_45)[name = tensor<string, []>("transpose_168")]; |
| tensor<fp32, [1, 2, ?]> var_964 = reshape(shape = var_963, x = var_961)[name = tensor<string, []>("op_964")]; |
| tensor<fp32, [1, 2, 1024]> input_135 = linear(bias = decoder_layers_5_encoder_attn_out_proj_bias, weight = decoder_layers_5_encoder_attn_out_proj_weight, x = var_964)[name = tensor<string, []>("linear_57")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_139 = layer_norm(axes = input_139_axes_0, beta = decoder_layers_5_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_5_final_layer_norm_weight, x = input_137)[name = tensor<string, []>("input_139")]; |
| tensor<fp32, [1, 2, 4096]> input_141 = linear(bias = decoder_layers_5_fc1_bias, weight = decoder_layers_5_fc1_weight, x = input_139)[name = tensor<string, []>("linear_58")]; |
| tensor<fp32, [1, 2, 4096]> input_143 = relu(x = input_141)[name = tensor<string, []>("input_143")]; |
| tensor<fp32, [1, 2, 1024]> input_147 = linear(bias = decoder_layers_5_fc2_bias, weight = decoder_layers_5_fc2_weight, x = input_143)[name = tensor<string, []>("linear_59")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_61 = layer_norm(axes = hidden_states_61_axes_0, beta = decoder_layers_6_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_6_self_attn_layer_norm_weight, x = input_149)[name = tensor<string, []>("hidden_states_61")]; |
| tensor<fp32, [1, 2, 1024]> var_1014 = 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_60")]; |
| tensor<int32, [4]> var_1015 = const()[name = tensor<string, []>("op_1015"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1016 = reshape(shape = var_1015, x = var_1014)[name = tensor<string, []>("op_1016")]; |
| tensor<fp32, [1, 2, 1024]> key_states_49 = 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_61")]; |
| tensor<fp32, [1, 2, 1024]> value_states_49 = 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_62")]; |
| tensor<int32, [4]> var_1024 = const()[name = tensor<string, []>("op_1024"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1025 = reshape(shape = var_1024, x = key_states_49)[name = tensor<string, []>("op_1025")]; |
| tensor<int32, [4]> var_1027 = const()[name = tensor<string, []>("op_1027"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1028 = reshape(shape = var_1027, x = value_states_49)[name = tensor<string, []>("op_1028")]; |
| 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_25_interleave_0 = const()[name = tensor<string, []>("key_25_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_129 = const()[name = tensor<string, []>("const_129"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_25 = concat(axis = const_129, interleave = key_25_interleave_0, values = var_1025)[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, 2, 64]> value_states_51 = transpose(perm = value_states_51_perm_0, x = var_1028)[name = tensor<string, []>("transpose_167")]; |
| tensor<fp32, [1, 16, 2, 64]> value_25 = concat(axis = var_13, interleave = value_25_interleave_0, values = value_states_51)[name = tensor<string, []>("value_25")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_12 = mul(x = var_1016, y = var_11)[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<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_85 = transpose(perm = transpose_85_perm_0, x = key_25)[name = tensor<string, []>("transpose_165")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = mul_12)[name = tensor<string, []>("transpose_166")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_12 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_84, y = transpose_85)[name = tensor<string, []>("matmul_12")]; |
| tensor<fp32, [1, 16, 2, 2]> add_12 = add(x = matmul_12, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1044_perm_0 = const()[name = tensor<string, []>("op_1044_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1046 = const()[name = tensor<string, []>("op_1046"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1044 = transpose(perm = var_1044_perm_0, x = attn_output_49)[name = tensor<string, []>("transpose_164")]; |
| tensor<fp32, [1, 2, 1024]> var_1047 = reshape(shape = var_1046, x = var_1044)[name = tensor<string, []>("op_1047")]; |
| tensor<fp32, [1, 2, 1024]> input_153 = linear(bias = decoder_layers_6_self_attn_out_proj_bias, weight = decoder_layers_6_self_attn_out_proj_weight, x = var_1047)[name = tensor<string, []>("linear_63")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_65 = layer_norm(axes = hidden_states_65_axes_0, beta = decoder_layers_6_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_6_encoder_attn_layer_norm_weight, x = input_155)[name = tensor<string, []>("hidden_states_65")]; |
| tensor<fp32, [1, 2, 1024]> var_1071 = 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_64")]; |
| tensor<int32, [4]> var_1072 = const()[name = tensor<string, []>("op_1072"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1073 = reshape(shape = var_1072, x = var_1071)[name = tensor<string, []>("op_1073")]; |
| tensor<int32, [4]> query_27_perm_0 = const()[name = tensor<string, []>("query_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_53 = linear(bias = decoder_layers_6_encoder_attn_k_proj_bias, weight = decoder_layers_6_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_65")]; |
| tensor<fp32, [1, ?, 1024]> value_states_53 = linear(bias = decoder_layers_6_encoder_attn_v_proj_bias, weight = decoder_layers_6_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_66")]; |
| tensor<int32, [4]> concat_22x = const()[name = tensor<string, []>("concat_22x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1082 = reshape(shape = concat_22x, x = key_states_53)[name = tensor<string, []>("op_1082")]; |
| 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]> concat_23x = const()[name = tensor<string, []>("concat_23x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1085 = reshape(shape = concat_23x, x = value_states_53)[name = tensor<string, []>("op_1085")]; |
| 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_27_interleave_0 = const()[name = tensor<string, []>("key_27_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_55 = transpose(perm = key_states_55_perm_0, x = var_1082)[name = tensor<string, []>("transpose_162")]; |
| tensor<fp32, [1, 16, ?, 64]> key_27 = concat(axis = var_13, interleave = key_27_interleave_0, values = key_states_55)[name = tensor<string, []>("key_27")]; |
| tensor<bool, []> value_27_interleave_0 = const()[name = tensor<string, []>("value_27_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_55 = transpose(perm = value_states_55_perm_0, x = var_1085)[name = tensor<string, []>("transpose_161")]; |
| tensor<fp32, [1, 16, ?, 64]> value_27 = concat(axis = var_13, interleave = value_27_interleave_0, values = value_states_55)[name = tensor<string, []>("value_27")]; |
| tensor<int32, [4]> var_1095_shape = shape(x = key_27)[name = tensor<string, []>("op_1095_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_1095_shape)[name = tensor<string, []>("gather_15")]; |
| 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_15))[name = tensor<string, []>("concat_24")]; |
| 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_24, end_mask = attention_mask_31_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_31")]; |
| tensor<fp32, [1, 16, 2, 64]> query_27 = transpose(perm = query_27_perm_0, x = var_1073)[name = tensor<string, []>("transpose_163")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_13 = mul(x = query_27, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1101_perm_0 = const()[name = tensor<string, []>("op_1101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1103 = const()[name = tensor<string, []>("op_1103"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1101 = transpose(perm = var_1101_perm_0, x = attn_output_53)[name = tensor<string, []>("transpose_160")]; |
| tensor<fp32, [1, 2, ?]> var_1104 = reshape(shape = var_1103, x = var_1101)[name = tensor<string, []>("op_1104")]; |
| tensor<fp32, [1, 2, 1024]> input_159 = linear(bias = decoder_layers_6_encoder_attn_out_proj_bias, weight = decoder_layers_6_encoder_attn_out_proj_weight, x = var_1104)[name = tensor<string, []>("linear_67")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_163 = layer_norm(axes = input_163_axes_0, beta = decoder_layers_6_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_6_final_layer_norm_weight, x = input_161)[name = tensor<string, []>("input_163")]; |
| tensor<fp32, [1, 2, 4096]> input_165 = linear(bias = decoder_layers_6_fc1_bias, weight = decoder_layers_6_fc1_weight, x = input_163)[name = tensor<string, []>("linear_68")]; |
| tensor<fp32, [1, 2, 4096]> input_167 = relu(x = input_165)[name = tensor<string, []>("input_167")]; |
| tensor<fp32, [1, 2, 1024]> input_171 = linear(bias = decoder_layers_6_fc2_bias, weight = decoder_layers_6_fc2_weight, x = input_167)[name = tensor<string, []>("linear_69")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_71 = layer_norm(axes = hidden_states_71_axes_0, beta = decoder_layers_7_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_7_self_attn_layer_norm_weight, x = input_173)[name = tensor<string, []>("hidden_states_71")]; |
| tensor<fp32, [1, 2, 1024]> var_1154 = 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_70")]; |
| tensor<int32, [4]> var_1155 = const()[name = tensor<string, []>("op_1155"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1156 = reshape(shape = var_1155, x = var_1154)[name = tensor<string, []>("op_1156")]; |
| tensor<fp32, [1, 2, 1024]> key_states_57 = 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_71")]; |
| tensor<fp32, [1, 2, 1024]> value_states_57 = 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_72")]; |
| tensor<int32, [4]> var_1164 = const()[name = tensor<string, []>("op_1164"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1165 = reshape(shape = var_1164, x = key_states_57)[name = tensor<string, []>("op_1165")]; |
| tensor<int32, [4]> var_1167 = const()[name = tensor<string, []>("op_1167"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1168 = reshape(shape = var_1167, x = value_states_57)[name = tensor<string, []>("op_1168")]; |
| 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_29_interleave_0 = const()[name = tensor<string, []>("key_29_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_130 = const()[name = tensor<string, []>("const_130"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_29 = concat(axis = const_130, interleave = key_29_interleave_0, values = var_1165)[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, 2, 64]> value_states_59 = transpose(perm = value_states_59_perm_0, x = var_1168)[name = tensor<string, []>("transpose_159")]; |
| tensor<fp32, [1, 16, 2, 64]> value_29 = concat(axis = var_13, interleave = value_29_interleave_0, values = value_states_59)[name = tensor<string, []>("value_29")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_14 = mul(x = var_1156, y = var_11)[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<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_87 = transpose(perm = transpose_87_perm_0, x = key_29)[name = tensor<string, []>("transpose_157")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = mul_14)[name = tensor<string, []>("transpose_158")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_14 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_86, y = transpose_87)[name = tensor<string, []>("matmul_14")]; |
| tensor<fp32, [1, 16, 2, 2]> add_14 = add(x = matmul_14, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1184_perm_0 = const()[name = tensor<string, []>("op_1184_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1186 = const()[name = tensor<string, []>("op_1186"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1184 = transpose(perm = var_1184_perm_0, x = attn_output_57)[name = tensor<string, []>("transpose_156")]; |
| tensor<fp32, [1, 2, 1024]> var_1187 = reshape(shape = var_1186, x = var_1184)[name = tensor<string, []>("op_1187")]; |
| tensor<fp32, [1, 2, 1024]> input_177 = linear(bias = decoder_layers_7_self_attn_out_proj_bias, weight = decoder_layers_7_self_attn_out_proj_weight, x = var_1187)[name = tensor<string, []>("linear_73")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_75 = layer_norm(axes = hidden_states_75_axes_0, beta = decoder_layers_7_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_7_encoder_attn_layer_norm_weight, x = input_179)[name = tensor<string, []>("hidden_states_75")]; |
| tensor<fp32, [1, 2, 1024]> var_1211 = 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_74")]; |
| tensor<int32, [4]> var_1212 = const()[name = tensor<string, []>("op_1212"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1213 = reshape(shape = var_1212, x = var_1211)[name = tensor<string, []>("op_1213")]; |
| tensor<int32, [4]> query_31_perm_0 = const()[name = tensor<string, []>("query_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_61 = linear(bias = decoder_layers_7_encoder_attn_k_proj_bias, weight = decoder_layers_7_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_75")]; |
| tensor<fp32, [1, ?, 1024]> value_states_61 = linear(bias = decoder_layers_7_encoder_attn_v_proj_bias, weight = decoder_layers_7_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_76")]; |
| tensor<int32, [4]> concat_25x = const()[name = tensor<string, []>("concat_25x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1222 = reshape(shape = concat_25x, x = key_states_61)[name = tensor<string, []>("op_1222")]; |
| 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]> concat_26x = const()[name = tensor<string, []>("concat_26x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1225 = reshape(shape = concat_26x, x = value_states_61)[name = tensor<string, []>("op_1225")]; |
| 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_31_interleave_0 = const()[name = tensor<string, []>("key_31_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_63 = transpose(perm = key_states_63_perm_0, x = var_1222)[name = tensor<string, []>("transpose_154")]; |
| tensor<fp32, [1, 16, ?, 64]> key_31 = concat(axis = var_13, interleave = key_31_interleave_0, values = key_states_63)[name = tensor<string, []>("key_31")]; |
| tensor<bool, []> value_31_interleave_0 = const()[name = tensor<string, []>("value_31_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_63 = transpose(perm = value_states_63_perm_0, x = var_1225)[name = tensor<string, []>("transpose_153")]; |
| tensor<fp32, [1, 16, ?, 64]> value_31 = concat(axis = var_13, interleave = value_31_interleave_0, values = value_states_63)[name = tensor<string, []>("value_31")]; |
| tensor<int32, [4]> var_1235_shape = shape(x = key_31)[name = tensor<string, []>("op_1235_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_1235_shape)[name = tensor<string, []>("gather_17")]; |
| 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_17))[name = tensor<string, []>("concat_27")]; |
| 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_27, end_mask = attention_mask_35_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_35")]; |
| tensor<fp32, [1, 16, 2, 64]> query_31 = transpose(perm = query_31_perm_0, x = var_1213)[name = tensor<string, []>("transpose_155")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_15 = mul(x = query_31, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1241_perm_0 = const()[name = tensor<string, []>("op_1241_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1243 = const()[name = tensor<string, []>("op_1243"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1241 = transpose(perm = var_1241_perm_0, x = attn_output_61)[name = tensor<string, []>("transpose_152")]; |
| tensor<fp32, [1, 2, ?]> var_1244 = reshape(shape = var_1243, x = var_1241)[name = tensor<string, []>("op_1244")]; |
| tensor<fp32, [1, 2, 1024]> input_183 = linear(bias = decoder_layers_7_encoder_attn_out_proj_bias, weight = decoder_layers_7_encoder_attn_out_proj_weight, x = var_1244)[name = tensor<string, []>("linear_77")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_187 = layer_norm(axes = input_187_axes_0, beta = decoder_layers_7_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_7_final_layer_norm_weight, x = input_185)[name = tensor<string, []>("input_187")]; |
| tensor<fp32, [1, 2, 4096]> input_189 = linear(bias = decoder_layers_7_fc1_bias, weight = decoder_layers_7_fc1_weight, x = input_187)[name = tensor<string, []>("linear_78")]; |
| tensor<fp32, [1, 2, 4096]> input_191 = relu(x = input_189)[name = tensor<string, []>("input_191")]; |
| tensor<fp32, [1, 2, 1024]> input_195 = linear(bias = decoder_layers_7_fc2_bias, weight = decoder_layers_7_fc2_weight, x = input_191)[name = tensor<string, []>("linear_79")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_81 = layer_norm(axes = hidden_states_81_axes_0, beta = decoder_layers_8_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_8_self_attn_layer_norm_weight, x = input_197)[name = tensor<string, []>("hidden_states_81")]; |
| tensor<fp32, [1, 2, 1024]> var_1294 = 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_80")]; |
| tensor<int32, [4]> var_1295 = const()[name = tensor<string, []>("op_1295"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1296 = reshape(shape = var_1295, x = var_1294)[name = tensor<string, []>("op_1296")]; |
| tensor<fp32, [1, 2, 1024]> key_states_65 = 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_81")]; |
| tensor<fp32, [1, 2, 1024]> value_states_65 = 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_82")]; |
| tensor<int32, [4]> var_1304 = const()[name = tensor<string, []>("op_1304"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1305 = reshape(shape = var_1304, x = key_states_65)[name = tensor<string, []>("op_1305")]; |
| tensor<int32, [4]> var_1307 = const()[name = tensor<string, []>("op_1307"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1308 = reshape(shape = var_1307, x = value_states_65)[name = tensor<string, []>("op_1308")]; |
| 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_33_interleave_0 = const()[name = tensor<string, []>("key_33_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_131 = const()[name = tensor<string, []>("const_131"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_33 = concat(axis = const_131, interleave = key_33_interleave_0, values = var_1305)[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, 2, 64]> value_states_67 = transpose(perm = value_states_67_perm_0, x = var_1308)[name = tensor<string, []>("transpose_151")]; |
| tensor<fp32, [1, 16, 2, 64]> value_33 = concat(axis = var_13, interleave = value_33_interleave_0, values = value_states_67)[name = tensor<string, []>("value_33")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_16 = mul(x = var_1296, y = var_11)[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<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_89 = transpose(perm = transpose_89_perm_0, x = key_33)[name = tensor<string, []>("transpose_149")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_88 = transpose(perm = transpose_88_perm_0, x = mul_16)[name = tensor<string, []>("transpose_150")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_88, y = transpose_89)[name = tensor<string, []>("matmul_16")]; |
| tensor<fp32, [1, 16, 2, 2]> add_16 = add(x = matmul_16, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1324_perm_0 = const()[name = tensor<string, []>("op_1324_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1326 = const()[name = tensor<string, []>("op_1326"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1324 = transpose(perm = var_1324_perm_0, x = attn_output_65)[name = tensor<string, []>("transpose_148")]; |
| tensor<fp32, [1, 2, 1024]> var_1327 = reshape(shape = var_1326, x = var_1324)[name = tensor<string, []>("op_1327")]; |
| tensor<fp32, [1, 2, 1024]> input_201 = linear(bias = decoder_layers_8_self_attn_out_proj_bias, weight = decoder_layers_8_self_attn_out_proj_weight, x = var_1327)[name = tensor<string, []>("linear_83")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_85 = layer_norm(axes = hidden_states_85_axes_0, beta = decoder_layers_8_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_8_encoder_attn_layer_norm_weight, x = input_203)[name = tensor<string, []>("hidden_states_85")]; |
| tensor<fp32, [1, 2, 1024]> var_1351 = 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_84")]; |
| tensor<int32, [4]> var_1352 = const()[name = tensor<string, []>("op_1352"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1353 = reshape(shape = var_1352, x = var_1351)[name = tensor<string, []>("op_1353")]; |
| tensor<int32, [4]> query_35_perm_0 = const()[name = tensor<string, []>("query_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_69 = linear(bias = decoder_layers_8_encoder_attn_k_proj_bias, weight = decoder_layers_8_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_85")]; |
| tensor<fp32, [1, ?, 1024]> value_states_69 = linear(bias = decoder_layers_8_encoder_attn_v_proj_bias, weight = decoder_layers_8_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_86")]; |
| tensor<int32, [4]> concat_28x = const()[name = tensor<string, []>("concat_28x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1362 = reshape(shape = concat_28x, x = key_states_69)[name = tensor<string, []>("op_1362")]; |
| 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]> concat_29x = const()[name = tensor<string, []>("concat_29x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1365 = reshape(shape = concat_29x, x = value_states_69)[name = tensor<string, []>("op_1365")]; |
| 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_35_interleave_0 = const()[name = tensor<string, []>("key_35_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_71 = transpose(perm = key_states_71_perm_0, x = var_1362)[name = tensor<string, []>("transpose_146")]; |
| tensor<fp32, [1, 16, ?, 64]> key_35 = concat(axis = var_13, interleave = key_35_interleave_0, values = key_states_71)[name = tensor<string, []>("key_35")]; |
| tensor<bool, []> value_35_interleave_0 = const()[name = tensor<string, []>("value_35_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_71 = transpose(perm = value_states_71_perm_0, x = var_1365)[name = tensor<string, []>("transpose_145")]; |
| tensor<fp32, [1, 16, ?, 64]> value_35 = concat(axis = var_13, interleave = value_35_interleave_0, values = value_states_71)[name = tensor<string, []>("value_35")]; |
| tensor<int32, [4]> var_1375_shape = shape(x = key_35)[name = tensor<string, []>("op_1375_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_1375_shape)[name = tensor<string, []>("gather_19")]; |
| 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_19))[name = tensor<string, []>("concat_30")]; |
| 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_30, end_mask = attention_mask_39_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_39")]; |
| tensor<fp32, [1, 16, 2, 64]> query_35 = transpose(perm = query_35_perm_0, x = var_1353)[name = tensor<string, []>("transpose_147")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_17 = mul(x = query_35, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1381_perm_0 = const()[name = tensor<string, []>("op_1381_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1383 = const()[name = tensor<string, []>("op_1383"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1381 = transpose(perm = var_1381_perm_0, x = attn_output_69)[name = tensor<string, []>("transpose_144")]; |
| tensor<fp32, [1, 2, ?]> var_1384 = reshape(shape = var_1383, x = var_1381)[name = tensor<string, []>("op_1384")]; |
| tensor<fp32, [1, 2, 1024]> input_207 = linear(bias = decoder_layers_8_encoder_attn_out_proj_bias, weight = decoder_layers_8_encoder_attn_out_proj_weight, x = var_1384)[name = tensor<string, []>("linear_87")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_211 = layer_norm(axes = input_211_axes_0, beta = decoder_layers_8_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_8_final_layer_norm_weight, x = input_209)[name = tensor<string, []>("input_211")]; |
| tensor<fp32, [1, 2, 4096]> input_213 = linear(bias = decoder_layers_8_fc1_bias, weight = decoder_layers_8_fc1_weight, x = input_211)[name = tensor<string, []>("linear_88")]; |
| tensor<fp32, [1, 2, 4096]> input_215 = relu(x = input_213)[name = tensor<string, []>("input_215")]; |
| tensor<fp32, [1, 2, 1024]> input_219 = linear(bias = decoder_layers_8_fc2_bias, weight = decoder_layers_8_fc2_weight, x = input_215)[name = tensor<string, []>("linear_89")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_91 = layer_norm(axes = hidden_states_91_axes_0, beta = decoder_layers_9_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_9_self_attn_layer_norm_weight, x = input_221)[name = tensor<string, []>("hidden_states_91")]; |
| tensor<fp32, [1, 2, 1024]> var_1434 = 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_90")]; |
| tensor<int32, [4]> var_1435 = const()[name = tensor<string, []>("op_1435"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1436 = reshape(shape = var_1435, x = var_1434)[name = tensor<string, []>("op_1436")]; |
| tensor<fp32, [1, 2, 1024]> key_states_73 = 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_91")]; |
| tensor<fp32, [1, 2, 1024]> value_states_73 = 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_92")]; |
| tensor<int32, [4]> var_1444 = const()[name = tensor<string, []>("op_1444"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1445 = reshape(shape = var_1444, x = key_states_73)[name = tensor<string, []>("op_1445")]; |
| tensor<int32, [4]> var_1447 = const()[name = tensor<string, []>("op_1447"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1448 = reshape(shape = var_1447, x = value_states_73)[name = tensor<string, []>("op_1448")]; |
| 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_37_interleave_0 = const()[name = tensor<string, []>("key_37_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_132 = const()[name = tensor<string, []>("const_132"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_37 = concat(axis = const_132, interleave = key_37_interleave_0, values = var_1445)[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, 2, 64]> value_states_75 = transpose(perm = value_states_75_perm_0, x = var_1448)[name = tensor<string, []>("transpose_143")]; |
| tensor<fp32, [1, 16, 2, 64]> value_37 = concat(axis = var_13, interleave = value_37_interleave_0, values = value_states_75)[name = tensor<string, []>("value_37")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_18 = mul(x = var_1436, y = var_11)[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<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_91 = transpose(perm = transpose_91_perm_0, x = key_37)[name = tensor<string, []>("transpose_141")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_90 = transpose(perm = transpose_90_perm_0, x = mul_18)[name = tensor<string, []>("transpose_142")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_18 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_90, y = transpose_91)[name = tensor<string, []>("matmul_18")]; |
| tensor<fp32, [1, 16, 2, 2]> add_18 = add(x = matmul_18, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1464_perm_0 = const()[name = tensor<string, []>("op_1464_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1466 = const()[name = tensor<string, []>("op_1466"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1464 = transpose(perm = var_1464_perm_0, x = attn_output_73)[name = tensor<string, []>("transpose_140")]; |
| tensor<fp32, [1, 2, 1024]> var_1467 = reshape(shape = var_1466, x = var_1464)[name = tensor<string, []>("op_1467")]; |
| tensor<fp32, [1, 2, 1024]> input_225 = linear(bias = decoder_layers_9_self_attn_out_proj_bias, weight = decoder_layers_9_self_attn_out_proj_weight, x = var_1467)[name = tensor<string, []>("linear_93")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_95 = layer_norm(axes = hidden_states_95_axes_0, beta = decoder_layers_9_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_9_encoder_attn_layer_norm_weight, x = input_227)[name = tensor<string, []>("hidden_states_95")]; |
| tensor<fp32, [1, 2, 1024]> var_1491 = 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_94")]; |
| tensor<int32, [4]> var_1492 = const()[name = tensor<string, []>("op_1492"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1493 = reshape(shape = var_1492, x = var_1491)[name = tensor<string, []>("op_1493")]; |
| tensor<int32, [4]> query_39_perm_0 = const()[name = tensor<string, []>("query_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_77 = linear(bias = decoder_layers_9_encoder_attn_k_proj_bias, weight = decoder_layers_9_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_95")]; |
| tensor<fp32, [1, ?, 1024]> value_states_77 = linear(bias = decoder_layers_9_encoder_attn_v_proj_bias, weight = decoder_layers_9_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_96")]; |
| tensor<int32, [4]> concat_31x = const()[name = tensor<string, []>("concat_31x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1502 = reshape(shape = concat_31x, x = key_states_77)[name = tensor<string, []>("op_1502")]; |
| 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]> concat_32x = const()[name = tensor<string, []>("concat_32x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1505 = reshape(shape = concat_32x, x = value_states_77)[name = tensor<string, []>("op_1505")]; |
| 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_39_interleave_0 = const()[name = tensor<string, []>("key_39_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_79 = transpose(perm = key_states_79_perm_0, x = var_1502)[name = tensor<string, []>("transpose_138")]; |
| tensor<fp32, [1, 16, ?, 64]> key_39 = concat(axis = var_13, interleave = key_39_interleave_0, values = key_states_79)[name = tensor<string, []>("key_39")]; |
| tensor<bool, []> value_39_interleave_0 = const()[name = tensor<string, []>("value_39_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_79 = transpose(perm = value_states_79_perm_0, x = var_1505)[name = tensor<string, []>("transpose_137")]; |
| tensor<fp32, [1, 16, ?, 64]> value_39 = concat(axis = var_13, interleave = value_39_interleave_0, values = value_states_79)[name = tensor<string, []>("value_39")]; |
| tensor<int32, [4]> var_1515_shape = shape(x = key_39)[name = tensor<string, []>("op_1515_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_1515_shape)[name = tensor<string, []>("gather_21")]; |
| tensor<int32, []> concat_33_values0_0 = const()[name = tensor<string, []>("concat_33_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_33_values1_0 = const()[name = tensor<string, []>("concat_33_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_33_values2_0 = const()[name = tensor<string, []>("concat_33_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_33_axis_0 = const()[name = tensor<string, []>("concat_33_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_33_interleave_0 = const()[name = tensor<string, []>("concat_33_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_33 = concat(axis = concat_33_axis_0, interleave = concat_33_interleave_0, values = (concat_33_values0_0, concat_33_values1_0, concat_33_values2_0, gather_21))[name = tensor<string, []>("concat_33")]; |
| 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_33, end_mask = attention_mask_43_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_43")]; |
| tensor<fp32, [1, 16, 2, 64]> query_39 = transpose(perm = query_39_perm_0, x = var_1493)[name = tensor<string, []>("transpose_139")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_19 = mul(x = query_39, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1521_perm_0 = const()[name = tensor<string, []>("op_1521_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1523 = const()[name = tensor<string, []>("op_1523"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1521 = transpose(perm = var_1521_perm_0, x = attn_output_77)[name = tensor<string, []>("transpose_136")]; |
| tensor<fp32, [1, 2, ?]> var_1524 = reshape(shape = var_1523, x = var_1521)[name = tensor<string, []>("op_1524")]; |
| tensor<fp32, [1, 2, 1024]> input_231 = linear(bias = decoder_layers_9_encoder_attn_out_proj_bias, weight = decoder_layers_9_encoder_attn_out_proj_weight, x = var_1524)[name = tensor<string, []>("linear_97")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_235 = layer_norm(axes = input_235_axes_0, beta = decoder_layers_9_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_9_final_layer_norm_weight, x = input_233)[name = tensor<string, []>("input_235")]; |
| tensor<fp32, [1, 2, 4096]> input_237 = linear(bias = decoder_layers_9_fc1_bias, weight = decoder_layers_9_fc1_weight, x = input_235)[name = tensor<string, []>("linear_98")]; |
| tensor<fp32, [1, 2, 4096]> input_239 = relu(x = input_237)[name = tensor<string, []>("input_239")]; |
| tensor<fp32, [1, 2, 1024]> input_243 = linear(bias = decoder_layers_9_fc2_bias, weight = decoder_layers_9_fc2_weight, x = input_239)[name = tensor<string, []>("linear_99")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_101 = layer_norm(axes = hidden_states_101_axes_0, beta = decoder_layers_10_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_10_self_attn_layer_norm_weight, x = input_245)[name = tensor<string, []>("hidden_states_101")]; |
| tensor<fp32, [1, 2, 1024]> var_1574 = 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_100")]; |
| tensor<int32, [4]> var_1575 = const()[name = tensor<string, []>("op_1575"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1576 = reshape(shape = var_1575, x = var_1574)[name = tensor<string, []>("op_1576")]; |
| tensor<fp32, [1, 2, 1024]> key_states_81 = 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_101")]; |
| tensor<fp32, [1, 2, 1024]> value_states_81 = 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_102")]; |
| tensor<int32, [4]> var_1584 = const()[name = tensor<string, []>("op_1584"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1585 = reshape(shape = var_1584, x = key_states_81)[name = tensor<string, []>("op_1585")]; |
| tensor<int32, [4]> var_1587 = const()[name = tensor<string, []>("op_1587"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1588 = reshape(shape = var_1587, x = value_states_81)[name = tensor<string, []>("op_1588")]; |
| 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_41_interleave_0 = const()[name = tensor<string, []>("key_41_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_133 = const()[name = tensor<string, []>("const_133"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_41 = concat(axis = const_133, interleave = key_41_interleave_0, values = var_1585)[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, 2, 64]> value_states_83 = transpose(perm = value_states_83_perm_0, x = var_1588)[name = tensor<string, []>("transpose_135")]; |
| tensor<fp32, [1, 16, 2, 64]> value_41 = concat(axis = var_13, interleave = value_41_interleave_0, values = value_states_83)[name = tensor<string, []>("value_41")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_20 = mul(x = var_1576, y = var_11)[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<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_93 = transpose(perm = transpose_93_perm_0, x = key_41)[name = tensor<string, []>("transpose_133")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_92 = transpose(perm = transpose_92_perm_0, x = mul_20)[name = tensor<string, []>("transpose_134")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_20 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_92, y = transpose_93)[name = tensor<string, []>("matmul_20")]; |
| tensor<fp32, [1, 16, 2, 2]> add_20 = add(x = matmul_20, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1604_perm_0 = const()[name = tensor<string, []>("op_1604_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1606 = const()[name = tensor<string, []>("op_1606"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1604 = transpose(perm = var_1604_perm_0, x = attn_output_81)[name = tensor<string, []>("transpose_132")]; |
| tensor<fp32, [1, 2, 1024]> var_1607 = reshape(shape = var_1606, x = var_1604)[name = tensor<string, []>("op_1607")]; |
| tensor<fp32, [1, 2, 1024]> input_249 = linear(bias = decoder_layers_10_self_attn_out_proj_bias, weight = decoder_layers_10_self_attn_out_proj_weight, x = var_1607)[name = tensor<string, []>("linear_103")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_105 = layer_norm(axes = hidden_states_105_axes_0, beta = decoder_layers_10_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_10_encoder_attn_layer_norm_weight, x = input_251)[name = tensor<string, []>("hidden_states_105")]; |
| tensor<fp32, [1, 2, 1024]> var_1631 = 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_104")]; |
| tensor<int32, [4]> var_1632 = const()[name = tensor<string, []>("op_1632"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1633 = reshape(shape = var_1632, x = var_1631)[name = tensor<string, []>("op_1633")]; |
| tensor<int32, [4]> query_43_perm_0 = const()[name = tensor<string, []>("query_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_85 = linear(bias = decoder_layers_10_encoder_attn_k_proj_bias, weight = decoder_layers_10_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_105")]; |
| tensor<fp32, [1, ?, 1024]> value_states_85 = linear(bias = decoder_layers_10_encoder_attn_v_proj_bias, weight = decoder_layers_10_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_106")]; |
| tensor<int32, [4]> concat_34x = const()[name = tensor<string, []>("concat_34x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1642 = reshape(shape = concat_34x, x = key_states_85)[name = tensor<string, []>("op_1642")]; |
| 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]> concat_35x = const()[name = tensor<string, []>("concat_35x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1645 = reshape(shape = concat_35x, x = value_states_85)[name = tensor<string, []>("op_1645")]; |
| 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_43_interleave_0 = const()[name = tensor<string, []>("key_43_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states_87 = transpose(perm = key_states_87_perm_0, x = var_1642)[name = tensor<string, []>("transpose_130")]; |
| tensor<fp32, [1, 16, ?, 64]> key_43 = concat(axis = var_13, interleave = key_43_interleave_0, values = key_states_87)[name = tensor<string, []>("key_43")]; |
| tensor<bool, []> value_43_interleave_0 = const()[name = tensor<string, []>("value_43_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states_87 = transpose(perm = value_states_87_perm_0, x = var_1645)[name = tensor<string, []>("transpose_129")]; |
| tensor<fp32, [1, 16, ?, 64]> value_43 = concat(axis = var_13, interleave = value_43_interleave_0, values = value_states_87)[name = tensor<string, []>("value_43")]; |
| tensor<int32, [4]> var_1655_shape = shape(x = key_43)[name = tensor<string, []>("op_1655_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_1655_shape)[name = tensor<string, []>("gather_23")]; |
| tensor<int32, []> concat_36_values0_0 = const()[name = tensor<string, []>("concat_36_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_36_values1_0 = const()[name = tensor<string, []>("concat_36_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_36_values2_0 = const()[name = tensor<string, []>("concat_36_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_36_axis_0 = const()[name = tensor<string, []>("concat_36_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_36_interleave_0 = const()[name = tensor<string, []>("concat_36_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_36 = concat(axis = concat_36_axis_0, interleave = concat_36_interleave_0, values = (concat_36_values0_0, concat_36_values1_0, concat_36_values2_0, gather_23))[name = tensor<string, []>("concat_36")]; |
| 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_36, end_mask = attention_mask_47_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask_47")]; |
| tensor<fp32, [1, 16, 2, 64]> query_43 = transpose(perm = query_43_perm_0, x = var_1633)[name = tensor<string, []>("transpose_131")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_21 = mul(x = query_43, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1661_perm_0 = const()[name = tensor<string, []>("op_1661_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1663 = const()[name = tensor<string, []>("op_1663"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1661 = transpose(perm = var_1661_perm_0, x = attn_output_85)[name = tensor<string, []>("transpose_128")]; |
| tensor<fp32, [1, 2, ?]> var_1664 = reshape(shape = var_1663, x = var_1661)[name = tensor<string, []>("op_1664")]; |
| tensor<fp32, [1, 2, 1024]> input_255 = linear(bias = decoder_layers_10_encoder_attn_out_proj_bias, weight = decoder_layers_10_encoder_attn_out_proj_weight, x = var_1664)[name = tensor<string, []>("linear_107")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_259 = layer_norm(axes = input_259_axes_0, beta = decoder_layers_10_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_10_final_layer_norm_weight, x = input_257)[name = tensor<string, []>("input_259")]; |
| tensor<fp32, [1, 2, 4096]> input_261 = linear(bias = decoder_layers_10_fc1_bias, weight = decoder_layers_10_fc1_weight, x = input_259)[name = tensor<string, []>("linear_108")]; |
| tensor<fp32, [1, 2, 4096]> input_263 = relu(x = input_261)[name = tensor<string, []>("input_263")]; |
| tensor<fp32, [1, 2, 1024]> input_267 = linear(bias = decoder_layers_10_fc2_bias, weight = decoder_layers_10_fc2_weight, x = input_263)[name = tensor<string, []>("linear_109")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_111 = layer_norm(axes = hidden_states_111_axes_0, beta = decoder_layers_11_self_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_11_self_attn_layer_norm_weight, x = input_269)[name = tensor<string, []>("hidden_states_111")]; |
| tensor<fp32, [1, 2, 1024]> var_1714 = 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_110")]; |
| tensor<int32, [4]> var_1715 = const()[name = tensor<string, []>("op_1715"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1716 = reshape(shape = var_1715, x = var_1714)[name = tensor<string, []>("op_1716")]; |
| tensor<fp32, [1, 2, 1024]> key_states_89 = 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_111")]; |
| tensor<fp32, [1, 2, 1024]> value_states_89 = 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_112")]; |
| tensor<int32, [4]> var_1724 = const()[name = tensor<string, []>("op_1724"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1725 = reshape(shape = var_1724, x = key_states_89)[name = tensor<string, []>("op_1725")]; |
| tensor<int32, [4]> var_1727 = const()[name = tensor<string, []>("op_1727"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1728 = reshape(shape = var_1727, x = value_states_89)[name = tensor<string, []>("op_1728")]; |
| 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_45_interleave_0 = const()[name = tensor<string, []>("key_45_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, []> const_134 = const()[name = tensor<string, []>("const_134"), val = tensor<int32, []>(1)]; |
| tensor<fp32, [1, 2, 16, 64]> key_45 = concat(axis = const_134, interleave = key_45_interleave_0, values = var_1725)[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, 2, 64]> value_states_91 = transpose(perm = value_states_91_perm_0, x = var_1728)[name = tensor<string, []>("transpose_127")]; |
| tensor<fp32, [1, 16, 2, 64]> value_45 = concat(axis = var_13, interleave = value_45_interleave_0, values = value_states_91)[name = tensor<string, []>("value_45")]; |
| tensor<fp32, [1, 2, 16, 64]> mul_22 = mul(x = var_1716, y = var_11)[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<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_95 = transpose(perm = transpose_95_perm_0, x = key_45)[name = tensor<string, []>("transpose_125")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_94 = transpose(perm = transpose_94_perm_0, x = mul_22)[name = tensor<string, []>("transpose_126")]; |
| tensor<fp32, [1, 16, 2, 2]> matmul_22 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_94, y = transpose_95)[name = tensor<string, []>("matmul_22")]; |
| tensor<fp32, [1, 16, 2, 2]> add_22 = add(x = matmul_22, y = reshape_4)[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, [1, 16, 2, 2]> 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, [1, 16, 2, 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_1744_perm_0 = const()[name = tensor<string, []>("op_1744_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1746 = const()[name = tensor<string, []>("op_1746"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1744 = transpose(perm = var_1744_perm_0, x = attn_output_89)[name = tensor<string, []>("transpose_124")]; |
| tensor<fp32, [1, 2, 1024]> var_1747 = reshape(shape = var_1746, x = var_1744)[name = tensor<string, []>("op_1747")]; |
| tensor<fp32, [1, 2, 1024]> input_273 = linear(bias = decoder_layers_11_self_attn_out_proj_bias, weight = decoder_layers_11_self_attn_out_proj_weight, x = var_1747)[name = tensor<string, []>("linear_113")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> hidden_states_115 = layer_norm(axes = hidden_states_115_axes_0, beta = decoder_layers_11_encoder_attn_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_11_encoder_attn_layer_norm_weight, x = input_275)[name = tensor<string, []>("hidden_states_115")]; |
| tensor<fp32, [1, 2, 1024]> var_1771 = 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_114")]; |
| tensor<int32, [4]> var_1772 = const()[name = tensor<string, []>("op_1772"), val = tensor<int32, [4]>([1, 2, -1, 64])]; |
| tensor<fp32, [1, 2, 16, 64]> var_1773 = reshape(shape = var_1772, x = var_1771)[name = tensor<string, []>("op_1773")]; |
| tensor<int32, [4]> query_perm_0 = const()[name = tensor<string, []>("query_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, ?, 1024]> key_states_93 = linear(bias = decoder_layers_11_encoder_attn_k_proj_bias, weight = decoder_layers_11_encoder_attn_k_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_115")]; |
| tensor<fp32, [1, ?, 1024]> value_states_93 = linear(bias = decoder_layers_11_encoder_attn_v_proj_bias, weight = decoder_layers_11_encoder_attn_v_proj_weight, x = encoder_hidden_states)[name = tensor<string, []>("linear_116")]; |
| tensor<int32, [4]> concat_37x = const()[name = tensor<string, []>("concat_37x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1782 = reshape(shape = concat_37x, x = key_states_93)[name = tensor<string, []>("op_1782")]; |
| 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]> concat_38x = const()[name = tensor<string, []>("concat_38x"), val = tensor<int32, [4]>([1, -1, 16, 64])]; |
| tensor<fp32, [1, ?, 16, 64]> var_1785 = reshape(shape = concat_38x, x = value_states_93)[name = tensor<string, []>("op_1785")]; |
| 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_interleave_0 = const()[name = tensor<string, []>("key_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> key_states = transpose(perm = key_states_perm_0, x = var_1782)[name = tensor<string, []>("transpose_122")]; |
| tensor<fp32, [1, 16, ?, 64]> key = concat(axis = var_13, interleave = key_interleave_0, values = key_states)[name = tensor<string, []>("key")]; |
| tensor<bool, []> value_interleave_0 = const()[name = tensor<string, []>("value_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp32, [1, 16, ?, 64]> value_states = transpose(perm = value_states_perm_0, x = var_1785)[name = tensor<string, []>("transpose_121")]; |
| tensor<fp32, [1, 16, ?, 64]> value = concat(axis = var_13, interleave = value_interleave_0, values = value_states)[name = tensor<string, []>("value")]; |
| tensor<int32, [4]> var_1795_shape = shape(x = key)[name = tensor<string, []>("op_1795_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_1795_shape)[name = tensor<string, []>("gather_25")]; |
| tensor<int32, []> concat_39_values0_0 = const()[name = tensor<string, []>("concat_39_values0_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_39_values1_0 = const()[name = tensor<string, []>("concat_39_values1_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_39_values2_0 = const()[name = tensor<string, []>("concat_39_values2_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, []> concat_39_axis_0 = const()[name = tensor<string, []>("concat_39_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<bool, []> concat_39_interleave_0 = const()[name = tensor<string, []>("concat_39_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> concat_39 = concat(axis = concat_39_axis_0, interleave = concat_39_interleave_0, values = (concat_39_values0_0, concat_39_values1_0, concat_39_values2_0, gather_25))[name = tensor<string, []>("concat_39")]; |
| 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_39, end_mask = attention_mask_end_mask_0, x = attention_mask_5)[name = tensor<string, []>("attention_mask")]; |
| tensor<fp32, [1, 16, 2, 64]> query = transpose(perm = query_perm_0, x = var_1773)[name = tensor<string, []>("transpose_123")]; |
| tensor<fp32, [1, 16, 2, 64]> mul_23 = mul(x = query, y = var_11)[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, 2, ?]> 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, 2, ?]> 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, 2, ?]> 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, 2, 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_1801_perm_0 = const()[name = tensor<string, []>("op_1801_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> var_1803 = const()[name = tensor<string, []>("op_1803"), val = tensor<int32, [3]>([1, 2, -1])]; |
| tensor<fp32, [?, 2, 16, 64]> var_1801 = transpose(perm = var_1801_perm_0, x = attn_output_93)[name = tensor<string, []>("transpose_120")]; |
| tensor<fp32, [1, 2, ?]> var_1804 = reshape(shape = var_1803, x = var_1801)[name = tensor<string, []>("op_1804")]; |
| tensor<fp32, [1, 2, 1024]> input_279 = linear(bias = decoder_layers_11_encoder_attn_out_proj_bias, weight = decoder_layers_11_encoder_attn_out_proj_weight, x = var_1804)[name = tensor<string, []>("linear_117")]; |
| tensor<fp32, [1, 2, 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, 2, 1024]> input_283 = layer_norm(axes = input_283_axes_0, beta = decoder_layers_11_final_layer_norm_bias, epsilon = var_9, gamma = decoder_layers_11_final_layer_norm_weight, x = input_281)[name = tensor<string, []>("input_283")]; |
| tensor<fp32, [1, 2, 4096]> input_285 = linear(bias = decoder_layers_11_fc1_bias, weight = decoder_layers_11_fc1_weight, x = input_283)[name = tensor<string, []>("linear_118")]; |
| tensor<fp32, [1, 2, 4096]> input_287 = relu(x = input_285)[name = tensor<string, []>("input_287")]; |
| tensor<fp32, [1, 2, 1024]> input_291 = linear(bias = decoder_layers_11_fc2_bias, weight = decoder_layers_11_fc2_weight, x = input_287)[name = tensor<string, []>("linear_119")]; |
| tensor<fp32, [1, 2, 1024]> input_293 = add(x = input_281, y = input_291)[name = tensor<string, []>("input_293")]; |
| tensor<int32, [1]> var_1838_axes_0 = const()[name = tensor<string, []>("op_1838_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp32, [1, 2, 1024]> var_1838 = layer_norm(axes = var_1838_axes_0, beta = decoder_layer_norm_bias, epsilon = var_9, gamma = decoder_layer_norm_weight, x = input_293)[name = tensor<string, []>("op_1838")]; |
| tensor<int32, [3]> var_1898_begin_0 = const()[name = tensor<string, []>("op_1898_begin_0"), val = tensor<int32, [3]>([0, -1, 0])]; |
| tensor<int32, [3]> var_1898_end_0 = const()[name = tensor<string, []>("op_1898_end_0"), val = tensor<int32, [3]>([1, 2, 1024])]; |
| tensor<bool, [3]> var_1898_end_mask_0 = const()[name = tensor<string, []>("op_1898_end_mask_0"), val = tensor<bool, [3]>([true, true, true])]; |
| tensor<fp32, [1, 1, 1024]> var_1898 = slice_by_index(begin = var_1898_begin_0, end = var_1898_end_0, end_mask = var_1898_end_mask_0, x = var_1838)[name = tensor<string, []>("op_1898")]; |
| tensor<fp32, [256206]> linear_120_bias_0 = const()[name = tensor<string, []>("linear_120_bias_0"), val = tensor<fp32, [256206]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1859891008)))]; |
| tensor<fp32, [1, 1, 256206]> logits = linear(bias = linear_120_bias_0, weight = decoder_embed_tokens_weight, x = var_1898)[name = tensor<string, []>("linear_120")]; |
| tensor<int32, []> var_1908_axis_0 = const()[name = tensor<string, []>("op_1908_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_102_perm_0 = const()[name = tensor<string, []>("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_107 = transpose(perm = transpose_107_perm_0, x = key_45)[name = tensor<string, []>("transpose_108")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_106 = transpose(perm = transpose_106_perm_0, x = key_41)[name = tensor<string, []>("transpose_109")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_105 = transpose(perm = transpose_105_perm_0, x = key_37)[name = tensor<string, []>("transpose_110")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_104 = transpose(perm = transpose_104_perm_0, x = key_33)[name = tensor<string, []>("transpose_111")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_103 = transpose(perm = transpose_103_perm_0, x = key_29)[name = tensor<string, []>("transpose_112")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_102 = transpose(perm = transpose_102_perm_0, x = key_25)[name = tensor<string, []>("transpose_113")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_101 = transpose(perm = transpose_101_perm_0, x = key_21)[name = tensor<string, []>("transpose_114")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_100 = transpose(perm = transpose_100_perm_0, x = key_17)[name = tensor<string, []>("transpose_115")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_99 = transpose(perm = transpose_99_perm_0, x = key_13)[name = tensor<string, []>("transpose_116")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_98 = transpose(perm = transpose_98_perm_0, x = key_9)[name = tensor<string, []>("transpose_117")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_97 = transpose(perm = transpose_97_perm_0, x = key_5)[name = tensor<string, []>("transpose_118")]; |
| tensor<fp32, [1, 16, 2, 64]> transpose_96 = transpose(perm = transpose_96_perm_0, x = key_1)[name = tensor<string, []>("transpose_119")]; |
| tensor<fp32, [12, 1, 16, 2, 64]> past_self_key = stack(axis = var_1908_axis_0, values = (transpose_96, transpose_97, transpose_98, transpose_99, transpose_100, transpose_101, transpose_102, transpose_103, transpose_104, transpose_105, transpose_106, transpose_107))[name = tensor<string, []>("op_1908")]; |
| tensor<int32, []> var_1911_axis_0 = const()[name = tensor<string, []>("op_1911_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, 2, 64]> past_self_value = stack(axis = var_1911_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_1911")]; |
| tensor<int32, []> var_1914_axis_0 = const()[name = tensor<string, []>("op_1914_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> past_cross_key = stack(axis = var_1914_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_1914")]; |
| tensor<int32, []> var_1917_axis_0 = const()[name = tensor<string, []>("op_1917_axis_0"), val = tensor<int32, []>(0)]; |
| tensor<fp32, [12, 1, 16, ?, 64]> past_cross_value = stack(axis = var_1917_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_1917")]; |
| } -> (logits, past_self_key, past_self_value, past_cross_key, past_cross_value); |
| } |
