hf-tiny-v2/tiny-random-SpeechEncoderDecoderModel-wav2vec2-bert

1{2  "architectures": [3    "SpeechEncoderDecoderModel"4  ],5  "decoder": {6    "_name_or_path": "/tmp/tmp73xden4h/decoder/BertLMHeadModel",7    "add_cross_attention": true,8    "architectures": [9      "BertLMHeadModel"10    ],11    "attention_probs_dropout_prob": 0.1,12    "bos_token_id": null,13    "chunk_size_feed_forward": 0,14    "classifier_dropout": null,15    "dtype": "float32",16    "eos_token_id": null,17    "hidden_act": "gelu",18    "hidden_dropout_prob": 0.1,19    "hidden_size": 32,20    "id2label": {21      "0": "LABEL_0",22      "1": "LABEL_1"23    },24    "initializer_range": 0.02,25    "intermediate_size": 37,26    "is_decoder": true,27    "is_encoder_decoder": false,28    "label2id": {29      "LABEL_0": 0,30      "LABEL_1": 131    },32    "layer_norm_eps": 1e-12,33    "max_position_embeddings": 512,34    "model_type": "bert",35    "num_attention_heads": 4,36    "num_hidden_layers": 2,37    "output_attentions": false,38    "output_hidden_states": false,39    "pad_token_id": 0,40    "problem_type": null,41    "return_dict": true,42    "tie_word_embeddings": true,43    "type_vocab_size": 16,44    "use_cache": true,45    "vocab_size": 3052246  },47  "dtype": "float32",48  "encoder": {49    "_name_or_path": "/tmp/tmp73xden4h/encoder/Wav2Vec2Model",50    "activation_dropout": 0.1,51    "adapter_attn_dim": null,52    "adapter_kernel_size": 3,53    "adapter_stride": 2,54    "add_adapter": false,55    "apply_spec_augment": true,56    "architectures": [57      "Wav2Vec2Model"58    ],59    "attention_dropout": 0.1,60    "bos_token_id": 1,61    "chunk_size_feed_forward": 0,62    "classifier_proj_size": 256,63    "codevector_dim": 256,64    "contrastive_logits_temperature": 0.1,65    "conv_bias": false,66    "conv_dim": [67      32,68      32,69      3270    ],71    "conv_kernel": [72      8,73      8,74      875    ],76    "conv_stride": [77      4,78      4,79      480    ],81    "ctc_loss_reduction": "sum",82    "ctc_zero_infinity": false,83    "diversity_loss_weight": 0.1,84    "do_stable_layer_norm": false,85    "dtype": "float32",86    "eos_token_id": 2,87    "feat_extract_activation": "gelu",88    "feat_extract_dropout": 0.0,89    "feat_extract_norm": "group",90    "feat_proj_dropout": 0.0,91    "feat_quantizer_dropout": 0.0,92    "final_dropout": 0.1,93    "hidden_act": "gelu",94    "hidden_dropout": 0.1,95    "hidden_dropout_prob": 0.1,96    "hidden_size": 16,97    "id2label": {98      "0": "LABEL_0",99      "1": "LABEL_1"100    },101    "initializer_range": 0.02,102    "intermediate_size": 20,103    "is_encoder_decoder": false,104    "label2id": {105      "LABEL_0": 0,106      "LABEL_1": 1107    },108    "layer_norm_eps": 1e-05,109    "layerdrop": 0.1,110    "mask_feature_length": 10,111    "mask_feature_min_masks": 0,112    "mask_feature_prob": 0.0,113    "mask_time_length": 2,114    "mask_time_min_masks": 2,115    "mask_time_prob": 0.5,116    "model_type": "wav2vec2",117    "num_adapter_layers": 1,118    "num_attention_heads": 2,119    "num_codevector_groups": 2,120    "num_codevectors_per_group": 320,121    "num_conv_pos_embedding_groups": 2,122    "num_conv_pos_embeddings": 16,123    "num_feat_extract_layers": 3,124    "num_hidden_layers": 2,125    "num_negatives": 100,126    "output_attentions": false,127    "output_hidden_size": 16,128    "output_hidden_states": false,129    "pad_token_id": 0,130    "problem_type": null,131    "proj_codevector_dim": 256,132    "return_dict": true,133    "tdnn_dilation": [134      1,135      2136    ],137    "tdnn_dim": [138      32,139      32140    ],141    "tdnn_kernel": [142      5,143      3144    ],145    "use_weighted_layer_sum": false,146    "vocab_size": 32,147    "xvector_output_dim": 32148  },149  "is_encoder_decoder": true,150  "model_type": "speech-encoder-decoder",151  "tie_word_embeddings": false,152  "transformers_version": "5.16.0.dev0"153}154