UBC-NLP/Simba-S
5246
1{2 "_name_or_path": "facebook/seamless-m4t-v2-large",3 "activation_dropout": 0.0,4 "activation_function": "relu",5 "adaptor_dropout": 0.1,6 "adaptor_kernel_size": 8,7 "adaptor_stride": 8,8 "add_adapter": true,9 "architectures": [10 "SeamlessM4Tv2ForSpeechToText"11 ],12 "attention_dropout": 0.1,13 "bos_token_id": 2,14 "char_vocab_size": 10943,15 "conv_depthwise_kernel_size": 31,16 "decoder_attention_heads": 16,17 "decoder_ffn_dim": 8192,18 "decoder_layerdrop": 0.05,19 "decoder_layers": 24,20 "decoder_start_token_id": 3,21 "dropout": 0.1,22 "encoder_attention_heads": 16,23 "encoder_ffn_dim": 8192,24 "encoder_layerdrop": 0.05,25 "encoder_layers": 24,26 "eos_token_id": 3,27 "feature_projection_input_dim": 160,28 "forced_decoder_ids": null,29 "hidden_size": 1024,30 "initializer_range": 0.02,31 "is_encoder_decoder": true,32 "lang_embed_dim": 256,33 "layer_norm_eps": 1e-05,34 "leaky_relu_slope": 0.1,35 "left_max_position_embeddings": 64,36 "max_new_tokens": 256,37 "max_position_embeddings": 4096,38 "model_type": "seamless_m4t_v2",39 "num_adapter_layers": 1,40 "num_attention_heads": 16,41 "num_hidden_layers": 24,42 "pad_token_id": 0,43 "position_embeddings_type": "relative_key",44 "resblock_dilation_sizes": [45 [46 1,47 3,48 549 ],50 [51 1,52 3,53 554 ],55 [56 1,57 3,58 559 ]60 ],61 "resblock_kernel_sizes": [62 3,63 7,64 1165 ],66 "right_max_position_embeddings": 8,67 "sampling_rate": 16000,68 "scale_embedding": true,69 "speech_encoder_attention_heads": 16,70 "speech_encoder_chunk_size": 20000,71 "speech_encoder_dropout": 0.0,72 "speech_encoder_hidden_act": "swish",73 "speech_encoder_intermediate_size": 4096,74 "speech_encoder_layerdrop": 0.1,75 "speech_encoder_layers": 24,76 "speech_encoder_left_chunk_num": 128,77 "spkr_embed_dim": 256,78 "t2u_bos_token_id": 0,79 "t2u_decoder_attention_heads": 16,80 "t2u_decoder_ffn_dim": 8192,81 "t2u_decoder_layers": 6,82 "t2u_encoder_attention_heads": 16,83 "t2u_encoder_ffn_dim": 8192,84 "t2u_encoder_layers": 6,85 "t2u_eos_token_id": 2,86 "t2u_max_position_embeddings": 4096,87 "t2u_pad_token_id": 1,88 "t2u_variance_pred_dropout": 0.5,89 "t2u_variance_predictor_embed_dim": 1024,90 "t2u_variance_predictor_hidden_dim": 256,91 "t2u_variance_predictor_kernel_size": 3,92 "t2u_vocab_size": 10082,93 "torch_dtype": "float32",94 "transformers_version": "4.48.1",95 "unit_embed_dim": 1280,96 "unit_hifi_gan_vocab_size": 10000,97 "upsample_initial_channel": 512,98 "upsample_kernel_sizes": [99 11,100 8,101 8,102 4,103 4104 ],105 "upsample_rates": [106 5,107 4,108 4,109 2,110 2111 ],112 "use_cache": true,113 "var_pred_dropout": 0.5,114 "variance_predictor_kernel_size": 3,115 "vocab_size": 256102,116 "vocoder_num_langs": 36,117 "vocoder_num_spkrs": 200,118 "vocoder_offset": 4119}120 