models/higgs-q8_0/higgs_audio_v2_tokenizer_config.json
| 1 | { |
| 2 | "acoustic_model_config": { |
| 3 | "codebook_dim": 8, |
| 4 | "codebook_loss_weight": 1.0, |
| 5 | "codebook_size": 1024, |
| 6 | "commitment_loss_weight": 0.25, |
| 7 | "decoder_hidden_size": 1024, |
| 8 | "downsampling_ratios": [8, 5, 4, 2, 3], |
| 9 | "encoder_hidden_size": 64, |
| 10 | "hidden_size": 256, |
| 11 | "hop_length": 960, |
| 12 | "model_type": "dac", |
| 13 | "n_codebooks": 9, |
| 14 | "quantizer_dropout": 0, |
| 15 | "sampling_rate": 16000, |
| 16 | "upsampling_ratios": [8, 5, 4, 2, 3] |
| 17 | }, |
| 18 | "architectures": ["HiggsAudioV2TokenizerModel"], |
| 19 | "block_dilations": [1, 1], |
| 20 | "channel_ratios": [1, 1], |
| 21 | "codebook_dim": 64, |
| 22 | "codebook_size": 1024, |
| 23 | "downsample_factor": 320, |
| 24 | "dtype": "float32", |
| 25 | "initializer_range": 0.02, |
| 26 | "kernel_size": 3, |
| 27 | "model_type": "higgs_audio_v2_tokenizer", |
| 28 | "sample_rate": 24000, |
| 29 | "semantic_model_config": { |
| 30 | "activation_dropout": 0.1, |
| 31 | "apply_spec_augment": true, |
| 32 | "attention_dropout": 0.1, |
| 33 | "bos_token_id": 1, |
| 34 | "classifier_proj_size": 256, |
| 35 | "conv_bias": false, |
| 36 | "conv_dim": [512, 512, 512, 512, 512, 512, 512], |
| 37 | "conv_kernel": [10, 3, 3, 3, 3, 2, 2], |
| 38 | "conv_pos_batch_norm": false, |
| 39 | "conv_stride": [5, 2, 2, 2, 2, 2, 2], |
| 40 | "ctc_loss_reduction": "sum", |
| 41 | "ctc_zero_infinity": false, |
| 42 | "do_stable_layer_norm": false, |
| 43 | "eos_token_id": 2, |
| 44 | "feat_extract_activation": "gelu", |
| 45 | "feat_extract_norm": "group", |
| 46 | "feat_proj_dropout": 0.0, |
| 47 | "feat_proj_layer_norm": true, |
| 48 | "final_dropout": 0.1, |
| 49 | "hidden_act": "gelu", |
| 50 | "hidden_dropout": 0.1, |
| 51 | "hidden_size": 768, |
| 52 | "initializer_range": 0.02, |
| 53 | "intermediate_size": 3072, |
| 54 | "layer_norm_eps": 1e-05, |
| 55 | "layerdrop": 0.1, |
| 56 | "mask_feature_length": 10, |
| 57 | "mask_feature_min_masks": 0, |
| 58 | "mask_feature_prob": 0.0, |
| 59 | "mask_time_length": 10, |
| 60 | "mask_time_min_masks": 2, |
| 61 | "mask_time_prob": 0.0, |
| 62 | "model_type": "hubert", |
| 63 | "num_attention_heads": 12, |
| 64 | "num_conv_pos_embedding_groups": 16, |
| 65 | "num_conv_pos_embeddings": 128, |
| 66 | "num_feat_extract_layers": 7, |
| 67 | "num_hidden_layers": 12, |
| 68 | "pad_token_id": 0, |
| 69 | "use_weighted_layer_sum": false, |
| 70 | "vocab_size": 32 |
| 71 | }, |
| 72 | "semantic_sample_rate": 16000, |
| 73 | "strides": [1, 1], |
| 74 | "target_bandwidths": [0.5, 1, 1.5, 2], |
| 75 | "transformers_version": "5.3.0.dev0", |
| 76 | "unit_kernel_size": 3 |
| 77 | } |
| 78 | |