config.json

{
  "model_type": "cosyvoice3",
  "version": "Fun-CosyVoice3-0.5B-2512",
  "llm": {
    "hidden_size": 896,
    "num_hidden_layers": 24,
    "num_attention_heads": 14,
    "num_key_value_heads": 2,
    "intermediate_size": 4864,
    "head_dim": 64,
    "max_position_embeddings": 32768,
    "vocab_size": 151936,
    "rms_norm_eps": 1e-06,
    "rope_theta": 1000000.0,
    "tie_word_embeddings": true,
    "speech_token_size": 6561,
    "text_token_size": 151936
  },
  "flow": {
    "input_size": 512,
    "output_size": 80,
    "vocab_size": 6561,
    "spk_embed_dim": 192,
    "token_frame_rate": 25,
    "token_mel_ratio": 2,
    "pre_lookahead_len": 3,
    "dit": {
      "dim": 1024,
      "depth": 22,
      "heads": 16,
      "dim_head": 64,
      "ff_mult": 2,
      "mel_dim": 80,
      "spk_dim": 80,
      "static_chunk_size": 50
    }
  },
  "hifigan": {
    "sampling_rate": 24000,
    "in_channels": 80,
    "base_channels": 512,
    "nb_harmonics": 8,
    "upsample_rates": [
      8,
      5,
      3
    ],
    "upsample_kernel_sizes": [
      16,
      11,
      7
    ],
    "istft_n_fft": 16,
    "istft_hop_len": 4,
    "resblock_kernel_sizes": [
      3,
      7,
      11
    ],
    "resblock_dilation_sizes": [
      [
        1,
        3,
        5
      ],
      [
        1,
        3,
        5
      ],
      [
        1,
        3,
        5
      ]
    ],
    "source_resblock_kernel_sizes": [
      7,
      7,
      11
    ],
    "nsf_alpha": 0.1,
    "nsf_sigma": 0.003,
    "nsf_voiced_threshold": 10,
    "audio_limit": 0.99
  },
  "mel": {
    "n_fft": 1920,
    "num_mels": 80,
    "hop_size": 480,
    "win_size": 1920,
    "sample_rate": 24000
  },
  "tokenizer": {
    "type": "fsq",
    "codebook_size": 6561,
    "frame_rate": 25
  },
  "quantization": {
    "bits": 4,
    "group_size": 64,
    "quantized_layers": [
      "q_proj",
      "k_proj",
      "v_proj",
      "o_proj",
      "gate_proj",
      "up_proj",
      "down_proj",
      "speech_head"
    ]
  }
}