config.json

{
  "model_type": "dots_tts",
  "latent_dim": 128,
  "patch_size": 4,
  "jit_mode": "1d",
  "encoder_type": "vse",
  "prediction_type": "v",
  "cfg_droprate": 0.2,
  "PatchEncoder": {
    "num_layers": 24,
    "num_heads": 16,
    "hidden_size": 1024,
    "ffn_hidden_size": 4096,
    "modulation": false,
    "qkv_bias": false,
    "qk_norm": true,
    "attn_dropout": 0.0,
    "dropout": 0.1,
    "norm_layer": "RMSNorm",
    "alibi_bias": false,
    "rotary_bias": true,
    "rotary_theta": 10000.0,
    "input_dim": 128,
    "causal": true
  },
  "DiT": {
    "num_layers": 18,
    "num_heads": 16,
    "hidden_size": 1024,
    "ffn_hidden_size": 4096,
    "modulation": true,
    "qkv_bias": false,
    "qk_norm": true,
    "attn_dropout": 0.0,
    "dropout": 0.0,
    "norm_layer": "RMSNorm",
    "alibi_bias": false,
    "rotary_bias": true,
    "rotary_theta": 10000.0,
    "attn_backend": "sdpa",
    "max_tokens": 8192
  },
  "vocoder": {
    "type": "audio_vae",
    "use_cuda_kernel": false,
    "sample_rate": 48000,
    "upsample_rates": [
      10,
      6,
      4,
      2,
      2,
      2
    ],
    "upsample_kernel_sizes": [
      20,
      12,
      8,
      4,
      4,
      4
    ],
    "upsample_initial_channel": 1536,
    "resblock": "1",
    "resblock_kernel_sizes": [
      3,
      7,
      11
    ],
    "resblock_dilation_sizes": [
      [
        1,
        3,
        5
      ],
      [
        1,
        3,
        5
      ],
      [
        1,
        3,
        5
      ]
    ],
    "downsample_rates": [
      2,
      2,
      2,
      4,
      6,
      10
    ],
    "downsample_channels": [
      12,
      24,
      48,
      96,
      192,
      384,
      768
    ],
    "activation": "snakebeta",
    "snake_logscale": true,
    "latent_dim": 128,
    "causal": true,
    "mi_num_layers": 4,
    "mi_skip": true,
    "causal_encoder": true,
    "use_bias_at_final": false,
    "use_tanh_at_final": false
  },
  "fm_sigma": 0.0,
  "t_sample_mode": "uniform",
  "P_mean": -0.8,
  "P_std": 0.8,
  "xvec_drop_rate": 0.2,
  "campplus_embedding_size": 512,
  "xvec_max_audio_seconds": 10.0,
  "architectures": [
    "DotsTTSForConditionalGeneration"
  ]
}