config.json

{
  "model_type": "echo_tts",
  "sample_rate": 44100,
  "max_text_length": 768,
  "max_speaker_latent_length": 6400,
  "audio_downsample_factor": 2048,
  "normalize_text": true,
  "delete_blockwise_modules": false,
  "pca_filename": "pca_state.safetensors",
  "fish_codec_repo": "jordand/fish-s1-dac-min",
  "model_path": "mlx_echo_tts_base",
  "dit": {
    "latent_size": 80,
    "model_size": 2048,
    "num_layers": 24,
    "num_heads": 16,
    "intermediate_size": 5888,
    "norm_eps": 1e-05,
    "text_vocab_size": 256,
    "text_model_size": 1280,
    "text_num_layers": 14,
    "text_num_heads": 10,
    "text_intermediate_size": 3328,
    "speaker_patch_size": 4,
    "speaker_model_size": 1280,
    "speaker_num_layers": 14,
    "speaker_num_heads": 10,
    "speaker_intermediate_size": 3328,
    "timestep_embed_size": 512,
    "adaln_rank": 256
  },
  "sampler": {
    "num_steps": 40,
    "cfg_scale_text": 3.0,
    "cfg_scale_speaker": 8.0,
    "cfg_min_t": 0.5,
    "cfg_max_t": 1.0,
    "truncation_factor": null,
    "rescale_k": null,
    "rescale_sigma": null,
    "speaker_kv_scale": null,
    "speaker_kv_max_layers": null,
    "speaker_kv_min_t": null,
    "sequence_length": 640
  }
}