config.json

{
  "_name_or_path": "suno/bark",
  "architectures": [
    "BarkModel"
  ],
  "coarse_acoustics_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "BarkCoarseModel"
    ],
    "bias": false,
    "hidden_size": 1024,
    "input_vocab_size": 12096,
    "model_type": "coarse_acoustics",
    "num_heads": 16,
    "num_layers": 24,
    "output_vocab_size": 12096,
    "torch_dtype": "float32"
  },
  "codec_config": {
    "_name_or_path": "facebook/encodec_24khz",
    "architectures": [
      "EncodecModel"
    ],
    "model_type": "encodec",
    "torch_dtype": "float32"
  },
  "fine_acoustics_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "BarkFineModel"
    ],
    "bias": false,
    "hidden_size": 1024,
    "input_vocab_size": 1056,
    "model_type": "fine_acoustics",
    "num_heads": 16,
    "num_layers": 24,
    "output_vocab_size": 1056,
    "torch_dtype": "float32"
  },
  "flash_attention_version": "2.1.0",
  "initializer_range": 0.02,
  "model_type": "bark",
  "semantic_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "BarkSemanticModel"
    ],
    "bias": false,
    "hidden_size": 1024,
    "input_vocab_size": 129600,
    "model_type": "semantic",
    "num_heads": 16,
    "num_layers": 24,
    "torch_dtype": "float32"
  },
  "torch_dtype": "float16",
  "transformers_version": "4.48.2",
  "use_flash_attention_2": true
}