{
  "architectures": [
    "VocosEncodecModel"
  ],
  "bandwidths": [
    1.5,
    3.0,
    6.0,
    12.0
  ],
  "codebook_dim": 128,
  "dtype": "float32",
  "hidden_act": "gelu",
  "hidden_size": 384,
  "hop_length": 320,
  "intermediate_size": 1152,
  "istft_padding": "same",
  "kernel_size": 7,
  "layer_norm_eps": 1e-06,
  "layer_scale_init_value": 0.125,
  "model_type": "vocos_encodec",
  "n_fft": 1280,
  "num_layers": 8,
  "num_quantizers": 16384,
  "padding": 3,
  "sample_rate": 24000,
  "transformers_version": "4.57.0.dev0"
}