config.json

{
  "adam_b1": 0.8,
  "adam_b2": 0.99,
  "architectures": [
    "Quantizer"
  ],
  "auto_map": {
    "AutoConfig": "quantizer_config.QuantizerConfig",
    "AutoModel": "models.Quantizer"
  },
  "custom_pipelines": {
    "prosody-embedding": {
      "impl": "prosody_embedding_pipeline.ProsodyEmbeddingPipeline",
      "pt": [
        "AutoModel"
      ],
      "tf": []
    }
  },
  "decoder_depth": 4,
  "decoder_dilation_growth_rate": 3,
  "decoder_downs_t": [
    4
  ],
  "decoder_input_emb_width": 3,
  "decoder_levels": 1,
  "decoder_m_conv": 1.0,
  "decoder_output_emb_width": 64,
  "decoder_strides_t": [
    2
  ],
  "decoder_width": 32,
  "emb_width": 64,
  "encoder_depth": 4,
  "encoder_dilation_growth_rate": 3,
  "encoder_downs_t": [
    4
  ],
  "encoder_input_emb_width": 3,
  "encoder_levels": 1,
  "encoder_m_conv": 1.0,
  "encoder_output_emb_width": 64,
  "encoder_strides_t": [
    2
  ],
  "encoder_width": 32,
  "f0_feats": false,
  "f0_median": false,
  "f0_normalize": true,
  "intensity_normalize": true,
  "l_bins": 320,
  "lambda_commit": 0.02,
  "learning_rate": 0.0002,
  "levels": 1,
  "lr_decay": 0.999,
  "model_type": "prosody_quantizer",
  "mu": 0.99,
  "multispkr": "single",
  "torch_dtype": "float32",
  "transformers_version": "4.53.3"
}