{
  "architectures": [
    "PeAudioEncoder"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "dac_config": {
    "codebook_dim": 32,
    "codebook_loss_weight": 1.0,
    "codebook_size": 512,
    "commitment_loss_weight": 0.25,
    "decoder_hidden_size": 16,
    "downsampling_ratios": [
      2,
      4,
      4
    ],
    "encoder_hidden_size": 16,
    "hidden_size": 128,
    "hop_length": 32,
    "model_type": "dac",
    "n_codebooks": 6,
    "quantizer_dropout": 0.0,
    "sampling_rate": 16000,
    "upsampling_ratios": [
      4,
      4,
      2
    ]
  },
  "dtype": "float32",
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 32,
  "initializer_range": 0.02,
  "intermediate_size": 37,
  "max_position_embeddings": 512,
  "max_window_layers": 28,
  "model_type": "pe_audio_encoder",
  "num_attention_heads": 2,
  "num_hidden_layers": 2,
  "num_key_value_heads": 2,
  "rms_norm_eps": 1e-05,
  "rope_parameters": {
    "rope_theta": 20000,
    "rope_type": "default"
  },
  "transformers_version": "5.16.0.dev0"
}