config.json

{
  "architectures": [
    "AudioEmbeddingModel"
  ],
  "auto_map": {
    "AutoConfig": "configuration_audio.AudioEmbeddingConfig",
    "AutoModel": "modeling_audio.AudioEmbeddingModel"
  },
  "dtype": "float32",
  "encoder_kwargs": {
    "attn_drop_rate": 0.0,
    "depth": 12,
    "drop_path_rate": 0.1,
    "drop_rate": 0.0,
    "embed_dim": 768,
    "mlp_bias": false,
    "mlp_ratio": 2.6666667,
    "mlp_type": "swiglu",
    "norm_type": "rmsnorm",
    "num_heads": 12,
    "num_patches": 128,
    "pos_embed_type": "rope",
    "proj_bias": false,
    "qk_norm": true,
    "qk_norm_type": "layernorm",
    "qkv_bias": false,
    "residual_type": "standard",
    "rope_mode": "1d"
  },
  "extraction_preset": "overlap50_two_phase",
  "feature_encoder_kwargs": {},
  "hidden_size": 768,
  "model_target": "src.models.best_rq2_module.BestRQ2Module",
  "model_type": "audio_embeddings",
  "patch_embed_kwargs": {
    "bias": false,
    "embed_dim": 768,
    "hmlp_kernel_schedule": null,
    "img_size": [
      128,
      512
    ],
    "in_chans": 1,
    "patch_size": [
      128,
      4
    ],
    "stem_type": "linear"
  },
  "sampling_rate": 16000,
  "spectrogram_adjustment_mode": "truncate",
  "spectrogram_kwargs": {
    "f_max": 8000,
    "f_min": 0,
    "hop_length_ms": 19.53125,
    "n_fft": 2048,
    "n_mels": 128,
    "power": 2.0,
    "sample_rate": 16000,
    "win_length_ms": 64
  },
  "transformers_version": "5.17.0"
}