{
  "cache_dir": "training/vision_spectrogram_vit/feature_cache",
  "display_name": "DINOv3 ViT-S/16 vision spectrogram",
  "feature_config": {
    "feature": "pooler_output concatenated with patch-token mean and std",
    "feature_shape": [
      1152
    ],
    "head": "MLP logit over cached DINOv3 spectrogram feature",
    "hop_length": 160,
    "image_size": 224,
    "mel_bins": 128,
    "model_id": "facebook/dinov3-vits16-pretrain-lvd1689m",
    "n_fft": 400,
    "sample_rate": 16000,
    "top_db": 80.0,
    "win_length": 400
  },
  "frozen_backbone": "facebook/dinov3-vits16-pretrain-lvd1689m",
  "method": "vision_spectrogram_vit",
  "train_config": {
    "batch_size": 8,
    "cache_root": "training",
    "dataset_root": "dataset",
    "device": "auto",
    "epochs": 20,
    "learning_rate": 0.0003,
    "num_workers": 0,
    "output_root": "final_models",
    "patience": 5,
    "print_progress": true,
    "seed": 153,
    "use_amp": true,
    "weight_decay": 0.0001
  }
}