{
  "model_name": "David-partial_shared-deep_efficiency",
  "run_id": "20251012_060013",
  "timestamp": "2025-10-12T06:40:29.691315",
  "best_val_acc": 83.04,
  "best_epoch": 9,
  "final_train_acc": 90.99773878034635,
  "final_train_loss": 0.9471227525236507,
  "scale_accuracies": {
    "256": 83.04,
    "512": 83.122,
    "768": 83.198,
    "1024": 83.21,
    "1280": 83.248,
    "1536": 83.132,
    "1792": 83.146,
    "2048": 83.144,
    "2304": 82.956,
    "2560": 82.708
  },
  "architecture": {
    "preset": "clip_vit_l14_deep",
    "sharing_mode": "partial_shared",
    "fusion_mode": "deep_efficiency",
    "scales": [
      256,
      512,
      768,
      1024,
      1280,
      1536,
      1792,
      2048,
      2304,
      2560
    ],
    "feature_dim": 768,
    "num_classes": 1000,
    "use_belly": true,
    "belly_expand": 2.0
  },
  "training": {
    "dataset": "AbstractPhil/imagenet-clip-features-orderly",
    "model_variant": "clip_vit_l14",
    "num_epochs": 10,
    "batch_size": 1024,
    "learning_rate": 0.001,
    "rose_weight": "0.1\u21920.5",
    "cayley_loss": false,
    "optimizer": "AdamW",
    "scheduler": "cosine_restarts"
  },
  "files": {
    "weights_pytorch": "weights/best_model.pth",
    "weights_safetensors": "weights/best_model.safetensors",
    "config": "weights/david_config.json",
    "training_config": "weights/train_config.json",
    "tensorboard": "runs/"
  }
}