best_model.json

{
  "model_name": "David-decoupled-deep_efficiency",
  "run_id": "20251013_004438",
  "timestamp": "2025-10-13T01:33:26.455926",
  "best_val_acc": 62.938,
  "best_epoch": 9,
  "final_train_acc": 61.07410925611831,
  "final_train_loss": 3.4686684005744417,
  "scale_accuracies": {
    "128": 62.938,
    "256": 71.08266666666667,
    "384": 73.44466666666666,
    "448": 74.28533333333333,
    "512": 74.61,
    "576": 75.04333333333334,
    "640": 75.17533333333333,
    "768": 75.57533333333333,
    "896": 75.904
  },
  "architecture": {
    "preset": "gated_expert_team",
    "sharing_mode": "decoupled",
    "fusion_mode": "deep_efficiency",
    "scales": [
      128,
      256,
      384,
      448,
      512,
      576,
      640,
      768,
      896
    ],
    "feature_dim": 512,
    "num_classes": 1000,
    "use_belly": true,
    "belly_expand": 4
  },
  "training": {
    "dataset": "AbstractPhil/imagenet-clip-features-orderly",
    "model_variant": [
      "clip_vit_b16",
      "clip_vit_laion_b32",
      "clip_vit_b32"
    ],
    "num_epochs": 10,
    "batch_size": 1024,
    "learning_rate": 0.01,
    "rose_weight": "0.1\u21920.8",
    "cayley_loss": false,
    "optimizer": "AdamW",
    "scheduler": "cosine_restarts"
  },
  "files": {
    "weights_safetensors": "weights/David-decoupled-deep_efficiency/20251013_004438/best_model_acc62.94.safetensors",
    "weights_pytorch": "weights/David-decoupled-deep_efficiency/20251013_004438/best_model.pth",
    "config": "weights/David-decoupled-deep_efficiency/20251013_004438/david_config.json",
    "training_config": "weights/David-decoupled-deep_efficiency/20251013_004438/train_config.json",
    "tensorboard": "runs/David-decoupled-deep_efficiency/20251013_004438/"
  }
}