weights/David-fully_shared-weighted_sum/20251012_235237/train_config.json

{
  "name": "david_training",
  "run_id": "20251012_235237",
  "dataset_name": "AbstractPhil/imagenet-clip-features-orderly",
  "model_variant": [
    "clip_vit_b16",
    "clip_vit_laion_b32",
    "clip_vit_b32"
  ],
  "num_classes": 1000,
  "preset": "small_fast",
  "custom_config_path": null,
  "num_classes_override": null,
  "use_belly_override": null,
  "belly_expand_override": null,
  "progressive_training_override": true,
  "scale_warmup_epochs_override": {
    "256": 0,
    "512": 2
  },
  "num_epochs": 10,
  "batch_size": 1024,
  "learning_rate": 0.01,
  "weight_decay": 1e-05,
  "warmup_epochs": 3,
  "use_rose_loss": true,
  "rose_initial_weight": 0.1,
  "rose_max_weight": 0.8,
  "rose_weight_schedule": "adaptive",
  "use_cayley_loss": false,
  "cayley_weight": 0.01,
  "scale_loss_balance": null,
  "use_mixed_precision": false,
  "gradient_clip": 15.0,
  "scheduler_type": "cosine_restarts",
  "min_lr": 1e-06,
  "freeze_strategy": "never",
  "freeze_threshold": 90.0,
  "unfreeze_on_plateau": true,
  "patience": 10,
  "track_gradients": true,
  "gradient_scale_threshold": 1e-05,
  "gradient_scale_multiplier": 10.0,
  "log_interval": 50,
  "val_interval": 1,
  "save_interval": 5,
  "log_fusion_weights": true,
  "log_loss_components": true,
  "save_format": "safetensors",
  "hf_repo": "AbstractPhil/david-shared-space",
  "upload_to_hub": true,
  "base_dir": "./david_training",
  "num_workers": 10,
  "pin_memory": true,
  "prefetch_factor": 4,
  "persistent_workers": true
}