Update train_config.json - Run 20251013_004438
Browse files
weights/David-decoupled-deep_efficiency/20251013_004438/train_config.json
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"name": "david_training",
|
| 3 |
+
"run_id": "20251013_004438",
|
| 4 |
+
"dataset_name": "AbstractPhil/imagenet-clip-features-orderly",
|
| 5 |
+
"model_variant": [
|
| 6 |
+
"clip_vit_b16",
|
| 7 |
+
"clip_vit_laion_b32",
|
| 8 |
+
"clip_vit_b32"
|
| 9 |
+
],
|
| 10 |
+
"num_classes": 1000,
|
| 11 |
+
"preset": "gated_expert_team",
|
| 12 |
+
"custom_config_path": null,
|
| 13 |
+
"num_classes_override": null,
|
| 14 |
+
"use_belly_override": null,
|
| 15 |
+
"belly_expand_override": null,
|
| 16 |
+
"progressive_training_override": true,
|
| 17 |
+
"scale_warmup_epochs_override": {
|
| 18 |
+
"128": 0,
|
| 19 |
+
"256": 0,
|
| 20 |
+
"384": 0,
|
| 21 |
+
"448": 0,
|
| 22 |
+
"512": 0,
|
| 23 |
+
"576": 0,
|
| 24 |
+
"640": 0,
|
| 25 |
+
"768": 0,
|
| 26 |
+
"896": 0
|
| 27 |
+
},
|
| 28 |
+
"num_epochs": 10,
|
| 29 |
+
"batch_size": 1024,
|
| 30 |
+
"learning_rate": 0.01,
|
| 31 |
+
"weight_decay": 1e-05,
|
| 32 |
+
"warmup_epochs": 3,
|
| 33 |
+
"use_rose_loss": true,
|
| 34 |
+
"rose_initial_weight": 0.1,
|
| 35 |
+
"rose_max_weight": 0.8,
|
| 36 |
+
"rose_weight_schedule": "adaptive",
|
| 37 |
+
"use_cayley_loss": false,
|
| 38 |
+
"cayley_weight": 0.01,
|
| 39 |
+
"scale_loss_balance": null,
|
| 40 |
+
"use_mixed_precision": false,
|
| 41 |
+
"gradient_clip": 10.0,
|
| 42 |
+
"scheduler_type": "cosine_restarts",
|
| 43 |
+
"min_lr": 1e-06,
|
| 44 |
+
"freeze_strategy": "never",
|
| 45 |
+
"freeze_threshold": 90.0,
|
| 46 |
+
"unfreeze_on_plateau": true,
|
| 47 |
+
"patience": 10,
|
| 48 |
+
"track_gradients": true,
|
| 49 |
+
"gradient_scale_threshold": 1e-05,
|
| 50 |
+
"gradient_scale_multiplier": 10.0,
|
| 51 |
+
"log_interval": 50,
|
| 52 |
+
"val_interval": 1,
|
| 53 |
+
"save_interval": 5,
|
| 54 |
+
"log_fusion_weights": true,
|
| 55 |
+
"log_loss_components": true,
|
| 56 |
+
"save_format": "safetensors",
|
| 57 |
+
"hf_repo": "AbstractPhil/david-shared-space",
|
| 58 |
+
"upload_to_hub": true,
|
| 59 |
+
"base_dir": "./david_training",
|
| 60 |
+
"num_workers": 10,
|
| 61 |
+
"pin_memory": true,
|
| 62 |
+
"prefetch_factor": 4,
|
| 63 |
+
"persistent_workers": true
|
| 64 |
+
}
|