Upload weights and configs - Run 20251012_050214

weights/best_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cedc682eef6b2abd97c43b3994b107bf28565e59082e0c4844bc193dedca1ab8
-size 64195804
+oid sha256:2adddce311fccc3820403cc4fca6539bbd4c15b1f9c498ec363f4afe58ad5787
+size 325845164

weights/david_config.json

@@ -1,20 +1,26 @@
 {
-  "name": "david_clip_vit_l14",
-  "uid": "c.david.clip_vit_l14",
+  "name": "david_clip_vit_l14_deep",
+  "uid": "c.david.clip_vit_l14_deep",
   "feature_dim": 768,
   "num_classes": 1000,
   "scales": [
-    384,
+    256,
+    512,
     768,
     1024,
-    1280
+    1280,
+    1536,
+    1792,
+    2048,
+    2304,
+    2560
   ],
   "sharing_mode": "partial_shared",
   "fusion_mode": "deep_efficiency",
   "use_belly": true,
   "belly_expand": 2.0,
   "shared_feature_dim": 1024,
-  "shared_layers": 2,
+  "shared_layers": 4,
   "shared_dropout": 0.1,
   "fusion_temperature": 1.0,
   "fusion_dropout": 0.1,
@@ -25,9 +31,15 @@
   "attention_dropout": 0.1,
   "progressive_training": true,
   "scale_warmup_epochs": {
-    "384": 0,
-    "768": 1,
-    "1024": 2,
-    "1280": 3
+    "256": 0,
+    "512": 1,
+    "768": 2,
+    "1024": 3,
+    "1280": 4,
+    "1536": 5,
+    "1792": 6,
+    "2048": 7,
+    "2304": 8,
+    "2560": 9
   }
 }

weights/train_config.json

@@ -1,26 +1,32 @@
 {
   "name": "david_training",
-  "run_id": "20251012_041353",
+  "run_id": "20251012_050214",
   "dataset_name": "AbstractPhil/imagenet-clip-features-orderly",
   "model_variant": "clip_vit_l14",
   "num_classes": 1000,
-  "preset": "clip_vit_l14",
+  "preset": "clip_vit_l14_deep",
   "custom_config_path": null,
   "num_classes_override": null,
   "use_belly_override": null,
   "belly_expand_override": null,
   "progressive_training_override": true,
   "scale_warmup_epochs_override": {
-    "384": 0,
-    "768": 1,
-    "1024": 2,
-    "1280": 3
+    "256": 0,
+    "512": 1,
+    "768": 2,
+    "1024": 3,
+    "1280": 4,
+    "1536": 5,
+    "1792": 6,
+    "2048": 7,
+    "2304": 8,
+    "2560": 9
   },
-  "num_epochs": 20,
+  "num_epochs": 10,
   "batch_size": 1024,
   "learning_rate": 0.01,
   "weight_decay": 1e-05,
-  "warmup_epochs": 3,
+  "warmup_epochs": 0,
   "use_rose_loss": true,
   "rose_initial_weight": 0.1,
   "rose_max_weight": 0.5,