Upload weights and configs - Run 20251012_060013

weights/best_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c8c1b0b0d1d118d7f880b490776a34704118f99f16b630e4db415a1ae617aec
-size 325845164
+oid sha256:f43bed2fac2969dcf5527abfb501d2730cd33ac87f0e9cd365432e486a8c1bc4
+size 2628344

weights/david_config.json

@@ -1,45 +1,29 @@
 {
-  "name": "david_clip_vit_l14_deep",
-  "uid": "c.david.clip_vit_l14_deep",
-  "feature_dim": 768,
+  "name": "david_small_fast",
+  "uid": "c.david.small_fast",
+  "feature_dim": 512,
   "num_classes": 1000,
   "scales": [
     256,
-    512,
-    768,
-    1024,
-    1280,
-    1536,
-    1792,
-    2048,
-    2304,
-    2560
+    512
   ],
-  "sharing_mode": "partial_shared",
-  "fusion_mode": "deep_efficiency",
-  "use_belly": true,
+  "sharing_mode": "fully_shared",
+  "fusion_mode": "weighted_sum",
+  "use_belly": false,
   "belly_expand": 2.0,
-  "shared_feature_dim": 1024,
-  "shared_layers": 4,
+  "shared_feature_dim": 512,
+  "shared_layers": 1,
   "shared_dropout": 0.1,
   "fusion_temperature": 1.0,
   "fusion_dropout": 0.1,
   "tree_depth": 3,
-  "num_experts": 4,
+  "num_experts": 3,
   "compression_ratio": 4,
   "expert_dropout": 0.1,
   "attention_dropout": 0.1,
-  "progressive_training": true,
+  "progressive_training": false,
   "scale_warmup_epochs": {
     "256": 0,
-    "512": 1,
-    "768": 2,
-    "1024": 3,
-    "1280": 4,
-    "1536": 5,
-    "1792": 6,
-    "2048": 7,
-    "2304": 8,
-    "2560": 9
+    "512": 0
   }
 }

weights/train_config.json

@@ -1,27 +1,16 @@
 {
   "name": "david_training",
-  "run_id": "20251012_050214",
+  "run_id": "20251012_060013",
   "dataset_name": "AbstractPhil/imagenet-clip-features-orderly",
-  "model_variant": "clip_vit_l14",
+  "model_variant": "clip_vit_b16",
   "num_classes": 1000,
-  "preset": "clip_vit_l14_deep",
+  "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": 1,
-    "768": 2,
-    "1024": 3,
-    "1280": 4,
-    "1536": 5,
-    "1792": 6,
-    "2048": 7,
-    "2304": 8,
-    "2560": 9
-  },
+  "progressive_training_override": false,
+  "scale_warmup_epochs_override": null,
   "num_epochs": 10,
   "batch_size": 1024,
   "learning_rate": 0.01,