Update README - Run 20251104_154540

README.md

@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 76.50
+      value: 76.60
 ---
 
 # David: Multi-Scale Feature Classifier
@@ -32,16 +32,16 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: clip_vit_b16_cantor
+- **Preset**: clip_vit_b16_cantor_big_window
 - **Sharing Mode**: decoupled
 - **Fusion Mode**: cantor_scale
-- **Scales**: [256, 512, 768, 1024]
+- **Scales**: [256, 512, 768, 1024, 2048, 4096]
 - **Feature Dim**: 512
-- **Parameters**: 9,057,029
+- **Parameters**: 60,452,103
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_laion_b32
+- **Model Variant**: clip_vit_b16
 - **Epochs**: 5
 - **Batch Size**: 512
 - **Learning Rate**: 0.001
@@ -51,15 +51,17 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 76.50%
-- **Best Epoch**: 4
-- **Final Train Accuracy**: 82.74%
+- **Validation Accuracy**: 76.60%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 75.20%
 
 ### Per-Scale Performance
-- **Scale 256**: 71.27%
+- **Scale 256**: 71.98%
 - **Scale 512**: 74.69%
-- **Scale 768**: 75.03%
-- **Scale 1024**: 75.53%
+- **Scale 768**: 75.59%
+- **Scale 1024**: 75.88%
+- **Scale 2048**: 76.12%
+- **Scale 4096**: 75.64%
 
 
 ## Usage
@@ -76,19 +78,19 @@ AbstractPhil/gated-david/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_clip_vit_b16_cantor/
-│       └── 20251104_152832/
+│   └── clip_vit_b16_cantor_big_window/
+│       └── 20251104_154540/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc76.50.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc76.50_metadata.json
+│           ├── best_model_acc76.60.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc76.60_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_clip_vit_b16_cantor/
-        └── 20251104_152832/
+    └── clip_vit_b16_cantor_big_window/
+        └── 20251104_154540/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -101,9 +103,9 @@ from huggingface_hub import hf_hub_download
 # Browse available models in MODELS_INDEX.json first!
 
 # Specify model variant and run
-model_name = "david_clip_vit_b16_cantor"
-run_id = "20251104_152832"
-accuracy = "76.50"  # From MODELS_INDEX.json
+model_name = "clip_vit_b16_cantor_big_window"
+run_id = "20251104_154540"
+accuracy = "76.60"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -152,7 +154,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (256, 512, 768, 1024),
+David processes inputs at multiple scales (256, 512, 768, 1024, 2048, 4096),
 allowing it to capture both coarse and fine-grained features.
 
 ### Feature Geometry
@@ -194,7 +196,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
   author = {AbstractPhil},
   year = {2025},
   url = {https://huggingface.co/AbstractPhil/gated-david},
-  note = {Run ID: 20251104_152832}
+  note = {Run ID: 20251104_154540}
 }
 ```
 
@@ -209,4 +211,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-11-04 15:39:28*
+*Generated on 2025-11-04 15:47:54*