Update README - Run 20251104_140742

README.md

@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 82.86
+      value: 83.52
 ---
 
 # David: Multi-Scale Feature Classifier
@@ -32,18 +32,18 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: clip_vit_l14_cantor
+- **Preset**: clip_vit_bigg14_cantor
 - **Sharing Mode**: partial_shared
 - **Fusion Mode**: cantor_scale
-- **Scales**: [384, 768, 1024, 1280, 1536]
-- **Feature Dim**: 768
-- **Parameters**: 32,436,998
+- **Scales**: [384, 512, 768, 1024, 1280, 1536, 1792, 2048]
+- **Feature Dim**: 1280
+- **Parameters**: 82,601,993
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_l14
+- **Model Variant**: clip_vit_laion_bigg14
 - **Epochs**: 5
-- **Batch Size**: 1024
+- **Batch Size**: 512
 - **Learning Rate**: 0.001
 - **Rose Loss Weight**: 0.1 → 0.5
 - **Cayley Loss**: False
@@ -51,16 +51,19 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 82.86%
-- **Best Epoch**: 3
-- **Final Train Accuracy**: 90.33%
+- **Validation Accuracy**: 83.52%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 82.03%
 
 ### Per-Scale Performance
-- **Scale 384**: 82.21%
-- **Scale 768**: 82.47%
-- **Scale 1024**: 82.25%
-- **Scale 1280**: 82.17%
-- **Scale 1536**: 82.25%
+- **Scale 384**: 82.86%
+- **Scale 512**: 83.22%
+- **Scale 768**: 83.09%
+- **Scale 1024**: 83.15%
+- **Scale 1280**: 83.17%
+- **Scale 1536**: 83.18%
+- **Scale 1792**: 82.98%
+- **Scale 2048**: 82.79%
 
 
 ## Usage
@@ -77,19 +80,19 @@ AbstractPhil/gated-david/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_clip_vit_l14_cantor/
-│       └── 20251104_133602/
+│   └── david_clip_vit_bigg14_cantor/
+│       └── 20251104_140742/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc82.86.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc82.86_metadata.json
+│           ├── best_model_acc83.52.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc83.52_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_clip_vit_l14_cantor/
-        └── 20251104_133602/
+    └── david_clip_vit_bigg14_cantor/
+        └── 20251104_140742/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -102,9 +105,9 @@ 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_l14_cantor"
-run_id = "20251104_133602"
-accuracy = "82.86"  # From MODELS_INDEX.json
+model_name = "david_clip_vit_bigg14_cantor"
+run_id = "20251104_140742"
+accuracy = "83.52"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -153,7 +156,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (384, 768, 1024, 1280, 1536),
+David processes inputs at multiple scales (384, 512, 768, 1024, 1280, 1536, 1792, 2048),
 allowing it to capture both coarse and fine-grained features.
 
 ### Feature Geometry
@@ -195,7 +198,7 @@ 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_133602}
+  note = {Run ID: 20251104_140742}
 }
 ```
 
@@ -210,4 +213,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-11-04 13:51:54*
+*Generated on 2025-11-04 14:16:43*