Update README - Run 20251104_133602

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-decoupled-cantor_scale
+- name: David-partial_shared-cantor_scale
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 78.87
+      value: 81.28
 ---
 
 # 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_b16_cantor_decoupled_massive
-- **Sharing Mode**: decoupled
+- **Preset**: clip_vit_l14_cantor
+- **Sharing Mode**: partial_shared
 - **Fusion Mode**: cantor_scale
-- **Scales**: [512, 4096, 8192, 16384]
-- **Feature Dim**: 512
-- **Parameters**: 751,239,941
+- **Scales**: [384, 768, 1024, 1280, 1536]
+- **Feature Dim**: 768
+- **Parameters**: 32,436,998
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_b16
+- **Model Variant**: clip_vit_l14
 - **Epochs**: 5
-- **Batch Size**: 256
+- **Batch Size**: 1024
 - **Learning Rate**: 0.001
 - **Rose Loss Weight**: 0.1 → 0.5
 - **Cayley Loss**: False
@@ -51,15 +51,16 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 78.87%
-- **Best Epoch**: 4
-- **Final Train Accuracy**: 87.75%
+- **Validation Accuracy**: 81.28%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 77.89%
 
 ### Per-Scale Performance
-- **Scale 512**: 77.27%
-- **Scale 4096**: 78.43%
-- **Scale 8192**: 78.03%
-- **Scale 16384**: 78.34%
+- **Scale 384**: 80.39%
+- **Scale 768**: 80.90%
+- **Scale 1024**: 80.75%
+- **Scale 1280**: 80.57%
+- **Scale 1536**: 80.98%
 
 
 ## Usage
@@ -76,19 +77,19 @@ AbstractPhil/gated-david/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_clip_vit_b16_cantor_decoupled_massive/
-│       └── 20251104_125328/
+│   └── david_clip_vit_l14_cantor/
+│       └── 20251104_133602/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc78.87.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc78.87_metadata.json
+│           ├── best_model_acc81.28.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc81.28_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_clip_vit_b16_cantor_decoupled_massive/
-        └── 20251104_125328/
+    └── david_clip_vit_l14_cantor/
+        └── 20251104_133602/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -101,9 +102,9 @@ 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_decoupled_massive"
-run_id = "20251104_125328"
-accuracy = "78.87"  # From MODELS_INDEX.json
+model_name = "david_clip_vit_l14_cantor"
+run_id = "20251104_133602"
+accuracy = "81.28"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -152,7 +153,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (512, 4096, 8192, 16384),
+David processes inputs at multiple scales (384, 768, 1024, 1280, 1536),
 allowing it to capture both coarse and fine-grained features.
 
 ### Feature Geometry
@@ -194,7 +195,7 @@ 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_125328}
+  note = {Run ID: 20251104_133602}
 }
 ```
 
@@ -209,4 +210,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-11-04 13:32:59*
+*Generated on 2025-11-04 13:41:01*