Update README - Run 20251104_152832

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-decoupled-deep_efficiency
+- name: David-decoupled-cantor_scale
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 75.77
+      value: 73.93
 ---
 
 # David: Multi-Scale Feature Classifier
@@ -32,12 +32,12 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: high_accuracy
+- **Preset**: clip_vit_b16_cantor
 - **Sharing Mode**: decoupled
-- **Fusion Mode**: deep_efficiency
-- **Scales**: [256, 512, 768, 1024, 1280]
+- **Fusion Mode**: cantor_scale
+- **Scales**: [256, 512, 768, 1024]
 - **Feature Dim**: 512
-- **Parameters**: 14,877,593
+- **Parameters**: 9,057,029
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
@@ -51,16 +51,15 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 75.77%
-- **Best Epoch**: 4
-- **Final Train Accuracy**: 81.50%
+- **Validation Accuracy**: 73.93%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 72.56%
 
 ### Per-Scale Performance
-- **Scale 256**: 72.68%
-- **Scale 512**: 75.10%
-- **Scale 768**: 75.77%
-- **Scale 1024**: 76.07%
-- **Scale 1280**: 76.00%
+- **Scale 256**: 68.30%
+- **Scale 512**: 72.27%
+- **Scale 768**: 72.71%
+- **Scale 1024**: 73.27%
 
 
 ## Usage
@@ -77,19 +76,19 @@ AbstractPhil/gated-david/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_high_accuracy/
-│       └── 20251104_151233/
+│   └── david_clip_vit_b16_cantor/
+│       └── 20251104_152832/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc75.77.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc75.77_metadata.json
+│           ├── best_model_acc73.93.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc73.93_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_high_accuracy/
-        └── 20251104_151233/
+    └── david_clip_vit_b16_cantor/
+        └── 20251104_152832/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -102,9 +101,9 @@ from huggingface_hub import hf_hub_download
 # Browse available models in MODELS_INDEX.json first!
 
 # Specify model variant and run
-model_name = "david_high_accuracy"
-run_id = "20251104_151233"
-accuracy = "75.77"  # From MODELS_INDEX.json
+model_name = "david_clip_vit_b16_cantor"
+run_id = "20251104_152832"
+accuracy = "73.93"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -153,7 +152,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (256, 512, 768, 1024, 1280),
+David processes inputs at multiple scales (256, 512, 768, 1024),
 allowing it to capture both coarse and fine-grained features.
 
 ### Feature Geometry
@@ -171,7 +170,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 ```
 
 ### Fusion Strategy
-**deep_efficiency**: Intelligently combines predictions from multiple scales.
+**cantor_scale**: Intelligently combines predictions from multiple scales.
 
 ## Training Details
 
@@ -195,7 +194,7 @@ 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_151233}
+  note = {Run ID: 20251104_152832}
 }
 ```
 
@@ -210,4 +209,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-11-04 15:26:20*
+*Generated on 2025-11-04 15:30:39*