Update README - Run 20251012_231445

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-decoupled-deep_efficiency
+- name: David-fully_shared-weighted_sum
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 66.84
+      value: 63.04
 ---
 
 # David: Multi-Scale Feature Classifier
@@ -36,12 +36,12 @@ exist simultaneously in the same shared space with the correct checks and spacin
 ## Model Details
 
 ### Architecture
-- **Preset**: high_accuracy
-- **Sharing Mode**: decoupled
-- **Fusion Mode**: deep_efficiency
-- **Scales**: [256, 512, 768, 1024, 1280]
+- **Preset**: small_fast
+- **Sharing Mode**: fully_shared
+- **Fusion Mode**: weighted_sum
+- **Scales**: [256, 512]
 - **Feature Dim**: 512
-- **Parameters**: 14,877,593
+- **Parameters**: 656,898
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
@@ -49,22 +49,19 @@ exist simultaneously in the same shared space with the correct checks and spacin
 - **Epochs**: 10
 - **Batch Size**: 1024
 - **Learning Rate**: 0.01
-- **Rose Loss Weight**: 0.2 → 0.8
+- **Rose Loss Weight**: 0.2 → 0.6
 - **Cayley Loss**: False
 
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 66.84%
-- **Best Epoch**: 9
-- **Final Train Accuracy**: 66.12%
+- **Validation Accuracy**: 63.04%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 54.92%
 
 ### Per-Scale Performance
-- **Scale 256**: 66.84%
-- **Scale 512**: 72.72%
-- **Scale 768**: 74.34%
-- **Scale 1024**: 75.09%
-- **Scale 1280**: 75.37%
+- **Scale 256**: 62.12%
+- **Scale 512**: 62.97%
 
 
 ## Usage
@@ -81,19 +78,19 @@ AbstractPhil/david-shared-space/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_high_accuracy/
-│       └── 20251012_221046/
+│   └── david_small_fast/
+│       └── 20251012_231445/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc66.84.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc66.84_metadata.json
+│           ├── best_model_acc63.04.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc63.04_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_high_accuracy/
-        └── 20251012_221046/
+    └── david_small_fast/
+        └── 20251012_231445/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -106,9 +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_high_accuracy"
-run_id = "20251012_221046"
-accuracy = "66.84"  # From MODELS_INDEX.json
+model_name = "david_small_fast"
+run_id = "20251012_231445"
+accuracy = "63.04"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -157,7 +154,7 @@ 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), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Shared Representation Space
@@ -178,20 +175,20 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 ```
 
 ### Fusion Strategy
-**deep_efficiency**: Intelligently combines predictions from multiple scales.
+**weighted_sum**: Intelligently combines predictions from multiple scales.
 
 ## Training Details
 
 ### Loss Components
 - **Cross-Entropy**: Standard classification loss
-- **Rose Loss**: Pentachora role-weighted margin loss (weight: 0.2→0.8)
+- **Rose Loss**: Pentachora role-weighted margin loss (weight: 0.2→0.6)
 - **Cayley Loss**: Geometric regularization (disabled)
 
 ### Optimization
 - **Optimizer**: AdamW
 - **Weight Decay**: 1e-05
 - **Scheduler**: cosine_restarts
-- **Gradient Clip**: 10.0
+- **Gradient Clip**: 5.0
 - **Mixed Precision**: False
 
 ## Citation
@@ -202,7 +199,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
   author = {AbstractPhil},
   year = {2025},
   url = {https://huggingface.co/AbstractPhil/david-shared-space},
-  note = {Run ID: 20251012_221046}
+  note = {Run ID: 20251012_231445}
 }
 ```
 
@@ -217,4 +214,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 22:58:07*
+*Generated on 2025-10-12 23:19:29*