Update README - Run 20251012_161107

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-partial_shared-hierarchical_tree
+- name: David-partial_shared-deep_efficiency
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 75.53
+      value: 83.71
 ---
 
 # David: Multi-Scale Crystal Classifier
@@ -32,16 +32,16 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: balanced
+- **Preset**: clip_vit_bigg14
 - **Sharing Mode**: partial_shared
-- **Fusion Mode**: hierarchical_tree
-- **Scales**: [256, 512, 768, 1024]
-- **Feature Dim**: 512
-- **Parameters**: 8,758,271
+- **Fusion Mode**: deep_efficiency
+- **Scales**: [384, 512, 768, 1024, 1280, 1536, 1792, 2048]
+- **Feature Dim**: 1280
+- **Parameters**: 60,814,373
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_laion_b32
+- **Model Variant**: clip_vit_laion_bigg14
 - **Epochs**: 10
 - **Batch Size**: 1024
 - **Learning Rate**: 0.001
@@ -51,15 +51,19 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 75.53%
-- **Best Epoch**: 8
-- **Final Train Accuracy**: 89.72%
+- **Validation Accuracy**: 83.71%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 81.03%
 
 ### Per-Scale Performance
-- **Scale 256**: 74.36%
-- **Scale 512**: 75.26%
-- **Scale 768**: 74.77%
-- **Scale 1024**: 74.70%
+- **Scale 384**: 83.52%
+- **Scale 512**: 83.54%
+- **Scale 768**: 83.71%
+- **Scale 1024**: 83.48%
+- **Scale 1280**: 83.63%
+- **Scale 1536**: 83.59%
+- **Scale 1792**: 83.62%
+- **Scale 2048**: 83.66%
 
 
 ## Usage
@@ -76,19 +80,19 @@ AbstractPhil/gated-david/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_balanced/
-│       └── 20251012_152245/
+│   └── david_clip_vit_bigg14/
+│       └── 20251012_161107/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc75.53.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc75.53_metadata.json
+│           ├── best_model_acc83.71.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc83.71_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_balanced/
-        └── 20251012_152245/
+    └── david_clip_vit_bigg14/
+        └── 20251012_161107/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -101,9 +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_balanced"
-run_id = "20251012_152245"
-accuracy = "75.53"  # From MODELS_INDEX.json
+model_name = "david_clip_vit_bigg14"
+run_id = "20251012_161107"
+accuracy = "83.71"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -152,7 +156,7 @@ 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 (384, 512, 768, 1024, 1280, 1536, 1792, 2048), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Crystal Geometry
@@ -170,7 +174,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 ```
 
 ### Fusion Strategy
-**hierarchical_tree**: Intelligently combines predictions from multiple scales.
+**deep_efficiency**: Intelligently combines predictions from multiple scales.
 
 ## Training Details
 
@@ -184,7 +188,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 - **Weight Decay**: 1e-05
 - **Scheduler**: cosine_restarts
 - **Gradient Clip**: 10.0
-- **Mixed Precision**: True
+- **Mixed Precision**: False
 
 ## Citation
 
@@ -194,7 +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: 20251012_152245}
+  note = {Run ID: 20251012_161107}
 }
 ```
 
@@ -209,4 +213,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 15:40:39*
+*Generated on 2025-10-12 16:14:46*