Update README - Run 20251012_065325

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: 75.41
+      value: 81.16
 ---
 
 # David: Multi-Scale Crystal Classifier
@@ -32,17 +32,17 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: balanced
+- **Preset**: clip_vit_l14_ultra_deep
 - **Sharing Mode**: partial_shared
-- **Fusion Mode**: hierarchical_tree
-- **Scales**: [256, 512, 768, 1024]
-- **Feature Dim**: 512
+- **Fusion Mode**: deep_efficiency
+- **Scales**: [256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2304, 2560]
+- **Feature Dim**: 768
 - **Parameters**: ~8.8M
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_laion_b32
-- **Epochs**: 20
+- **Model Variant**: clip_vit_l14
+- **Epochs**: 10
 - **Batch Size**: 1024
 - **Learning Rate**: 0.001
 - **Rose Loss Weight**: 0.1 → 0.5
@@ -51,15 +51,12 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 75.41%
-- **Best Epoch**: 9
-- **Final Train Accuracy**: 87.91%
+- **Validation Accuracy**: 81.16%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 78.10%
 
 ### Per-Scale Performance
-- **Scale 256**: 74.79%
-- **Scale 512**: 75.39%
-- **Scale 768**: 75.40%
-- **Scale 1024**: 73.42%
+- **Scale 256**: 81.16%
 
 
 ## Usage
@@ -69,7 +66,7 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ```
 AbstractPhil/gated-david/
 ├── weights/
-│   └── david_balanced/
+│   └── david_clip_vit_l14_ultra_deep/
 │       └── 20251012_065325/
 │           ├── best_model.safetensors
 │           ├── best_model_metadata.json
@@ -78,7 +75,7 @@ AbstractPhil/gated-david/
 │           ├── david_config.json
 │           └── train_config.json
 ├── runs/
-│   └── david_balanced/
+│   └── david_clip_vit_l14_ultra_deep/
 │       └── 20251012_065325/
 │           └── events.out.tfevents.*
 ├── README.md
@@ -92,7 +89,7 @@ from geovocab2.train.model.core.david import David, DavidArchitectureConfig
 from huggingface_hub import hf_hub_download
 
 # Specify model variant and run
-model_name = "david_balanced"
+model_name = "david_clip_vit_l14_ultra_deep"
 run_id = "20251012_065325"
 
 # Download config
@@ -136,7 +133,7 @@ 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 (256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2304, 2560), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Crystal Geometry
@@ -154,7 +151,7 @@ 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
 
@@ -193,4 +190,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 07:35:40*
+*Generated on 2025-10-12 07:38:25*