Update README - Run 20251012_132646

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-partial_shared-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: 83.10
+      value: 68.12
 ---
 
 # David: Multi-Scale Crystal Classifier
@@ -32,17 +32,17 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Model Details
 
 ### Architecture
-- **Preset**: clip_vit_l14_ultra_deep
-- **Sharing Mode**: partial_shared
-- **Fusion Mode**: deep_efficiency
-- **Scales**: [256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2304, 2560]
-- **Feature Dim**: 768
+- **Preset**: small_fast
+- **Sharing Mode**: fully_shared
+- **Fusion Mode**: weighted_sum
+- **Scales**: [256, 512]
+- **Feature Dim**: 512
 - **Parameters**: ~8.8M
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
-- **Model Variant**: clip_vit_l14
-- **Epochs**: 10
+- **Model Variant**: clip_vit_laion_b32
+- **Epochs**: 20
 - **Batch Size**: 1024
 - **Learning Rate**: 0.001
 - **Rose Loss Weight**: 0.1 → 0.5
@@ -51,20 +51,13 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 83.10%
-- **Best Epoch**: 8
-- **Final Train Accuracy**: 94.52%
+- **Validation Accuracy**: 68.12%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 63.38%
 
 ### Per-Scale Performance
-- **Scale 256**: 83.10%
-- **Scale 512**: 83.24%
-- **Scale 768**: 83.06%
-- **Scale 1024**: 83.03%
-- **Scale 1280**: 82.92%
-- **Scale 1536**: 82.95%
-- **Scale 1792**: 82.98%
-- **Scale 2048**: 82.97%
-- **Scale 2304**: 82.89%
+- **Scale 256**: 67.00%
+- **Scale 512**: 67.96%
 
 
 ## Usage
@@ -74,8 +67,8 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ```
 AbstractPhil/gated-david/
 ├── weights/
-│   └── david_clip_vit_l14_ultra_deep/
-│       └── 20251012_065325/
+│   └── david_small_fast/
+│       └── 20251012_132646/
 │           ├── best_model.safetensors
 │           ├── best_model_metadata.json
 │           ├── final_model.safetensors
@@ -83,8 +76,8 @@ AbstractPhil/gated-david/
 │           ├── david_config.json
 │           └── train_config.json
 ├── runs/
-│   └── david_clip_vit_l14_ultra_deep/
-│       └── 20251012_065325/
+│   └── david_small_fast/
+│       └── 20251012_132646/
 │           └── events.out.tfevents.*
 ├── README.md
 └── best_model.json
@@ -97,8 +90,8 @@ from geovocab2.train.model.core.david import David, DavidArchitectureConfig
 from huggingface_hub import hf_hub_download
 
 # Specify model variant and run
-model_name = "david_clip_vit_l14_ultra_deep"
-run_id = "20251012_065325"
+model_name = "david_small_fast"
+run_id = "20251012_132646"
 
 # Download config
 config_path = hf_hub_download(
@@ -141,7 +134,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2304, 2560), 
+David processes inputs at multiple scales (256, 512), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Crystal Geometry
@@ -159,7 +152,7 @@ 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
 
@@ -183,7 +176,7 @@ 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_065325}
+  note = {Run ID: 20251012_132646}
 }
 ```
 
@@ -198,4 +191,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 08:02:12*
+*Generated on 2025-10-12 13:30:05*