Update README - Run 20251012_065325

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-partial_shared-deep_efficiency
+- name: David-decoupled-deep_efficiency
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 83.04
+      value: 69.49
 ---
 
 # 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_deep
-- **Sharing Mode**: partial_shared
+- **Preset**: high_accuracy
+- **Sharing Mode**: decoupled
 - **Fusion Mode**: deep_efficiency
-- **Scales**: [256, 512, 768, 1024, 1280, 1536, 1792, 2048, 2304, 2560]
-- **Feature Dim**: 768
+- **Scales**: [256, 512, 768, 1024, 1280]
+- **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
@@ -51,21 +51,12 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 83.04%
-- **Best Epoch**: 9
-- **Final Train Accuracy**: 91.00%
+- **Validation Accuracy**: 69.49%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 65.85%
 
 ### Per-Scale Performance
-- **Scale 256**: 83.04%
-- **Scale 512**: 83.12%
-- **Scale 768**: 83.20%
-- **Scale 1024**: 83.21%
-- **Scale 1280**: 83.25%
-- **Scale 1536**: 83.13%
-- **Scale 1792**: 83.15%
-- **Scale 2048**: 83.14%
-- **Scale 2304**: 82.96%
-- **Scale 2560**: 82.71%
+- **Scale 256**: 69.49%
 
 
 ## Usage
@@ -75,17 +66,20 @@ as class prototypes with role-weighted similarity computation (Rose Loss).
 ```
 AbstractPhil/gated-david/
 ├── weights/
-│   ├── best_model.pth              # Best model weights (PyTorch)
-│   ├── best_model.safetensors      # Best model weights (SafeTensors)
-│   ├── best_model_metadata.json    # Training metadata
-│   ├── final_model.pth             # Final epoch weights
-│   ├── final_model.safetensors
-│   ├── david_config.json           # Model architecture config
-│   └── train_config.json           # Training configuration
+│   └── david_high_accuracy/
+│       └── 20251012_065325/
+│           ├── best_model.safetensors
+│           ├── best_model_metadata.json
+│           ├── final_model.safetensors
+│           ├── checkpoint_epoch_X.safetensors
+│           ├── david_config.json
+│           └── train_config.json
 ├── runs/
-│   └── events.out.tfevents.*       # TensorBoard logs
-├── README.md                        # This file
-└── best_model.json                 # Performance summary
+│   └── david_high_accuracy/
+│       └── 20251012_065325/
+│           └── events.out.tfevents.*
+├── README.md
+└── best_model.json
 ```
 
 ### Loading the Model
@@ -94,19 +88,27 @@ AbstractPhil/gated-david/
 from geovocab2.train.model.core.david import David, DavidArchitectureConfig
 from huggingface_hub import hf_hub_download
 
+# Specify model variant and run
+model_name = "david_high_accuracy"
+run_id = "20251012_065325"
+
 # Download config
-config_path = hf_hub_download(repo_id="AbstractPhil/gated-david", 
-                               filename="weights/david_config.json")
+config_path = hf_hub_download(
+    repo_id="AbstractPhil/gated-david", 
+    filename=f"weights/{model_name}/{run_id}/david_config.json"
+)
 config = DavidArchitectureConfig.from_json(config_path)
 
 # Download weights
-weights_path = hf_hub_download(repo_id="AbstractPhil/gated-david", 
-                                filename="weights/best_model.pth")
+weights_path = hf_hub_download(
+    repo_id="AbstractPhil/gated-david", 
+    filename=f"weights/{model_name}/{run_id}/best_model.safetensors"
+)
 
-# Initialize model
+# Load model
+from safetensors.torch import load_file
 david = David.from_config(config)
-checkpoint = torch.load(weights_path)
-david.load_state_dict(checkpoint['model_state_dict'])
+david.load_state_dict(load_file(weights_path))
 david.eval()
 ```
 
@@ -131,7 +133,7 @@ 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, 768, 1024, 1280), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Crystal Geometry
@@ -162,7 +164,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 - **Optimizer**: AdamW
 - **Weight Decay**: 1e-05
 - **Scheduler**: cosine_restarts
-- **Gradient Clip**: 5.0
+- **Gradient Clip**: 10.0
 - **Mixed Precision**: False
 
 ## Citation
@@ -173,7 +175,7 @@ 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_060013}
+  note = {Run ID: 20251012_065325}
 }
 ```
 
@@ -188,4 +190,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-12 06:40:54*
+*Generated on 2025-10-12 06:57:45*