Update README - Run 20251013_004438

README.md

@@ -12,7 +12,7 @@ datasets:
 metrics:
 - accuracy
 model-index:
-- name: David-fully_shared-weighted_sum
+- name: David-decoupled-deep_efficiency
   results:
   - task:
       type: image-classification
@@ -21,7 +21,7 @@ model-index:
       type: imagenet-1k
     metrics:
     - type: accuracy
-      value: 66.52
+      value: 58.40
 ---
 
 # 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**: small_fast
-- **Sharing Mode**: fully_shared
-- **Fusion Mode**: weighted_sum
-- **Scales**: [256, 512]
+- **Preset**: gated_expert_team
+- **Sharing Mode**: decoupled
+- **Fusion Mode**: deep_efficiency
+- **Scales**: [128, 256, 384, 448, 512, 576, 640, 768, 896]
 - **Feature Dim**: 512
-- **Parameters**: 656,898
+- **Parameters**: 22,133,801
 
 ### Training Configuration
 - **Dataset**: AbstractPhil/imagenet-clip-features-orderly
@@ -55,13 +55,20 @@ exist simultaneously in the same shared space with the correct checks and spacin
 ## Performance
 
 ### Best Results
-- **Validation Accuracy**: 66.52%
-- **Best Epoch**: 9
-- **Final Train Accuracy**: 63.87%
+- **Validation Accuracy**: 58.40%
+- **Best Epoch**: 0
+- **Final Train Accuracy**: 51.66%
 
 ### Per-Scale Performance
-- **Scale 256**: 65.96%
-- **Scale 512**: 66.43%
+- **Scale 128**: 58.40%
+- **Scale 256**: 67.03%
+- **Scale 384**: 69.55%
+- **Scale 448**: 70.34%
+- **Scale 512**: 70.84%
+- **Scale 576**: 71.29%
+- **Scale 640**: 71.60%
+- **Scale 768**: 72.03%
+- **Scale 896**: 72.25%
 
 
 ## Usage
@@ -78,19 +85,19 @@ AbstractPhil/david-shared-space/
 ├── README.md                         # This file
 ├── best_model.json                   # Latest best model info
 ├── weights/
-│   └── david_small_fast/
-│       └── 20251012_235237/
+│   └── david_gated_expert_team/
+│       └── 20251013_004438/
 │           ├── MODEL_SUMMARY.txt     # 🎯 Human-readable performance summary
 │           ├── training_history.json # 📈 Epoch-by-epoch training curve
-│           ├── best_model_acc66.52.safetensors  # ⭐ Accuracy in filename!
-│           ├── best_model_acc66.52_metadata.json
+│           ├── best_model_acc58.40.safetensors  # ⭐ Accuracy in filename!
+│           ├── best_model_acc58.40_metadata.json
 │           ├── final_model.safetensors
 │           ├── checkpoint_epoch_X_accYY.YY.safetensors
 │           ├── david_config.json
 │           └── train_config.json
 └── runs/
-    └── david_small_fast/
-        └── 20251012_235237/
+    └── david_gated_expert_team/
+        └── 20251013_004438/
             └── events.out.tfevents.* # TensorBoard logs
 ```
 
@@ -103,9 +110,9 @@ from huggingface_hub import hf_hub_download
 # Browse available models in MODELS_INDEX.json first!
 
 # Specify model variant and run
-model_name = "david_small_fast"
-run_id = "20251012_235237"
-accuracy = "66.52"  # From MODELS_INDEX.json
+model_name = "david_gated_expert_team"
+run_id = "20251013_004438"
+accuracy = "58.40"  # From MODELS_INDEX.json
 
 # Download config
 config_path = hf_hub_download(
@@ -154,7 +161,7 @@ with torch.no_grad():
 ## Architecture Overview
 
 ### Multi-Scale Processing
-David processes inputs at multiple scales (256, 512), 
+David processes inputs at multiple scales (128, 256, 384, 448, 512, 576, 640, 768, 896), 
 allowing it to capture both coarse and fine-grained features.
 
 ### Shared Representation Space
@@ -175,7 +182,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 ```
 
 ### Fusion Strategy
-**weighted_sum**: Intelligently combines predictions from multiple scales.
+**deep_efficiency**: Intelligently combines predictions from multiple scales.
 
 ## Training Details
 
@@ -188,7 +195,7 @@ score = w_anchor * sim(z, anchor) + w_need * sim(z, need) + ...
 - **Optimizer**: AdamW
 - **Weight Decay**: 1e-05
 - **Scheduler**: cosine_restarts
-- **Gradient Clip**: 15.0
+- **Gradient Clip**: 10.0
 - **Mixed Precision**: False
 
 ## Citation
@@ -199,7 +206,7 @@ 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_235237}
+  note = {Run ID: 20251013_004438}
 }
 ```
 
@@ -214,4 +221,4 @@ Special thanks to Claude (Anthropic) for debugging assistance.
 
 ---
 
-*Generated on 2025-10-13 00:38:47*
+*Generated on 2025-10-13 00:49:36*