Update main README

README.md

@@ -9,132 +9,50 @@ tags:
   - cantor-routing
   - pentachoron
   - multi-scale
-datasets:
-  - cifar100
-metrics:
-  - accuracy
-model-index:
-  - name: DavidBeans
-    results:
-      - task:
-          type: image-classification
-          name: Image Classification
-        dataset:
-          name: CIFAR-100
-          type: cifar100
-        metrics:
-          - type: accuracy
-            value: 70.05
-            name: Top-1 Accuracy
 ---
 
-# At long last my goal of 70% accuracy cifar100 on a geometric vit barrier breached
+# 🫘💎 DavidBeans: Unified Vision-to-Crystal Architecture
 
-David's primary metrics were based on clip-vit (the original gated-david repo), and now geovit-david-beans can exist as a stand-in for a vit!
+This repository contains training runs for DavidBeans - a unified geometric deep learning architecture combining:
 
-This is my first legitimate example of a geovit that I can attest is nearly to the expectations of geometric encoding in a useful fashion.
+- **BEANS (ViT Backbone)**: Cantor-routed sparse attention
+- **DAVID (Classifier)**: Multi-scale crystal projection with Cayley-Menger geometric regularization
 
-ITS NOT THERE YET, but it's a definite benchmark I've been trying to achieve for quite some time, so I'm glad this one finally worked.
-
-# 💎 DavidBeans: Unified Vision-to-Crystal Architecture
-
-DavidBeans combines **ViT-Beans** (Cantor-routed sparse attention) with **David** (multi-scale crystal classification) into a unified geometric deep learning architecture.
-
-## Model Description
-
-This model implements several novel techniques:
-
-- **Hybrid Cantor Routing**: Combines fractal Cantor set distances with positional proximity for sparse attention patterns
-- **Pentachoron Experts**: 5-vertex simplex structure with Cayley-Menger geometric regularization
-- **Multi-Scale Crystal Projection**: Projects features to multiple representation scales with learned fusion
-- **Cross-Contrastive Learning**: Aligns patch-level features with crystal anchors
-
-## Architecture
+## Repository Structure
 
 ```
-Image [B, 3, 32, 32]
-       │
-       ▼
-┌─────────────────────────────────────────┐
-│  BEANS BACKBONE                         │
-│  ├─ Patch Embed → [64 patches, 512d]
-│  ├─ Hybrid Cantor Router (α=0.3)
-│  ├─ 8 × Attention Blocks (8 heads)
-│  └─ 8 × Pentachoron Expert Layers
-└─────────────────────────────────────────┘
-       │
-       ▼
-┌─────────────────────────────────────────┐
-│  DAVID HEAD                             │
-│  ├─ Multi-scale projection: [256, 512, 768]
-│  ├─ Per-scale Crystal Heads
-│  └─ Geometric Fusion (learned weights)
-└─────────────────────────────────────────┘
-       │
-       ▼
-    [100 classes]
-```
-
-## Training Details
-
-| Parameter | Value |
-|-----------|-------|
-| Dataset | CIFAR-100 |
-| Classes | 100 |
-| Image Size | 32×32 |
-| Patch Size | 4×4 |
-| Embedding Dim | 512 |
-| Layers | 8 |
-| Attention Heads | 8 |
-| Experts | 5 (pentachoron) |
-| Sparse Neighbors | k=32 |
-| Scales | [256, 512, 768] |
-| Epochs | 200 |
-| Batch Size | 128 |
-| Learning Rate | 0.0005 |
-| Weight Decay | 0.1 |
-| Mixup α | 0.3 |
-| CutMix α | 1.0 |
-| Label Smoothing | 0.1 |
-
-## Results
-
-| Metric | Value |
-|--------|-------|
-| **Top-1 Accuracy** | **70.05%** |
-
-## TensorBoard Logs
-
-Training logs are included in the `tensorboard/` directory. To view:
-
-```bash
-tensorboard --logdir tensorboard/
+AbstractPhil/geovit-david-beans/
+├── README.md (this file)
+└── weights/
+    ├── run_001_baseline_YYYYMMDD_HHMMSS/
+    │   ├── best.safetensors
+    │   ├── epoch_010.safetensors
+    │   ├── config.json
+    │   ├── training_config.json
+    │   └── tensorboard/
+    ├── run_002_5expert_5scale_YYYYMMDD_HHMMSS/
+    │   └── ...
+    └── ...
 ```
 
 ## Usage
 
 ```python
-import torch
 from safetensors.torch import load_file
 from david_beans import DavidBeans, DavidBeansConfig
+import json
+
+# Pick a run
+run_path = "weights/run_002_5expert_5scale_20251129_171229"
 
 # Load config
-config = DavidBeansConfig(
-    image_size=32,
-    patch_size=4,
-    dim=512,
-    num_layers=8,
-    num_heads=8,
-    num_experts=5,
-    k_neighbors=32,
-    cantor_weight=0.3,
-    scales=[256, 512, 768],
-    num_classes=100
-)
-
-# Create model and load weights
+with open(f"{run_path}/config.json") as f:
+    config_dict = json.load(f)
+config = DavidBeansConfig(**config_dict)
+
+# Load model
 model = DavidBeans(config)
-state_dict = load_file("model.safetensors")
+state_dict = load_file(f"{run_path}/best.safetensors")
 model.load_state_dict(state_dict)
 
 # Inference
@@ -144,16 +62,37 @@ with torch.no_grad():
     predictions = output['logits'].argmax(dim=-1)
 ```
 
-## Citation
+## Training Runs
+
+| Run | Name | Accuracy | Notes |
+|-----|------|----------|-------|
+| 001 | baseline | 70.05% | Initial CIFAR-100 run |
+| 002 | 5expert_5scale | 68.34% | 5 experts, 5 scales |
+
+## Architecture
 
-```bibtex
-@misc{davidbeans2025,
-  author = {AbstractPhil},
-  title = {DavidBeans: Unified Vision-to-Crystal Architecture},
-  year = {2025},
-  publisher = {HuggingFace},
-  url = {https://huggingface.co/AbstractPhil/geovit-david-beans}
-}
+```
+Image [B, 3, 32, 32]
+       │
+       ▼
+┌─────────────────────────────────────────┐
+│  BEANS BACKBONE                         │
+│  ├─ Patch Embed → [64 patches, dim]     │
+│  ├─ Hybrid Cantor Router                │
+│  ├─ N × Attention Blocks                │
+│  └─ N × Pentachoron Expert Layers       │
+└─────────────────────────────────────────┘
+       │
+       ▼
+┌─────────────────────────────────────────┐
+│  DAVID HEAD                             │
+│  ├─ Multi-scale projection              │
+│  ├─ Per-scale Crystal Heads             │
+│  └─ Geometric Fusion                    │
+└─────────────────────────────────────────┘
+       │
+       ▼
+    [num_classes]
 ```
 
 ## License