Upload folder using huggingface_hub

README.md

@@ -0,0 +1,66 @@
+---
+license: apache-2.0
+library_name: pytorch
+tags:
+  - chess
+  - transformer
+  - causal-lm
+  - world-model
+datasets:
+  - random-self-play
+model-index:
+  - name: pawn-large
+    results:
+      - task:
+          type: next-move-prediction
+        metrics:
+          - name: Val Loss
+            type: loss
+            value: 3.091
+          - name: Val Accuracy
+            type: accuracy
+            value: 7.1
+---
+
+# PAWN-LARGE
+
+A causal transformer trained on random chess games, designed as a testbed for finetuning and augmentation methods at small scales.
+
+## Model Details
+
+| | |
+|---|---|
+| **Parameters** | 68.4M |
+| **Architecture** | Decoder-only transformer (RMSNorm, SwiGLU, RoPE) |
+| **d_model** | 640 |
+| **Layers** | 10 |
+| **Heads** | 8 |
+| **Vocabulary** | 4,278 tokens (4,096 grid + 176 promotions + 5 outcomes + 1 PAD) |
+| **Sequence length** | 256 |
+| **Training steps** | 85K/100K |
+| **Best val loss** | 3.091 (step 85,000) |
+| **Best val accuracy** | 7.1% |
+
+## Usage
+
+```python
+import torch
+from pawn.config import CLMConfig
+from pawn.model import PAWNCLM
+
+cfg = CLMConfig.large()
+model = PAWNCLM(cfg)
+
+ckpt = torch.load("model.pt", map_location="cpu", weights_only=False)
+model.load_state_dict(ckpt["model_state_dict"])
+model.eval()
+```
+
+## Training
+
+Trained from scratch on random self-play games generated by a Rust chess engine (shakmaty).
+See the [PAWN repository](https://github.com/thomas-schweich/PAWN) for training code, data pipeline, and evaluation suite.
+
+## License
+
+Apache 2.0

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:891707422c0e63a3bfc91080cdfeae34033f9e3647c3c673b366786061706782
+size 820643835