model_680m_1.6B card

model_680m_1.6B/README.md

@@ -0,0 +1,42 @@
+---
+license: apache-2.0
+tags: [chess, reinforcement-learning, grpo]
+---
+
+# model_680m_1.6B — RL (GRPO) checkpoints
+
+RL post-training trajectory for the chess pre-to-post compute-allocation study.
+The pretraining base and SFT init for this model are
+[`model_680m_1.6B`](https://huggingface.co/pavelslab-nyu/Chess-Pretrain-Models/tree/main/model_680m_1.6B) and
+[`model_680m_1.6B`](https://huggingface.co/pavelslab-nyu/Chess-SFT-Models/tree/main/model_680m_1.6B).
+
+| | |
+|---|---|
+| run id | `C6p5e18_680m_alpha1.000_beta0.296` |
+| parameters | 680m |
+| pretraining tokens | 1,596,445,842 (1.6B) |
+| compute class | 6p5e18 |
+| alpha (pretrain fraction) | 1.0 |
+| beta (SFT fraction) | 0.296 |
+| checkpoints here | 1 (steps 2000–2000) |
+| checkpoints saved by the run | 3 |
+
+## Steps
+
+`2000`
+
+## Loading
+
+Each `global_step_N/` folder is self-contained. The models use a custom
+tokenizer (`tokenizer.py`), and the remote-code resolver ignores `subfolder=`,
+so download the folder first and load the **local path**:
+
+```python
+from huggingface_hub import snapshot_download
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+step = "global_step_2000"
+p = snapshot_download("Pre2Post-Chess-RL/Chess-RL-Models", allow_patterns=f"model_680m_1.6B/{step}/*") + f"/model_680m_1.6B/{step}"
+model = AutoModelForCausalLM.from_pretrained(p, trust_remote_code=True)
+tok = AutoTokenizer.from_pretrained(p, trust_remote_code=True)
+```