metadata
license: apache-2.0
tags:
- chess
- reinforcement-learning
- grpo
model_680m_6.4B — 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_6.4B and
model_680m_6.4B.
| parameters | 680m |
| pretraining tokens | 6,397,966,000 (6.4B) |
| checkpoints here | 10 (steps 100–1000) |
| checkpoints saved by the run | 20 |
Steps
100, 200, 300, 400, 500, 600, 700, 800, 900, 1000
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:
from huggingface_hub import snapshot_download
from transformers import AutoModelForCausalLM, AutoTokenizer
step = "global_step_1000"
p = snapshot_download("Pre2Post-Chess-RL/Chess-RL-Models", allow_patterns=f"model_680m_6.4B/{step}/*") + f"/model_680m_6.4B/{step}"
model = AutoModelForCausalLM.from_pretrained(p, trust_remote_code=True)
tok = AutoTokenizer.from_pretrained(p, trust_remote_code=True)