Evangelinejy's picture
model_680m_0.32B card
d986098 verified
|
Raw
History Blame Contribute Delete
1.32 kB
metadata
license: apache-2.0
tags:
  - chess
  - reinforcement-learning
  - grpo

model_680m_0.32B — 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_0.32B and model_680m_0.32B.

parameters 680m
pretraining tokens 319,289,168 (0.319B)
checkpoints here 5 (steps 100–500)
checkpoints saved by the run 10

Steps

100, 200, 300, 400, 500

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_500"
p = snapshot_download("Pre2Post-Chess-RL/Chess-RL-Models", allow_patterns=f"model_680m_0.32B/{step}/*") + f"/model_680m_0.32B/{step}"
model = AutoModelForCausalLM.from_pretrained(p, trust_remote_code=True)
tok = AutoTokenizer.from_pretrained(p, trust_remote_code=True)