🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
Commits on master
Merge pull request #840 from JustinGuese/readme-benchmark-resolution
4d7a74a jingyaogong committed 2 months ago
[refactor] remove redundant rollout update_policy after torch.compile
0c91dea TOPSAIL\78190 committed 2 months ago
Merge pull request #814 from FloatFrank/fix/ppo-logp-autocast-align-v3
e9f4650 jingyaogong committed 4 months ago
[fix] PPO: align mb_resp_logp with old_resp_logp under autocast
1191fc8 FloatFrank committed 4 months ago
fix: guard against empty/filtered LLM responses in chat_api.py
182bff1 Jonathan Hill committed 6 months ago
Merge pull request #759 from TKiteRunner/fix/grpo-oom-reorder-rewards
e73a407 jingyaogong committed 6 months ago
fix: resolve CUDA OOM in train_grpo.py on GPUs with <=8GB VRAM
1077641 TKiteRunner committed 6 months ago