Logo
Add
M
M
mail.dark4scope@gmail.com User ID: 1
Settings Admin
Sign out
07 · DAPO — Open-Source LLM RL System at Scale
Edit
Share Download Star Archive
Collection
Default
Edit
Tags
#bytedance #grpo #post-training #rlvr #verl
Edit
Description
GRPO 规模化会崩，这里给 4 个修法：clip-higher/动态采样/token-level loss/超长奖励整形。跑 verl 时读
Edit
Notes
no notes available
Edit
Views
8
Progress
6% - Page 1 of 16
Date added
Aug. 14, 2026, 9:33 a.m.
Confirm Cancel Delete