SunJin Kim

icon location Seoul icon location let's code

Results 1 comments of SunJin Kim

Add asynchronous rollout + reward stage to PPOTrainer

Has been implemented in https://github.com/agentica-project/verl-pipeline, which was used to make the the [DeepCoder-14B](https://pretty-radio-b75.notion.site/DeepCoder-A-Fully-Open-Source-14B-Coder-at-O3-mini-Level-1cf81902c14680b3bee5eb349a512a51) model. They saw 2.5x speedup in code RL training. ![Image](https://github.com/user-attachments/assets/95777ac8-4e0a-4ec6-820d-9f08b78f8472)