modelscope / modelscope/ms-swift
加速reward模型（非LLM）计算与LLM GRPO的高效协同训练
Open
Nobody has claimed this yet.
stale
- Dominant language
- Python
- Stars
- 15.7k
- Forks
- 1.7k
- Avg merge
- 1d 16h
- Merged PRs (30d)
- 136
Description
我有个场景是用Diffusion模型作为reward，导致reward计算相对较慢，且占用额外的显卡。请问是否有方式可以加速这类的reward的计算，因为在计算reward时，训练GRPO的显卡处于等待中。
我已经做的是: 对Diffusion模型本身和Diffusion去噪做了加速，也做成了一个reward server、异步请求reward计算等。
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
No file, test, or entry point is identified. Start by locating the GRPO reward-computation path and reviewing how the existing reward server and asynchronous requests interact with training; define measurable goals for GPU utilization and reward latency before proposing a coordinated solution.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100