modelscope / modelscope/ms-swift

Question about New model and algorithms support

Open
#6,694 1 comment 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

stale
Dominant language
Python
Stars
15.7k
Forks
1.7k
Avg merge
1d 16h
Merged PRs (30d)
136

Description

I want to adapt my RL algorithm, specifically designed for DLLM (Diffusion Language Model), using the SDAR-8B-Chat model and its SFT and RL-trained versions. Currently, this model is compatible with the Lmdeploy inference engine, which can be used as a rollout tool. However, I don't know how to adapt it to ms-swift, because for RL algorithms like GRPO, it requires redesigning the attention mechanism and forward process, meaning I need to manually modify the intermediate training details. I want to adapt it to ms-swift because your framework supports lmdeploy rollout. How can I customize this RL training process?

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing ms-swift's GRPO training path and its lmdeploy rollout integration, then identify how the SDAR-8B-Chat model's custom attention and forward process could fit. Done would be a documented customization path or a clear determination that this model requires new framework support.

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
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.