modelscope / modelscope/ms-swift
支持任意 ViT Encoder 与任意 LLM 自由组合构建多模态模型
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 15.7k
- Forks
- 1.7k
- Avg merge
- 1d 16h
- Merged PRs (30d)
- 136
Description
Checklist / 检查清单
- I have searched existing issues, and this is a new feature request. / 我已经搜索过现有的 issues,确认这是一个新的 Feature Request。
Feature Request Description / Feature Request 描述
冒昧提一个需求建议:不知 ms-swift 能否参考 LoongForge 的设计,支持通过配置文件或命令行参数,将代码库中已支持的任意 ViT 编码器(如 CLIP-ViT、Qwen-ViT、InternViT 等)与任意 LLM(如 Qwen、LLaMA、InternLM 等)自由组合成多模态模型并直接训练,同时仍能对 ViT、Aligner、LLM 各自独立控制冻存与学习率?目前虽有 MultiModelKeys 注册机制,但每新增组合仍需手写适配代码,若能做到配置驱动的即插即用,对我们做实验对比会非常方便。当然,这只是个人的一点想法,不一定符合项目的设计方向,如果团队有别的考量也完全理解,还是非常感谢 ms-swift 已经提供的强大支持!
Pull Request / Pull Request 信息
No response
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
Start by reviewing the existing MultiModelKeys registration mechanism and the current model-combination adapters mentioned in the issue. Define how configuration or CLI parameters would compose supported ViT encoders with LLMs and independently control ViT, aligner, and LLM freezing and learning rates; done means arbitrary supported combinations can train without handwritten adapters.
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
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100