modelscope / modelscope/ms-swift
Bidirectional mask for embedding training w/ megatron
Open
Nobody has claimed this yet.
question
stale
- 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 question or discussion topic. / 我已经搜索过现有的 issues，确认这是一个新的问题与讨论。
Question Description / 问题描述
I want to finetune Qwen/Qwen3-embedding-8B using megatron backend, but i want to modify forward to use bidirectional attention mask, is there a way to do it?
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 locating the Megatron backend forward path and the attention-mask handling used when finetuning Qwen/Qwen3-embedding-8B. Determine whether bidirectional attention can be configured there; done means the requested embedding training setup supports this mask and its behavior is verified.
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