注意力头维度为32会报错，有什么办法可以支持注意力维度为32的吗
Open
- Dominant language
- Python
- Stars
- 1.3k
- Forks
- 275
- Avg merge
- 3d 17h
- Merged PRs (30d)
- 33
Description
模型框架是基于千问3-0.6b的，修改了部分结构，其中注意力头维度为32会报错，如果修改Warp Group参数，后面有很多问题，有什么办法可以支持注意力32维度的吗
Contributor guide
No contributing guide indexed for this repository
Research direction
Start by reproducing the error with the modified Qwen3-0.6B-based model using an attention head dimension of 32, then inspect the Warp Group parameter handling. Done means dimension 32 is supported without requiring problematic Warp Group changes or causing the reported downstream issues.
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