Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Feature] mask_strategy可以改成可学参数吗？ #219

Open
wangshankun opened this issue Feb 26, 2025 · 5 comments
Open

[Feature] mask_strategy可以改成可学参数吗？ #219

wangshankun opened this issue Feb 26, 2025 · 5 comments

Comments

@wangshankun
Copy link

Motivation

mask_strategy可以作为一层可学的参数吗？类似MOBA中的gate，在finetune或者distill里面直接学习进去，这样工程适配工作就少了很多

Related resources

No response
@BrianChen1129
Copy link
Collaborator

That is a great idea! But we haven't tried it.

@foreverpiano
Copy link
Collaborator

Yes, a gated sparse pattern would be cool and easy to adapt with.

@wangshankun
Copy link
Author

That is a great idea! But we haven't tried it.
@BrianChen1129
我想试试，但看起来sta算子实现没有backward，应该可以用FlexAttention先做Finetune吧。
我想请教一下FlexAttention和现在STA算子执行效率有多少差异

@foreverpiano
Copy link
Collaborator

Image7.30x vs 10.45x

@yinian-lw
Copy link

That is a great idea! But we haven't tried it.
@BrianChen1129
我想试试，但看起来sta算子实现没有backward，应该可以用FlexAttention先做Finetune吧。
我想请教一下FlexAttention和现在STA算子执行效率有多少差异

hi～ 这个你有实现吗，效果怎么样？

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
4 participants
@wangshankun @foreverpiano @yinian-lw @BrianChen1129