Skip to content

Conversation

@xinxilwl
Copy link
Contributor

@xinxilwl xinxilwl commented Jun 4, 2025

Given suggestions from Dr. Chen in #17883 , some commits are withdrawn. And #17486 is solved.

@tqchen
Copy link
Member

tqchen commented Jun 4, 2025

Sorry i take a closer look, seems a better fix would be to refactor and remove attention_bias op, and ensure that the attention operator would accept three or four arguments, so it would either take three or four argument here.

cc @cyx-6

@tqchen
Copy link
Member

tqchen commented Jun 5, 2025

actually seems there is a need to introduce attention_bias variant here, so let us first go with this version

Copy link
Contributor

@cyx-6 cyx-6 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for fixing this! This fix looks good to me.

@cyx-6 cyx-6 merged commit 76946b4 into apache:main Jun 5, 2025
11 checks passed
ShiboXing pushed a commit to ShiboXing/tvm that referenced this pull request Aug 10, 2025
* Resolving inconsistency between attention/attention_bias

* reformat

* reduce the length of line

---------

Co-authored-by: taylor <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants