-
Notifications
You must be signed in to change notification settings - Fork 145
Fix prebuild for cktile moe kernel #1423
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.
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
base: main
Are you sure you want to change the base?
Conversation
zhiding512
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
1f88988 to
c2ca6ac
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Pull Request Overview
This PR fixes a missing pybind registration for the CKtile 2-stage MOE kernel that was not properly registered when ported from the 355_wip branch. The fix enables the CKtile MOE functions to be accessible from Python.
- Adds the missing header include for
moe_cktile2stages.h - Registers the
MOE_CKTILE_2STAGES_PYBINDmacro to exposecktile_moe_gemm1andcktile_moe_gemm2functions
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
@zhiding512 / @valarLip PTAL, can we merge to unblock vllm/355_wip? Unrelated CI failures due to git ssh outage earlier today |
Motivation
While working on https://github.com/ROCm/vllm/pull/808/files, we've found that building the aiter wheel then linking it into the final docker build stage leads to a broken build, which does not occur when building the library in-place. cc @dllehr-amd @zhiding512
It looks like the CKtile 2stage MOE was not registered in pybind when it was ported over from 355_wip: https://github.com/ROCm/aiter/blob/355_wip/csrc/rocm_ops.cpp#L80; linking it fixes the issue.
Also updated the CK 2stage MOE to use FLATMM_HIP_CLANG_PATH.Technical Details
Test Plan
Test Result
Submission Checklist