[MoE-FP8-modelopt] Add FlashInfer alignment padding for intermediate dimensions #29748
+48
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Daniel Afrimi <dafrimi@pool0-00589.cm.cluster>
Signed-off-by: Daniel Afrimi <dafrimi@pool0-00589.cm.cluster>
danielafrimi
requested review from
mgoin,
pavanimajety,
robertgshaw2-redhat,
tlrmchlsmth and
yewentao256
as code owners
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Code Review
This pull request introduces padding for the intermediate dimension of MoE weights to meet the alignment requirements of FlashInfer FP8 kernels. This is a necessary fix to prevent kernel launch failures for certain models and tensor parallelism configurations. The changes are well-contained within the ModelOptFp8MoEMethod, and the new _maybe_pad_intermediate_for_flashinfer function correctly pads the w13 and w2 weights with zeros and updates the intermediate_size_per_partition attribute on the layer. The implementation appears correct and robust.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds support for FlashInfer MoE FP8 kernels that require the intermediate (gated) dimension to be aligned.
Some models break when the intermediate size isn’t divisible by 16, especially under different TP.
This PR introduces _maybe_pad_intermediate_for_flashinfer, which pads w13 and w2 along the intermediate dim so the weights meet FlashInfer’s alignment constraints.
If padding is needed, we zero-pad the up/gate and down projection weights and update intermediate_size_per_partition accordingly. This keeps the model numerically correct while avoiding kernel launch failures.