Fix fused indices conversion for non-power-of-two topk/local experts#150
Open
zhujian19891203 wants to merge 1 commit into
Open
Fix fused indices conversion for non-power-of-two topk/local experts#150zhujian19891203 wants to merge 1 commit into
zhujian19891203 wants to merge 1 commit into
Conversation
Pad Triton arange ranges for topk and local expert counts to the next power of two, while masking out padded lanes. This avoids Triton compilation failures when DeepEP/flex dispatch uses values such as topk=7. Upstream-source: NVIDIA/Megatron-LM@bc70535 Local-port-of: 00b19053b091205316732411e0c5c6dfed355525
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Port the Megatron-LM fix for fused indices-to-multihot conversion when topk or num_of_local_experts is not a power of two.
This fixes a Triton compilation failure seen with DeepSeek V2 MoE models when using DeepEP/flex MoE token dispatch, for example with topk=7.
Source
Upstream fx: NVIDIA/Megatron-LM@bc70535.