Optimize top-2 expert group score computation in MoE routing. - #5124
Merged
Merged
Conversation
copybara-service
Bot
requested review from
RissyRan,
darisoy,
gagika,
igorts-git,
michelle-yooh,
richjames0,
shralex,
shuningjin and
zxhe-sean
as code owners
September 3, 2026 04:46
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
copybara-service
Bot
force-pushed
the
test_975290726
branch
3 times, most recently
from
September 3, 2026 21:25
f5e53f7 to
95d4b67
Compare
In DeepSeek-style expert group routing, groups of experts are selected based on the sum of their top-2 expert scores. When using `jax.lax.top_k(scores_grouped, k=2)` with small group sizes (such as 32 experts per group in DeepSeek-V3), XLA:TPU falls back to `BitonicSortEmitter` rather than `SortPrefixFusion`, introducing significant sort overhead and register padding. PiperOrigin-RevId: 975986068
copybara-service
Bot
force-pushed
the
test_975290726
branch
from
September 3, 2026 23:34
95d4b67 to
5033136
Compare
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Optimize top-2 expert group score computation in MoE routing.
In DeepSeek-style expert group routing, groups of experts are selected based on the sum of their top-2 expert scores. When using
jax.lax.top_k(scores_grouped, k=2)with small group sizes (such as 32 experts per group in DeepSeek-V3), XLA:TPU falls back toBitonicSortEmitterrather thanSortPrefixFusion, introducing significant sort overhead and register padding.