Fix ragged gather reduce correctness#4262
Open
NuojCheng wants to merge 2 commits into
Open
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
NuojCheng
force-pushed
the
chengnuojin-ragged-followup
branch
3 times, most recently
from
3b816f6 to
e4b450a
Compare
NuojCheng
force-pushed
the
chengnuojin-ragged-followup
branch
from
e4b450a to
8c14942
Compare
NuojCheng
force-pushed
the
chengnuojin-ragged-followup
branch
2 times, most recently
from
64a11c0 to
9e10916
Compare
When prefuse_moe_weights=True (requires sparse_matmul=True), the two FFN1 expert weight matrices [G,K,N] are concatenated into [G,K,2N] and dispatched as a single grouped GEMM, then split. This halves FFN1 kernel launches and reads input activations from HBM once instead of twice. Backend-agnostic: works with Megablox, Tokamax, and jax.lax.ragged_dot. When attention=vllm_rpa the fused tensor is passed directly to the vLLM-TPU serving kernel. correct sort kernel convergence
NuojCheng
force-pushed
the
chengnuojin-ragged-followup
branch
from
9e10916 to
bd32f80
Compare
NuojCheng
requested review from
A9isha,
RissyRan,
SurbhiJainUSC,
abhinavclemson,
aireenmei,
bvandermoon,
gagika,
gobbleturk,
hengtaoguo,
khatwanimohit,
richjames0,
shralex and
vipannalla
as code owners
NuojCheng
requested review from
NicoGrande,
darisoy,
dipannita08,
igorts-git,
jiangjy1982 and
suexu1025
as code owners
gobbleturk
approved these changes
Jun 29, 2026
Shuwen-Fang
approved these changes
Jun 29, 2026
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
4 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
This PR does three things:
ragged_gather_reducekernel solving correctness using ragged sort kernels;ragged_gather_reducekernel of small tensors. It breaks our ragged sort unit test;Tests
CI tests
Checklist
Before submitting this PR, please make sure (put X in square brackets):
gemini-reviewlabel.