vllm
[Model][Perf] Fuse Beam's TP all-reduce with its sandwich norms and MoE finalize
#62366
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
6
Changes
View On
GitHub
[Model][Perf] Fuse Beam's TP all-reduce with its sandwich norms and MoE finalize
#62366
WoosukKwon
merged 6 commits into
main
from
woosuk/beam-fused-allreduce
mergify
added
ci/build
mergify
added
nvidia
WoosukKwon
force pushed
from
f1d7247e
to
32fc9c80
33 days ago
Base automatically changed from
woosuk/beam-model
to
main
33 days ago
[Model][Perf] Fuse Beam's TP all-reduce with its sandwich norms and M…
4fb4e753
mergify
added
documentation
mergify
added
new-model
mergify
added
speculative-decoding
mergify
added
dflash
mergify
added
mrv2
mergify
added
needs-rebase
WoosukKwon
force pushed
from
32fc9c80
to
4fb4e753
33 days ago
mergify
removed
needs-rebase
[Model][Perf] Build Beam's fused all-reduce ahead of time under csrc/…
5d683553
[Model][Perf] Note FlashInfer v0.7.2 as the vendored MNNVL source
3ec2cf25
[Model][Perf] Keep Beam's deferred MoE finalize TP-only
9fdc9ed5
[Build] Gate Beam's csrc kernels under one CMake switch
df583a4b
WoosukKwon
marked this pull request as ready for review
30 days ago
WoosukKwon
requested a review
from
DarkLight1337
30 days ago
WoosukKwon
requested a review
from
ywang96
30 days ago
WoosukKwon
requested a review
from
AndreasKaratzas
30 days ago
WoosukKwon
requested a review
from
tlrmchlsmth
30 days ago
WoosukKwon
requested a review
from
LucasWilkinson
30 days ago
WoosukKwon
requested a review
from
Harry-Chen
30 days ago
Merge branch 'main' into woosuk/beam-fused-allreduce
73195883
claude
commented on 2026-10-21
WoosukKwon
added
ready
WoosukKwon
merged
e4e6bcaa
into main
30 days ago
WoosukKwon
deleted the woosuk/beam-fused-allreduce branch
30 days ago
Login to write a write a comment.
Login via GitHub
Reviewers
claude
DarkLight1337
ywang96
AndreasKaratzas
tlrmchlsmth
LucasWilkinson
Harry-Chen
Assignees
No one assigned
Labels
documentation
new-model
speculative-decoding
ready
ci/build
nvidia
mrv2
dflash
Milestone
No milestone
Login to write a write a comment.
Login via GitHub