vllm
[Kernel] Refactor FlashInfer allreduce for mnnvl backend
#34109

Merged

[Kernel] Refactor FlashInfer allreduce for mnnvl backend #34109

ProExpertProg merged 9 commits into vllm-project:main from CentML:hjjq/ar

mergify added nvidia

gemini-code-assist commented on 2026-02-09

mergify added needs-rebase

mergify removed needs-rebase

mergify added documentation

mergify added ci/build

mergify added deepseek

mergify added frontend

mergify added llama

mergify added multi-modality

mergify added new-model

mergify added performance

mergify added qwen

mergify added gpt-oss

mergify added rocm

mergify added cpu

mergify added speculative-decoding

mergify added v1

squash merge

b2c39595

hjjq force pushed to b2c39595 105 days ago

hjjq marked this pull request as ready for review 105 days ago

hjjq requested a review from

zou3519 105 days ago

hjjq requested a review from

youkaichao 105 days ago

hjjq requested a review from

ProExpertProg 105 days ago

Allow both backends to be used at the same time

a118b20f

Merge branch 'main' into hjjq/ar

159bdd6b

Test both trtllm and mnnvl backends in test_fusion_all_reduce.py

fe3458c3

mgoin requested a review from

mgoin 98 days ago

Merge branch 'main' into hjjq/ar

922a4dbf

ProExpertProg commented on 2026-02-24

ilmarkov commented on 2026-02-24

ProExpertProg commented on 2026-02-25

Add flashinfer AR to benchmark_device_communicators.py

955630f1

wzhao18 force pushed to 955630f1 92 days ago

ilmarkov approved these changes on 2026-02-25

mergify added needs-rebase

ProExpertProg approved these changes on 2026-02-25

Merge main

8948a590

mergify removed needs-rebase

wzhao18 commented on 2026-02-25

Special warning for multicast check in AR workspace initialization fa…

3f3dc6f8

ProExpertProg approved these changes on 2026-02-25

ProExpertProg enabled auto-merge (squash) 91 days ago

github-actions added ready

Merge branch 'main' into hjjq/ar

746c8fed

mgoin removed documentation

mgoin removed new-model

mgoin removed rocm

mgoin removed frontend

mgoin removed speculative-decoding

mgoin removed ci/build

mgoin removed v1

mgoin removed multi-modality

mgoin removed llama

mgoin removed qwen

mgoin removed deepseek

mgoin removed cpu

mgoin removed gpt-oss

mgoin approved these changes on 2026-02-26

ProExpertProg merged 71dfce6a into main 91 days ago

hjjq deleted the hjjq/ar branch 72 days ago

Reviewers

mgoin

ProExpertProg

ilmarkov

wzhao18

gemini-code-assist

zou3519

youkaichao

Assignees

No one assigned

Labels

performance ready nvidia

Milestone

No milestone

vllm [Kernel] Refactor FlashInfer allreduce for mnnvl backend #34109 Merged

[Kernel] Refactor FlashInfer allreduce for mnnvl backend #34109

vllm
[Kernel] Refactor FlashInfer allreduce for mnnvl backend
#34109

Merged