vllm
[MLA Attention Backend] Add TOKENSPEED_MLA backend for DSR1/Kimi K25 prefill + decode on Blackwell
#41778

Merged

[MLA Attention Backend] Add TOKENSPEED_MLA backend for DSR1/Kimi K25 prefill + decode on Blackwell #41778

ywang96 merged 23 commits into vllm-project:main from zyongye:feat/tokenspeed_mla_upstream

[Attention] Add TOKENSPEED_MLA backend for DeepSeek R1 prefill + deco…

d0e6514b

[Attention][TokenSpeed MLA] Fix decode FP8 numerics: pass output_scal…

964c6eb4

[Attention][TokenSpeed MLA] Warm up BF16 prefill compile, drop seq_le…

73cd7e25

[Attention][TokenSpeed MLA] Fix trtllm LSE parity test: log2 → natura…

00128182

[Attention][TokenSpeed MLA] Surface install hint when package missing

de6bc297

[Attention][TokenSpeed MLA] Force prefill V tensor contiguous before …

c0015350

zyongye requested a review from

mgoin 71 days ago

zyongye requested a review from

tlrmchlsmth 71 days ago

zyongye requested a review from

WoosukKwon 71 days ago

zyongye requested a review from

yewentao256 71 days ago

zyongye requested a review from

pavanimajety 71 days ago

zyongye requested a review from

LucasWilkinson 71 days ago

zyongye requested a review from

MatthewBonanni 71 days ago

claude commented on 2026-05-06

mergify added documentation

mergify added v1

gemini-code-assist commented on 2026-05-06

fix precommit

97cd2c41

ywang96 added ready

adding to cuda.txt dependency

91ef0afc

mergify added ci/build

mergify added nvidia

update version

c4547482

[Attention][TokenSpeed MLA] Also warm up prefill kernel from decode impl

58c8a5ea

MatthewBonanni commented on 2026-05-06

[Attention][TokenSpeed MLA] Address PR review: priorities, tests, ben…

29690bfa

zyongye force pushed from 388598f6 to 29690bfa 70 days ago

mergify added performance

update benchmarks

371d11af