vllm
[MISC] Fix Tensor Parallelism for Quantized Mamba Models with n_groups=1
#33257

Merged

[MISC] Fix Tensor Parallelism for Quantized Mamba Models with n_groups=1 #33257

tlrmchlsmth merged 3 commits into vllm-project:main from CentML:vadim/fix-falcon-fp8-tp

fix tp>1 for quntized mamba models

88743733

vadiklyutiy requested a review from

tdoublep 119 days ago

vadiklyutiy requested a review from

tlrmchlsmth 119 days ago

vadiklyutiy assigned

vadiklyutiy 119 days ago

gemini-code-assist commented on 2026-01-28

cursor commented on 2026-01-28

fix

b3878efe

tomeras91 commented on 2026-01-28

vadiklyutiy force pushed to b3878efe 115 days ago

Unify MambaMixer2 TP sharding to use custom weight loader

d5d6d0b8

vadiklyutiy added ready

mgoin added bug

tlrmchlsmth commented on 2026-02-03

tlrmchlsmth approved these changes on 2026-02-03

tlrmchlsmth merged a372f3f4 into main 113 days ago

vadiklyutiy deleted the vadim/fix-falcon-fp8-tp branch 78 days ago

Reviewers

tlrmchlsmth

tomeras91

cursor

gemini-code-assist

tdoublep

Assignees

vadiklyutiy

Labels

bug ready

Milestone

No milestone