vllm
[Kernel] Refactor CUTLASS kernels to always take scales that reside on the GPU
#5137

Merged

[Kernel] Refactor CUTLASS kernels to always take scales that reside on the GPU #5137

robertgshaw2-redhat merged 7 commits into vllm-project:main from neuralmagic:tms/cutlass_scale_fixes

Refactor CUTLASS kernels to always take scales that reside on the GPU

07d12785

tlrmchlsmth force pushed to 07d12785 2 years ago

Allowing subtile in codespell

57cbe258

Extend cutlass unit test for cuda graphs to check the scalar case

24ff23cc

tlrmchlsmth marked this pull request as ready for review 2 years ago

bnellnm commented on 2024-05-30

bnellnm approved these changes on 2024-05-30

pcmoritz commented on 2024-05-31

Merge branch 'upstream-main' into tms/cutlass_scale_fixes

6c5dcf62

review comments, revert formatting changes from upstream

f7484bc6

fixup, tag cutlass version

52100b1e

fixup comment

aaf5fdec

pcmoritz approved these changes on 2024-06-01

robertgshaw2-redhat enabled auto-merge (squash) 2 years ago

mgoin approved these changes on 2024-06-01

robertgshaw2-redhat merged 260d119e into main 2 years ago

robertgshaw2-redhat deleted the tms/cutlass_scale_fixes branch 2 years ago

Reviewers

mgoin

pcmoritz

bnellnm

Assignees

No one assigned

Labels

None yet

Milestone

No milestone