llama.cpp
Add sm70 FlashAttention config case for DKQ=256, DV=256, ncols=64
#27997

Merged

Add sm70 FlashAttention config case for DKQ=256, DV=256, ncols=64 #27997

ggerganov merged 2 commits into ggml-org:master from mistrjirka:perf/volta-fa-256x256

CUDA : add sm70 FlashAttention config case for the 256x256 head

327f0b14

mistrjirka requested a review 75 days ago

github-actions added ggml

github-actions added CUDA

mistrjirka closed this 75 days ago

mistrjirka reopened this 75 days ago

mistrjirka marked this pull request as draft 75 days ago

JohannesGaessler commented on 2026-10-24

CUDA : tune sm70 FlashAttention for 256x256 ncols=32

35931592

mistrjirka marked this pull request as ready for review 19 days ago

JohannesGaessler approved these changes on 2026-11-05

JohannesGaessler added merge ready

ggerganov merged 77e2db8c into master 2 days ago

Reviewers

JohannesGaessler

Assignees

No one assigned

Labels

ggml merge ready CUDA

Milestone

No milestone