llama.cpp
CUDA: tune fp16 tile FlashAttention configs for head sizes 40-112
#26289

Merged

CUDA: tune fp16 tile FlashAttention configs for head sizes 40-112 #26289

ggerganov merged 2 commits into ggml-org:master from animeshsri14:perf/pascal-fattn-tile-fp16-tune

CUDA: tune fp16 tile FlashAttention configs for head sizes 40-112

8575ea06

animeshsri14 requested a review 86 days ago

github-actions added ggml

github-actions added CUDA

Merge branch 'ggml-org:master' into perf/pascal-fattn-tile-fp16-tune

750d90d7

JohannesGaessler approved these changes on 2026-09-24

JohannesGaessler added merge ready

ggerganov merged 36d7b083 into master 26 days ago

Reviewers

JohannesGaessler

Assignees

No one assigned

Labels

ggml merge ready CUDA

Milestone

No milestone