llama.cpp
CUDA: FA support for Deepseek (Ampere or newer)
#13306

Merged

CUDA: FA support for Deepseek (Ampere or newer) #13306

JohannesGaessler merged 4 commits into ggml-org:master from JohannesGaessler:cuda-deepseek-fa-4

CUDA: FA support for Deepseek (Ampere or newer)

d19838e9

github-actions added Nvidia GPU

github-actions added python

github-actions added ggml

CISC commented on 2025-05-04

JohannesGaessler force pushed 268 days ago

wrap __cvta_generic_to_shared for HIP

187054a7

JohannesGaessler force pushed to 187054a7 268 days ago

fix loop unrolling for KV data load

dd054465

slaren approved these changes on 2025-05-08

do loop unrolling via C++ template

fe2b775a

JohannesGaessler merged 0cf6725e into master 265 days ago

Reviewers

slaren

CISC

Assignees

No one assigned

Labels

Nvidia GPU python ggml

Milestone

No milestone