FlashMLA-3 for DeepSeek models on CUDA#386

Merged

ikawrakow merged 7 commits intomainfrom

ik/cuda_flash_mla3

May 7, 2025

Commits on May 6, 2025

CUDA WIP: support for FlashMLA-3
Iwan Kawrakow
committed
Much better
Iwan Kawrakow
committed
Sadly, the previous commit was wrong
Iwan Kawrakow
committed
Finalizing
Iwan Kawrakow
committed
Also add these
Iwan Kawrakow
committed
Minor
Iwan Kawrakow
committed

Commits on May 7, 2025

Minor tweak
Iwan Kawrakow
committed