Commit 75aa7b4

JohannesGaessler

authored and

committed

CUDA: faster FlashAttention, kernel for bs == 1

1 parent 08e69c5 commit 75aa7b4Copy full SHA for 75aa7b4

1 file changed

+906

-451

lines changed

ggml-cuda
- fattn.cu

1 file changed

+906

-451

lines changed

Comments

(0)