Skip to content

b1663

Latest

Choose a tag to compare

@github-actions github-actions released this 20 Dec 17:17
799fc22
CUDA: Faster Mixtral prompt processing (#4538)

* CUDA: make MoE tensors contiguous for batch size>1

* Update ggml-cuda.cu

Co-authored-by: slaren <slarengh@gmail.com>

---------

Co-authored-by: slaren <slarengh@gmail.com>