ggml-cuda: size MMQ tile to MoE tokens-per-expert - #39

ggml-cuda: size MMQ tile to MoE tokens-per-expert#39

roberteg16 merged 1 commit into

gfx11AMD-Ecosystem/llama.cpp:gfx11from

rogarcia.mmq-moe-tile-sizeAMD-Ecosystem/llama.cpp:rogarcia.mmq-moe-tile-sizeCopy head branch name to clipboard