2 ms·
on my M1 Max, MTP consistently lowers my performance! I’ve tried both llama-cpp’s recently landed MTP support (cloned and built Tuesday) as well as one of the o
by gcr 5mo ago
on my M1 Max, MTP consistently lowers my performance! I’ve tried both llama-cpp’s recently landed MTP support (cloned and built Tuesday) as well as one of the other forks a few weeks ago. Suspect nobody’s done a comparison on hardware like mine.