fix: place each MoE expert-cache pack on the GPU that runs its layer - #16

fix: place each MoE expert-cache pack on the GPU that runs its layer#16

thecodacus wants to merge 3 commits into

perfthecodacus/llama.cpp:perffrom

perf-multigpu-cachethecodacus/llama.cpp:perf-multigpu-cacheCopy head branch name to clipboard