perf(kernels): faster q8_0 mat-vec and iq2_s MoE gathers - #92

perf(kernels): faster q8_0 mat-vec and iq2_s MoE gathers#92

asher merged 4 commits into

mainasher/mlx-kquant:mainfrom

muse-spark-decodeasher/mlx-kquant:muse-spark-decodeCopy head branch name to clipboard