perf(kernels): faster sparse-attention top-k and scoring, slot-parallel MoE down gathers - #86

perf(kernels): faster sparse-attention top-k and scoring, slot-parallel MoE down gathers#86

asher merged 5 commits into

mainasher/mlx-kquant:mainfrom

perf/q4-selectasher/mlx-kquant:perf/q4-selectCopy head branch name to clipboard