Optimize Qwen3.6-27B dense 4-bit prefill/decode (affine-4 BF16 QMM) - #64

Optimize Qwen3.6-27B dense 4-bit prefill/decode (affine-4 BF16 QMM)#64

goniz merged 9 commits into

maingoniz/mlx-vulkan:mainfrom

cursor/qwen36-27b-dense-opt-2578goniz/mlx-vulkan:cursor/qwen36-27b-dense-opt-2578Copy head branch name to clipboard