perf(cuda/moe): backend-aware default for the fused decode-MoE Dff threshold - #643

perf(cuda/moe): backend-aware default for the fused decode-MoE Dff threshold#643

inureyes merged 1 commit into

mainlablup/mlxcel:mainfrom

feature/issue-626-cuda-fused-moe-dfflablup/mlxcel:feature/issue-626-cuda-fused-moe-dffCopy head branch name to clipboard