[moe] Fuse per-slot routed-expert launches: 8 kernel launches per layer to 2 (+15.4%) - #1

[moe] Fuse per-slot routed-expert launches: 8 kernel launches per layer to 2 (+15.4%)#1

MRPRESIDENT66 merged 1 commit into

MRPRESIDENT66:moe-routed-expert-fusionMRPRESIDENT66/GPULlama3.java:moe-routed-expert-fusionfrom

mikepapadim:opt/moe-fused-expert-launchesmikepapadim/GPULlama3.javaF:opt/moe-fused-expert-launchesCopy head branch name to clipboard