perf(verl): reduce peak memory when decoding routed experts - #636

perf(verl): reduce peak memory when decoding routed experts#636

Bozhen Peng (kiteretsu903) wants to merge 1 commit into

microsoft:mainmicrosoft/agent-lightning:mainfrom

kiteretsu903:perf-stream-routed-expertskiteretsu903/agent-lightning:perf-stream-routed-expertsCopy head branch name to clipboard