Skip to content

397B v7x RL trainer: GDN kernel, MoE routing and engine speedups (10.62 -> 9.80 s per micro-batch on the 128-chip bench) - #5759

Closed
khatwanimohit wants to merge 11 commits into
atwigg/mlperffrom
mohit/trainer-perf-v7x
Closed

khatwanimohit wants to merge 11 commits into
atwigg/mlperffrom
mohit/trainer-perf-v7x

GDN CP backward: dS_ext_local as a reverse Pallas scan

4026f50
Select commit
Loading
Failed to load commit list.