Reduce peak vram in moe.py by avoiding cached router indices - #464
Open
shedrachIkenna wants to merge 1 commit into
Open
Reduce peak vram in moe.py by avoiding cached router indices #464shedrachIkenna wants to merge 1 commit into
shedrachIkenna wants to merge 1 commit into
