-
Fei Sun authored
Summary: Pull Request resolved: https://github.com/facebookresearch/d2go/pull/526 Add a config variable: DDP_GRADIENT_AS_BUCKET_VIEW. Pass it to DDP. This variable reduces the memory consumption of the model. Reviewed By: tglik Differential Revision: D44273339 fbshipit-source-id: 272e2ffbea89532a55df0ebdb3bd49f0df7d78a5
8353ad23
