How to use this in PyTorch DDP?

Mikoto10032 / AutomaticWeightedLoss

Multi-task learning using uncertainty to weigh losses for scene geometry and semantics, Auxiliary Tasks in Multi-task Learning

Apache License 2.0

575 stars 81 forks source link

Closed ygtxr1997 closed 2 years ago

ygtxr1997 commented 2 years ago

DDP means DistributedDataParallel

ygtxr1997 commented 2 years ago

awl = AutomaticWeightedLoss(2).cuda()

is ok