LARS LAMB optimizers

I searched inside https://pytorch.org/docs/master/optim.html but haven’t found any Layer-wise AdaptiveRate Scaling optimizer LARS implemented. Same for LAMB. Are there any plans soon? Should we add FR?

Hi,
LARC is not available directly in PyTorch but it is in Apex https://github.com/NVIDIA/apex/blob/master/apex/parallel/LARC.py

1 Like