returnn
returnn copied to clipboard
RF PT ignores parameter weight decay, only uses the global optimizer setting
rf.Parameter.weight_decay is ignored in the PyTorch engine.