Hi all
I have a model which has an encoder and two decoders. Each decoder has a separate loss function and target, so I think I can run both decoders separately. Is this possible using torch.multiprocessing?
So I want to do something like
code = self.encoder(input0)
decode1 = self.decoder1(code, input1)
decode2 = self.decoder2(code, input2)
loss1 = loss_fn(decode1, output1)
loss2 = loss_fn(decode2, output2)
total_loss = loss1 + loss2
self.optimizer.zero_grad()
total_loss.backward()
self.optimizer.step()
and concurrently run self.decoder1 and self.decoder2. Is this possible?