1f8ddd6d1c71fc2d08999e0e49193ffc269c86d7,python/baseline/pytorch/tagger/train.py,TaggerTrainerPyTorch,_train,#TaggerTrainerPyTorch#Any#,123

Before Change

            self.optimizer.zero_grad()
            loss = self.model.compute_loss(inputs)
            loss.backward()
            torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.clip)
            self.optimizer.step()
            bsz = self._get_batchsz(batch_dict)
            report_loss = loss.item() * bsz
            epoch_loss += report_loss

After Change

            loss = self.model.compute_loss(inputs)
            loss.backward()

            if (i+1) % self.grad_accum == 0 or (i+1) == steps:
                torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.clip)
                self.optimizer.step()
                self.optimizer.zero_grad()

            bsz = self._get_batchsz(batch_dict)
            report_loss = loss.item() * bsz
            epoch_loss += report_loss
            epoch_norm += bsz

Instances