Issue about the backward()

[PolarisV]

When I Calculate the loss using this moddle,the error is RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation . But When I Use CrossEntropyLoss it runs successfully.My code :

lossF = CRF(num_tags=num_tags, batch_first=True)
for batch in tqdm(trainloader, desc=f"Training Epoch {epoch}"):
            optimizer.zero_grad()
            inputs, targets = [x.to(device) for x in batch]
            if use_crf:
                targets = targets
            else:
                targets = targets.view(-1)
            if use_crf:
                bert_output = model(inputs, use_crf=use_crf)
                loss = lossF(bert_output, targets)
            loss.backward()

[Kemal Kurniawan]

I'm guessing you have in-place operations inside your model forward(). That error is usually caused by such operations.