'BCEWithLogitsLoss' object has no attribute 'backward'

nlp
1

Hello guys,
I’m trying to fine-tune the Bert model i.e., bert-base-uncased for a text classification task. I’m getting a strange error while calling the backward() on loss, which is as follows

torch.nn.modules.module.ModuleAttributeError: ‘BCEWithLogitsLoss’ object has no attribute ‘backward’.

I can’t find any syntax error and also checked the inputs(outputs and targets) to the loss function i.e, nn.BCEWithLogitsLoss(outputs, targets) and I found then to be in the correct format. I don’t know what is causing this error. I would be great if any of you could help me with this…thanks in advance :slight_smile:

Here is the code

def train_loop_fn(data_loader, model, optimizer, device, scheduler = None):
model.train()
for bi, d in enumerate(data_loader):
ids = d[“ids”]
mask = d[“mask”]
token_type_ids = d[“token_type_ids”]
targets = d[“targets”]

    ids = ids.to(device, dtype=torch.long)
    mask = mask.to(device, dtype=torch.long)
    token_type_ids = token_type_ids.to(device, dtype=torch.long)
    targets = targets.to(device, dtype=torch.float)

    optimizer.zero_grad()
    outputs = model(ids=ids, mask = mask, token_type_ids = token_type_ids)
    outputs = outputs.reshape(1)
    print("here is the outputs")
    print(outputs)
    print("here is the targets")
    print(targets)
    loss = nn.BCEWithLogitsLoss(outputs, targets)
    print("this is the loss")
    print(loss)
    loss.backward()
    optimizer.step()
    if scheduler is not None:
        scheduler.step()
2

You have to create an object of the criterion first and later call it with your tensors.
Change:

loss = nn.BCEWithLogitsLoss(outputs, targets)

to

criterion = nn.BCEWithLogitsLoss()
for data in loader:
    ...
    loss = criterion(output, target)
    loss.backward()