AttributeError: 'str' object has no attribute 'view'

[wrecka98]

Hi! I am trying to customize a YOLOv11 training loop, but I am running into some issues when computing loss from the prediction tensors. The error happens at this point in the library:

File ~/.local/lib/python3.8/site-packages/ultralytics/utils/loss.py:211, in (.0)
209 loss = torch.zeros(3, device=self.device) # box, cls, dfl
210 feats = preds[1] if isinstance(preds, tuple) else preds
--> 211 pred_distri, pred_scores = torch.cat([xi.view(feats[0].shape[0], self.no, -1) for xi in feats], 2).split(
212 (self.reg_max * 4, self.nc), 1

I have tried several, ensuring that preds is indeed of type Torch.Tensor. Another thread (#10 ) suggested loading the model using YOLOv10 instead in YOLO. However, I am using YOLOv11 and understand that the YOLOv10 model loading is an experimental feature. I cannot implement it in my newest version of ultralytics.

Here is the code to my custom training loop:

def train_sparse_yolo_l1(model, dataloader, optimizer, device, num_epochs=5, l1_lambda=0.0001):

model = model.model.train()

for epoch in range(num_epochs):
    for batch in dataloader:
        # Move entire batch to device
        batch = {k: v.to(device) for k, v in batch.items()}
        images = batch["img"]  # Extract images from batch

        optimizer.zero_grad()

        # Forward pass
        preds = model(images)
        
        print(preds)
        

        # Compute loss (including L1 regularization)
        loss, _ = model.loss(tuple(preds), batch)

        # L1 Regularization on model weights
        l1_loss = sum(param.abs().sum() for param in model.parameters())
        total_loss = loss + l1_lambda * l1_loss

        # Backpropagation
        total_loss.backward()
        optimizer.step()

A swift answer would be highly appreciated. Thank you in advance!