R: element 0 of tensors does not require grad and does not have a grad_fn

So i am trying to Run a neural network that tries to detect the YOGA style. I scrapped dataset from google for 13 diiferent YOGA positions. Here is my NN. Kindly help as i am getting this error
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn


dl = torch.utils.data.DataLoader(YogaPoseDataset('classifier_train_data2.csv', trn_tfms, 
                                                           pose_id_to_name, pose_name_to_id),
                                           bs, 
                                           shuffle=True)
model = resnet34(pretrained=True)
model.fc = nn.Linear(512, 14)   
model_utils.freeze_all_layers(model)
model = model.to(device)

model.train()
n_epochs = 2
lr = 0.00001
loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr = lr)

print("Fine till here")

total_steps = 0
for e in range(n_epochs):
    avg_batch_accuracy = []
    for batch, labels in dl:
        optimizer.zero_grad()
        
        batch = batch.to(device)
        labels = labels.to(device)
        print(labels)
        
        preds = model(batch)
        print(preds)
        loss = loss_fn(preds, labels)
        loss.backward()
        optimizer.step()
                
        pred_labels = preds.argmax(dim=1)
        total_steps +=1
        batch_accuracy = (labels == pred_labels).float().mean()   
        avg_batch_accuracy.append(batch_accuracy.item())
        model_utils.print_training_loss_summary(loss.item(), total_steps, e+1, n_epochs, len(dl))
    
    print('avg batch accuracy:{:.2f}'.format(np.array(avg_batch_accuracy).mean()))

The error message wants to tell you it doesn’t have anything w.r.t. which you would want to compute grads.
One might naïvely think freeze_all_layers could cause training to bite on ice here.

Best regards

Thomas