Break does not work as expected

1

When I add the break in the inner for loop, I expect to get the loss for only 1 batch. Not sure why I get 10 different lines with different loss values

loss_list=[]
for epoch in range(num_epochs):
    model.train()
    l=0
    for i, values in enumerate(train_iter):
        input_data = []
        for v in values:
            input_data.append(v.to(device)) #you should load here to GPU's if you have more than 1 GPU
            
        p_pos = [model(*t) for t in zip(*input_data[0:-1])]
        p_neg = [model(*t) for t in zip(*input_data[0:-2], input_data[-1])]
        ls = [loss(p, n) for p, n in zip(p_pos, p_neg)]
        optimizer.zero_grad()
        [l.backward(retain_graph=False) for l in ls]
        l += sum([l for l in ls])  #you should normalize by the number of devices if you have more than 1 GPU
        optimizer.step()        
        print(i, l)
        break
    loss_list.append(l)

Here are the results:
0 tensor([[572.0556]], device=‘cuda:0’, grad_fn=)
0 tensor([[560.5846]], device=‘cuda:0’, grad_fn=)
0 tensor([[564.1666]], device=‘cuda:0’, grad_fn=)
0 tensor([[562.2694]], device=‘cuda:0’, grad_fn=)
0 tensor([[546.3769]], device=‘cuda:0’, grad_fn=)
0 tensor([[564.4094]], device=‘cuda:0’, grad_fn=)
0 tensor([[552.5000]], device=‘cuda:0’, grad_fn=)
0 tensor([[537.7695]], device=‘cuda:0’, grad_fn=)
0 tensor([[540.4388]], device=‘cuda:0’, grad_fn=)
0 tensor([[543.6255]], device=‘cuda:0’, grad_fn=)

2

It would seem that break successfully breaks from the for loop on train_iter, but you’d still get the results for every epoch (the outer for loop).