Loss backward from sum of all values in array

Hi all,

I have an array of loss values called loss_arr:
[tensor([0.6828], grad_fn=<SubBackward0>), tensor([0.152], grad_fn=<SubBackward0>)]

When I tried to backward the total loss, I tried to do:

loss = sum(loss_arr)

However it gave me:

RuntimeError: Trying to backward through the graph a second time, but the buffers have already been freed. Specify retain_graph=True when calling backward the first time.

If I backward like:


for example, there is no problem.

So, how can I backward the total loss?
And, why am I failed to backward?

Thank you all, have a nicely quarantined days…


Can you try torch.sum instead of sum ?