A problem when training.

[Du]

Hi there!
I've recently read your paper Sequential and Diverse Recommendation with Long Tail published on the SIGIR.
It's undoubtedly a great work. But when I downloaded the code here and tried to deploy the GRU4DIV model without content embedding on the MovieLens dataset, some problems occurred.
Of course I did change some code in the pre-process.py but for the main.py, nearly no changes in the core codes.
I noticed that you had used a linklist data structure to store the item sequences which is a good job for saving memory.
But when it came to the second iteration, at the loss.backward() line, there was a "RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed)." message.
Actually, I've read the whole script but I didn't find why it was the second time to backward.
Any tips for this problem would be helpful.
Thank you.