This is a problem that has come up before. The issue is that apparently the grader uses an older version of pytorch that is less forgiving in how torch.cat handles inputs of different types. Here’s an earlier thread on this. Please have a look and see if that is relevant to your case.