Hi,
Thank you for providing the back translation code.
What are those assertion errors mentioned in your HW5 code?

Regards
Utkarsh

When i run the HW7 TA example code, my cpu ram keep increasing until out of my 64G ram.I think this problem is coming from the training process.
train_loss += output.loss # store the complete compution graph leading to out of memory
train_loss += output.loss.detach() or .item() if i modify original code to this, problem solved.

So, my question is, didnt you come across this problem? If not, why this happened to me, but TA and your code works well.
Just assume TAs example code should not be wrong, so afraid of myself may ignore something important.
Really hope you can help me, thanks in advance.

Hi,
Thank you for the hw7/hw7_macbert4.ipynb code. When I run it it gives error pointing to this line in Training cell "dev_acc += evaluate(data, output, doc_stride, dev_paragraphs[dev_questions[i]['paragraph_id']],
dev_paragraphs_tokenized[dev_questions[i]['paragraph_id']].tokens) == dev_questions[i]["answer_text"]"

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking arugment for argument mask in method wrapper_masked_select)

Please give some suggestion on how it can be solved.