Hello, I am encountering the following error while fine-tuning a large language model. The code initially ran without issues, but when I executed it again two hours later, I received this error: Expected input batch_size (220) to match target batch_size (63).
This error is often caused by view or reshape operations manipulating the batch size while in fact other dimensions should be changed.
E.g. are you using x = x.view(-1, ...) somewhere in your model (where x could be an activation or target)? If so, replace it with x = x.view(x.size(0), -1, ...) to keep the batch size.
I can’t find where this applies in my code. Could you help me locate it?