Dan Kondratyuk

How's the training going? Any problems?

This seems to me like a newer version of PyTorch made an incompatible change `torch.load`, which leads to it saying that the file might be corrupted. It seems unlikely that...

Hmm, this seems tricky. Looks like some others report issues with the main HuggingFace library: https://github.com/huggingface/transformers/issues/6620 https://github.com/huggingface/transformers/issues/1491 There are a few solutions posed, but I'm not sure how applicable they...

Seems like it stops at `deserialized_objects[key]._set_from_file(f, offset, f_should_read_directly)`. Can you set a breakpoint/print statement and list out what the input variables are? Maybe it could give a clue. Or maybe...

Sorry for the late reply. I think there might be a bug in how the multiword IDs are handled. In this case, you don't have any multiword IDs because you...

I've seen NaN loss with this code before, but there are a lot of things that can cause it to happen. Sometimes it's an uninitialized variable that is set to...

@jbrry Thanks for looking into it. So it appears to be due to a recent PyTorch update. I updated the `requirements.txt` to explicitly use `1.4.0`. Hope that fixes it for...

It looks like the `reqs.txt` you used has an earlier version of torch (1.3.1). Would it work if you change the `requirements.txt` to `torch==1.3.1`?

Are you having problems downloading the files, or are you having trouble finding them? I can verify that I can download the weights at the bottom of the link. Sometimes...

The code assumes all paths point inside the repo for simplicity. One easy workaround is to create a symlink inside the repo that points to your desired directory/file. Otherwise, it...