Tristan Bepler

I added this in commit 752c140a709c745dabdcc2232b6e9444a11e1ef1

Thanks for the usability improvement pull request. I'm happy to accept this if you can change the version requirements to be minimum version numbers where possible (e.g. torch>=0.4.0).

The model produces an embedding for each position of the sequence, so a sequence of length L will always produce an LxD (where D is the dimension of the embedding)...

I recommend PCA.

If you are training a new model from scratch, you can adjust the hidden dimension and the number of layers to change D. These are already fixed for the pre-trained...

There is no length limit.

@pconesa Did you figure out a solution to this? It sounds like an issue with your CUDA and/or pytorch installation rather than topaz itself.

We have mid-term plans to upstream this into topaz, but I don't have a concrete timeline for it. If someone from the community wants to volunteer to update the filament...

This is annoying, but it's nothing to worry about. The warning doesn't cause any problems.

Thanks for reporting this and opening a PR. We'll take a look.