Protein Gym Eval script, gitignore

[Muedi]

Hi,

as discussed, here is a basic start for the proteingym eval script :) As is it runs on a rtxx 4090 but the VRAM is full.

The script has a header with a discription.

Things to do/discuss downstream:

• Implement our own trainingloop -> Pascal has some code ready that would fit already apparently
• Change Model to a smaller version to be able to handle it on smaller machines during the starting phase and get back to the bigger version for later phase? (Not sure if necessary)
• Introduce some folder structure in the repo. On my local repo I have a data/ProteinGym/ folder, I'd suggest to add these, I just kept them in the gitignore for now :)

[pascalnotin]

Lgtm for the base ProteinGym supervised eval script!

[jamaliki]

If I may make a small change, the following are the HuggingFace Esm tokens:

['<cls>', '<pad>', '<eos>', '<unk>', 'L', 'A', 'G', 'V', 'S', 'E', 'R', 'T', 'I', 'D', 'P', 'K', 'Q', 'N', 'F', 'Y', 'M', 'H', 'W', 'C', 'X', 'B', 'U', 'Z', 'O', '.', '-', '<null_1>', '<mask>']

So we want to change "[CLS]" and "[EOS]"

[Muedi]

Thanks, I'll change it in the evening :)