Offline run of DiffDock-pp

[slieped]

Consider to add utilities / modify code to work with offline computing resources.
Most HPC do not have direct internet conection to internet.
Thus, the use of torch.hub to download ESM model, might be problematic!

I came up with a simple solution that could be integrated (or at least mentioned in furhter examples)

0. Instal esm package via pip:
   pip install fair-esm
   (https://github.com/facebookresearch/esm)

1. Download model and regression .pt files:
   https://dl.fbaipublicfiles.com/fair-esm/models/esm2_t33_650M_UR50D.pt
   https://dl.fbaipublicfiles.com/fair-esm/regression/esm2_t33_650M_UR50D-contact-regression.pt

2. Import esm function to load precomputed models:
   from esm.pretrained import load_model_and_alphabet_local

3. Modify data.train.utils.compute_emedding function [325-327]:
   modelpath = 'path/to/model/esm2_t33_650M_UR50D.pt'
   esm_model, alphabet = load_model_and_alphabet_local(modelpath)

I do know that with torch.hub its possible to pre-cache the files. And then just load the pre-downloaded ones, but its not ideal
This is just a consideration, to make the tool more scalable and useful for other teams!

Victor M

[ketatam]

Hi!

Thanks a lot for the detailed suggestion. I will test your approach and then update the README.md accordingly.