🐅 Epic: Fix critical bugs in Ersilia

[GemmaTuron]

Summary

There have been some bugs that cause the models not to work at the general Ersilia Level. This is hard to un-link from other issues but I will try to collect here bugs that need to be fixed:

• Automatic deduplication of SMILES: 🐛 Bug: Variable number of rows between input and output #1535
• Parsing of None outputs to proper format: Descriptors do not parse Nones to h5 zairachem-docker#7
• Automatic selection of BentoML or FastAPI packaging: 🐈 Task: Revise resolution method for Dockerized images in models #1509

It would be good to add these checks to the model test or the Ersilia CLI workflows to ensure they do not repeat. @Abellegese for the parsing to None outputs, we thought we could run a one single check in the Ersilia CLI with one model and one molecule we know fails in this model and ensure the None output is properly parsed. Thoughts?

Objective(s)

• Ensure Ersilia is fully functional

Documentation

Ersilia's gitbook

[Abellegese]

Yes @GemmaTuron I agree with that. Whats even best is I think to create a small functions that enforces a correct parsing for all models. How do we do this? One way is to use the schema of the api to construct parsing rules the None as well as for the correct values.