We found that there are many pitfalls with reproducibility and in particular ESM C out of the box _will not_ give you reproducible embeddings. A major problem if you don't pay attention to it. Fortunately there are easy fixes.

Thanks. It would be interesting to use it for macrocycles like a recent Baker Lab paper that used AIMNET. Also, maybe for calculating energies for protein ligand systems as shown is Figure 6b.

Awesome, congratulations! Are there going to be some tutorials on setting up simulations (especially Biomolecular systems) at some point?