[Feature Request] Early stopping

[akwarii]

Summary

Adding early stopping callback and add it as an additional training parameter for training

Detailed Description

As of now one need to train a model until numb_steps is reached. It could be convenient to have a option to enable early stopping (with a tunable patience parameter) in order to accelerate the model training. It's especially useful if one wants to perform hyperparameters search.

Further Information, Files, and Links

No response

[njzjz]

I think the strategy of early stopping needs discussion. In the actual training, the prefactor of energies and forces changes with the number of steps, so the total loss may always decrease.