IssueTracker

This feature request is for a user friendly way to be able to stop a training of a Natural Language model while preserving its training state at the moment of stopping.

Without losing the model.

Reasoning:

1. The initial results are satisfactory, and the remainder of the training is spent on improving impurities, which can lead to overfitting.

2. The training picks a wrong direction, and there is no point in wasting anymore time on training it.

3. Pricing.

4. Training fails on an easy set of examples, and there is no point to continue training it on a more complex one.