Enhance Documentation: Specify Model Saving Criteria Post-Training

[GrunCrow]

The current documentation lacks information about the type of model saved after training—whether it's the last checkpoint, the best checkpoint based on a specific metric, or another criterion. This crucial detail is essential for users familiar with deep learning who require clarity on model saving procedures. Adding this information to the documentation would greatly benefit users seeking to understand and optimize their workflow without needing to delve into the codebase.