How are evals done on trained models?

Thanks for putting this together. I am wondering how are evals done on trained models.

Are there some third-party evaluation libraries that you use to measure trained model performance/metric, or some other eval code available?

Thanks in advance.

