Thanks for the pointer to Gradio! Looks neat :)

However, if I understood you correctly, you are misleading readers in this post - cross validation does not eliminate the need for the `test set`. All it does, is replacement of a single validation set with a technique that can estimate performance of a set of parameters betters, but it does so by training a set of **different** models, not a single model. That's why when using cross-validation approach, once the best parameters are selected, you should train a final model on the whole training dataset and evaluate it on **test set** to get the "real world" performance measurement.

Thanks for putting this into a single post!

All the best!