Add minimal sample dataset to Training Example

[EduardoPach]

It would be great if we could add a minimal sample dataset to the example training notebook introduced at #77 so we can run it end to end without any modifications and to inspect the code.

As context I am actually asking this so I can easily test the code and use that as a baseline to fix huggingface/transformers#31434

c.c. @SahilCarterr