If you try to run this in #Colab it will probably fail due to PIL.Image.ANTIALIAS and export_model errors

(deprecated libraries from 2021)

I fixed those here:
colab.research.google.com/github/tomas...

That should take you as far as "Training", but other bugs are in progress.

You bring up a good point. When I said

"Got matching answers -> likely to be accurate information"

I should have said

"Got matching answers -> undetermined"

My point was that if LLM is giving you contradictory answers it's definitely confused. If no contradictions ... who knows.