Fit generator throws ValueError: Sample larger than population or is negative whenever I try to re-build same network created previously

Question

Whenever I try to replicate neural networks code on kaggle (i.e. https://www.kaggle.com/code/mohamedmzoughi/vggface-cv-focal-loss/edit
https://www.kaggle.com/code/janpreets/just-another-feature-extractor-0-824-lb/notebookhttps://www.kaggle.com/code/hsinwenchang/vggface-baseline-197x197/notebook)

fit_generator throws :

ValueError: Sample larger than population or is negative

and I don't know how to solve it.

I tried looking for solution yet all of them relates to random module.