How do I generate batches of data using Keras from huge Parquet part files

Question

I have a dataset of 100 Parquet files(each of size ~10 GB) in a directory, on which I want to train a DSSM based model

Now, given the huge amount of data, I am confused a