How to train the react model on retrieval dataset

I have a question.

After retrieving data using React retrieval at /react_retrieval, how can I proceed with training the dataset retrieved at /react_customization?

 Do I need to create a separate web dataset? When I look at the ImageNet10M data at /react_customization, it downloads a parquet file and then dataset with img2dataset. How was the parquet file created?  

Thank you.