How to handle IterableDataset with HuggingFace trainer and num_workers in DDP setup

🤗Datasets
6

But if split by node, the trainer should not skip examples?

So, how do you implement fast processing in DDP?