- Explain that the expressions used for splitting shards into train/val during prepare are regexes - Explain how the seed offset can be used to get new random orders, and that by default the order is always the same - Mention SkipSample and explain when to use it - Update list of sample types in `basic/data_prep`. Some are missing. - Add little how to on automated dataset preparation (non-interactive) - Metadataset options not fully documented (e.g. `dataset_config`) - Difference between `get_val_dataset` and `get_val_datasets` and also fix their docstrings
