Hello, thanks for the lib!
Something I haven't quite figured out is how to handle parquet dataset splits. Hugging face is notorious for these, splitting those big dataset is ~150Mb parquet files each, eg: https://huggingface.co/datasets/Lichess/chess-puzzles/tree/main/data.
However I haven't found if there's a way to merge handle those separate files as one dataset with hyperparquet.