How do I add custom metadata fields to datasets?

Basically, I am using Dataset.from_dict() to create datasets, but I want to save two strings as global metadata, without making them column headers or making them repeated as values in every row. What would be the correct approach to this? A naive idea would be to encode them into the “description” field of DatasetInfo, but that seems like it can’t be intended.

1 Like

Storing arbitrary metadata is currently not supported but should eventually be implemented as part of Metadata for each column · Issue #5575 · huggingface/datasets · GitHub.