huggingface / huggingface/datasets
Support saving datasets with custom formatting
Open
- Dominant language
- Python
- Stars
- 22k
- Forks
- 3.4k
- Avg merge
- 5d 7h
- Merged PRs (30d)
- 17
Description
Requested in https://discuss.huggingface.co/t/using-set-transform-on-a-dataset-leads-to-an-exception/53036.
I am not sure if supporting this is the best idea for the following reasons:
>For this to work, we would have to pickle a custom transform, which means the transform and the objects it references need to be serializable. Also, deserializing these bytes would make `load_from_disk` unsafe, so I'm not sure this is a good idea.
@lhoestq WDYT?
Contributor guide
Assessment
This issue has not been assessed yet.