The short answer
Usually, after a check. Might hold something you care about. Look inside before removing it.
What is it?
The Hugging Face datasets library converts downloaded data into Arrow files and stores them, along with processed versions (after map, filter and similar steps), in ~/.cache/huggingface/datasets. This processed copy can be larger than the original download.
Datasets converted by the datasets library. Rebuilt from the hub when a script loads them again.
What happens if you delete it?
The datasets library rebuilds the Arrow files from the hub cache, or downloads the data again, the next time a script calls load_dataset. Expensive processing steps are recomputed.
How to remove it properly
Quit any notebook or script using datasets, then move the folder to the Trash in Finder (Go › Go to Folder… ~/.cache/huggingface/datasets), or remove only the dataset subfolders you no longer use.
Quit the app that uses it, then move it to the Trash. Empty the Trash when you’re sure.
Before you delete it
Datasets saved locally with save_to_disk elsewhere are not affected, but long preprocessing results cached here will need to be recomputed.
How MoreRoom helps
MoreRoom recognises this item and labels it “Review first”: it shows what’s inside, its size and last use, so you can decide. Nothing is removed without your say, and everything goes to the Trash first, with Undo.
Frequently asked questions
Can I delete ~/.cache/huggingface/datasets on Mac?
Usually, after a check. Might hold something you care about. Look inside before removing it. The datasets library rebuilds the Arrow files from the hub cache, or downloads the data again, the next time a script calls load_dataset. Expensive processing steps are recomputed.
Will Hugging Face processed datasets come back after I delete it?
Yes. The datasets library rebuilds the Arrow files from the hub cache, or downloads the data again, the next time a script calls load_dataset. Expensive processing steps are recomputed.
How do I remove Hugging Face processed datasets properly?
Quit any notebook or script using datasets, then move the folder to the Trash in Finder (Go › Go to Folder… ~/.cache/huggingface/datasets), or remove only the dataset subfolders you no longer use.