Is it safe to delete Hugging Face cache?
Models and datasets downloaded from Hugging Face, shared by every library and tool that uses them.
Review first
It's your own data, or hard to recreate. Decide item by item.
~/
What it is
Python libraries such as Transformers and Diffusers, Hugging Face's hf command-line tool, and apps such as llama.cpp download models and datasets from the Hugging Face Hub into this cache, in a folder for each repository. New revisions of a model are stored next to the old ones, and nothing is removed automatically. Files download again when a program asks for them, but some models require you to sign in or accept a license first.
If it's removed
Programs download the models and datasets they need again, which can take a long time. Gated models need you to be signed in to Hugging Face.
How it's removed
Moved to the Trash, where you can restore it.
The free analyzer finds it and explains it. Removing it is Pro, after a review sheet that lists every move and command. It's never checked for you: it waits for your choice.
Sources
- huggingface.co/docs/huggingface_hub/guides/manage-cache
- huggingface.co/docs/huggingface_hub/package_reference/environment_variables
Last checked on 26 Sep 2026. Every entry is written and checked by hand; the app and the command-line tool use the same words.