Prune

How-to

How to clear the Hugging Face cache (Mac & Windows)

Transformers and diffusers quietly fill ~/.cache/huggingface. Ten to fifty GB is common. Here is the manual fix, then the one-click option.

Prune scan results showing reclaimable AI and developer caches

1. Find the cache

On Mac and Linux:

du -sh ~/.cache/huggingface
ls ~/.cache/huggingface/hub

On Windows (PowerShell):

Get-ChildItem "$env:USERPROFILE\.cache\huggingface" -Recurse -ErrorAction SilentlyContinue |
  Measure-Object -Property Length -Sum

2. Clear it safely

Deleting the Hub cache does not delete your project code. The next from_pretrained call re-downloads weights. Prefer removing unused snapshot folders under hub/ if you still need a few models offline.

# macOS / Linux
rm -rf ~/.cache/huggingface/hub

# Windows
Remove-Item -Recurse -Force "$env:USERPROFILE\.cache\huggingface\hub"

3. Or let Prune do the review

Prune’s free scan shows the exact gigabytes under Hugging Face and related AI caches. You pick paths. Nothing is deleted blind.

FAQ

Where is the Hugging Face cache?

On both macOS and Windows it is usually under ~/.cache/huggingface (or %USERPROFILE%\.cache\huggingface on Windows). Clearing it is safe; checkpoints re-download when needed.

Will clearing the cache break my training jobs?

Running jobs that already loaded weights into memory keep working. The next cold start will re-download missing files.

Can Prune clear Hugging Face cache for me?

Yes. Prune lists the exact size and paths, then only deletes what you select. Nothing is removed blind.

Related