Machine Learning Caches
Manage Hugging Face Cache on macOS
The transformers, diffusers, and datasets Python packages cache model snapshots in ~/.cache/huggingface, frequently consuming 30 GB to 80 GB without warning.
Understanding the Hugging Face Hub Cache Structure
Hugging Face uses a content-addressable storage structure with blobs and symlink snapshots. Deleting individual files manually can break existing Python scripts. DiskWarren identifies complete abandoned revisions so they can be recycled safely.
Using the Hugging Face CLI Cache Tool
Hugging Face Cache CLI
# Inspect and manage Hugging Face cache interactively
huggingface-cli delete-cache