Hugging Face Hub Model Cache Folder (~/.cache/huggingface/hub)

Bottom Line: It is 100% safe to clear the Hugging Face Hub model cache. Running huggingface-cli delete-cache will NOT delete your Python machine learning scripts, datasets, or fine-tuned checkpoints.

What is the Hugging Face Hub Cache Folder?

Located at ~/.cache/huggingface/hub on Linux/macOS (and %USERPROFILE%\.cache\huggingface\hub on Windows), this folder is managed by the Hugging Face transformers and diffusers libraries.

When downloading open-source models from Hugging Face (BERT, RoBERTa, Stable Diffusion, Llama, Falcon), Hugging Face uses a symlinked cache architecture inside models--[author]--[repo]. Over time, duplicate model commits and unlinked snapshot revisions hoard 20 GB to 100 GB+ of drive space.

Can You Delete Hugging Face Hub Cache?

✅ Yes, it is completely safe to delete.

• No code risk: Your Python scripts, fine-tuning scripts, and training code stay 100% safe.
• Auto-downloaded: Calling AutoModel.from_pretrained("...") will re-fetch required model weights automatically.

How to Safely Clean Up Hugging Face Cache

Method 1: Using Hugging Face CLI Interactive Cache Cleaner (Recommended)

1. Open your terminal.
2. Launch the interactive cache manager:

   huggingface-cli delete-cache

3. Select older model revisions to delete and press Enter.

Method 2: Manual Directory Wipe

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

# Windows (PowerShell)
Remove-Item -Recurse -Force $env:USERPROFILE\.cache\huggingface\hub\*

Related Guides

• Learn about deleting Hugging Face cache.
• How to clean up PyTorch model cache.
• Check if you can clean Ollama models cache.