The Hugging Face cache is filling my drive

Short answer: models downloaded with transformers, diffusers or huggingface_hub go to %USERPROFILE%\.cache\huggingface\hub. When a model is updated, the old revision often stays next to the new one, so the same model can take space twice.

By hand

1. See what is there: hf cache scan (older versions: huggingface-cli scan-cache).
2. Remove models or old revisions, choosing from a list: huggingface-cli delete-cache
3. To keep the cache on another drive, set HF_HOME to a folder there.

With Sparewise

• Lists each Hugging Face model with its size and last use, next to Ollama and LM Studio.
• Shows old revisions you no longer need, and how much they take.
• Gives the command to remove each. Models are slow to download again, so it leaves the choice to you.

Sparewise does this safely

Download for Windows

Free · No admin rights · No telemetry · Everything it moves can be put back