Where does Hugging Face store downloaded models?

In ~/.cache/huggingface. Use 'huggingface-cli scan-cache' for a per-model breakdown and 'huggingface-cli delete-cache' for interactive removal. Models re-download on the next from_pretrained() call.

How do I delete Ollama models?

List them with 'ollama list' and remove with 'ollama rm '. The files live in ~/.ollama/models.

← All guides

Where AI models hide on your Mac

CacheCleaner guides · Updated July 18, 2026

Every model you ever tried is still on disk. A couple of 7B experiments here, one 70B download there – AI tooling quietly stacks up to 100+ GB across four or five separate cache locations.

Hugging Face

du -sh ~/.cache/huggingface
huggingface-cli scan-cache        # per-model breakdown
huggingface-cli delete-cache      # interactive removal

Everything re-downloads on the next from_pretrained().

Ollama

ollama list                       # models with sizes
ollama rm llama3:70b              # delete one
du -sh ~/.ollama/models

LM Studio

du -sh ~/.lmstudio/models         # or ~/.cache/lm-studio on older versions

Delete models from the app's My Models tab, or remove the folders directly.

PyTorch, pip & friends

~/.cache/torch                    # torch hub checkpoints
~/Library/Caches/pip              # wheel cache – pip cache purge
~/.cache/whisper                  # whisper checkpoints

Claude Code & other CLI agents

Agent CLIs keep session logs and caches too – smaller, but they add up on heavy use: ~/.claude, ~/.codex.

CacheCleaner finds all of these in one scan – Hugging Face, Ollama and LM Studio models are listed per-location and marked ⚠️, since models are downloads you chose, not disposable caches. You decide what goes.

Get CacheCleaner for Mac