How to delete Ollama and Hugging Face models on a Mac

Running language models locally is great, until your disk is full of weights you tried once. A single model can be several gigabytes, and quantised variants of the same model add up.

Ollama

List what you have, then remove models by name:

ollama list
ollama rm <model-name>

Models live in ~/.ollama/models. Using ollama rm is better than deleting files, because it keeps the model index consistent.

Hugging Face cache

Libraries such as transformers download models to ~/.cache/huggingface/hub. The Hugging Face command-line tool has an interactive cleaner:

huggingface-cli delete-cache

Newer versions of the CLI may provide the same under a different command, so check its help if that one isn’t found. You can also delete a model’s folder (named models--<org>--<name>) by hand; it will be downloaded again when needed.

LM Studio

LM Studio keeps its models in a folder shown in its My Models tab. Delete models from inside the app.

Check before you delete

A model you downloaded to fine-tune, or one without a public source, may be hard to get back. Models from public hubs can simply be downloaded again, so they’re the safest to remove when you’re short on space. Make sure the machine isn’t in the middle of a download or a running job.

How Gardner treats model stores

Gardner finds Ollama, LM Studio and Hugging Face model stores as part of the Clutter tool and always marks them “your call”, never ticked, because weights are expensive to download again. You choose which models to move to the Trash.