How to find and delete downloaded AI models on a Mac
Local models are easy to download and easy to forget. A single model is often several gigabytes, and the tools that fetch them keep the files in hidden folders you never look at.
Where the files live
| Tool | Location |
|---|---|
| Hugging Face | ~/.cache/huggingface/hub |
| LM Studio | ~/.lmstudio/models (older versions: ~/.cache/lm-studio/models) |
| Ollama | ~/.ollama/models |
| PyTorch | ~/.cache/torch |
| Whisper | ~/.cache/whisper |
| Keras | ~/.keras |
| GPT4All | ~/Library/Application Support/nomic.ai/GPT4All |
Most of these start with a dot, so Finder hides them. Press Command, Shift and period in a Finder window to show hidden files, or use Go to Folder with Command, Shift and G.
See what is big
du -sh ~/.cache/huggingface/hub/* | sort -hEach Hugging Face repository gets its own folder named like models--org--name. Deleting that folder removes the model. A script that needs it later will download it again.
Remove models from Ollama
ollama list
ollama rm <model>Do not delete single files inside ~/.ollama/models by hand. Its blobs are shared between models, so removing the wrong one can break another model. Let ollama rm do it.
Remove models from LM Studio
Open My Models inside the app and delete from there, or delete a model's folder under the models directory.
Before you delete
Nothing here is lost for good, but a large model can take a long time to download again, and some gated models need you to be signed in to the provider. Delete what you know you will not use, and keep the one you reach for every week.
Doing it with Vervain
The AI Models module lists these downloads with their sizes: Hugging Face per repository, LM Studio per model, and PyTorch, Whisper, Keras and GPT4All as their own entries. Ollama shows up as a single item, because its files are shared, and the module tells you to use ollama rm for individual models. Nothing is ticked by default, since downloading a model again can take so long. What you remove goes to the Trash.