Where Ollama stores models on macOS

Ollama: ~/.ollama/models

Ollama keeps everything under one folder in your home directory:

~/.ollama/models/
├── blobs/       # the actual model weights, content-addressed by SHA-256
└── manifests/   # one small file per model tag, pointing at its blobs

Models that share layers share blobs, so the folder is usually smaller than the sum of the model sizes Ollama reports. That is also why you should not delete files in blobs/ by hand: a blob can belong to several models, and removing it breaks every manifest that references it.

List and remove models the right way

ollama list            # every model with its size
ollama ps              # what is loaded right now
ollama rm llama3:70b   # delete one model and any blobs only it used

Move the store to another drive

Set OLLAMA_MODELS before the Ollama server starts. For the menu bar app, set it with launchctl so the GUI process sees it:

launchctl setenv OLLAMA_MODELS /Volumes/External/ollama-models

Then quit and relaunch Ollama. Existing models are not moved automatically; copy the models folder across first or pull them again.

LM Studio: ~/.lmstudio/models

LM Studio downloads to ~/.lmstudio/models/<publisher>/<model>/. Older installs used ~/.cache/lm-studio/models, and if you upgraded you may have both. The models directory can be changed and models deleted from the My Models view inside the app, which is the safest route. Deleting a model folder from Finder while LM Studio is closed also works.

Hugging Face: ~/.cache/huggingface/hub

Anything that uses the transformers, diffusers, sentence-transformers or mlx-lm libraries downloads into the Hugging Face hub cache. This is the one most people forget, because no app put it there. On a machine used for a few experiments it is routinely 10–20 GB.

~/.cache/huggingface/
├── hub/        # models--org--name folders with snapshots and blobs
├── xet/        # chunk cache for the newer download protocol
└── datasets/   # downloaded datasets

The official CLI has an interactive cleaner that shows each cached repo with its size and lets you pick what to remove:

pip install -U huggingface_hub
huggingface-cli scan-cache
huggingface-cli delete-cache

To move the cache, set HF_HOME in your shell profile. Everything in it is re-downloaded on the next use, so deleting the whole folder is safe, just slow to recover from.

What is safe to delete

Common questions

Where does Ollama store models on a Mac?

In ~/.ollama/models, with weights in blobs/ and one small manifest per model tag in manifests/. Set OLLAMA_MODELS to move the store.

How do I delete an Ollama model?

Run ollama list to see models and ollama rm <name> to remove one. Do not delete files in blobs/ by hand: a blob can be shared by several models.

Where does Hugging Face cache models?

In ~/.cache/huggingface/hub, with xet/ and datasets/ alongside. Set HF_HOME to move it, or run huggingface-cli delete-cache to remove individual repos.