Duplicate models across Ollama and LM Studio: stop storing them twice

Part of: Local AI models on Mac: where Ollama, LM Studio and Hugging Face store them (AI models)

Yes, the same model can be on your Mac twice. Ollama and LM Studio keep their models in separate folders, and neither tool’s documentation says they share files. If you pulled a model in both, you are paying for it in both places. Compare the two folders, keep the copy you actually use, and remove the other through its own tool.

Where each tool stores models

• Ollama: ~/.ollama/models, according to the Ollama FAQ.
• LM Studio: ~/.lmstudio/models/, according to LM Studio’s documentation, arranged as publisher, then model, then file.

These are separate folders. See the overview for more locations, including the Hugging Face cache, which a third tool may fill with the same weights.

To see how big each is, run:

du -sh ~/.ollama/models ~/.lmstudio/models

du is read-only. I could not run it against real installs, because neither tool is installed on the Mac I used for this article, so the paths come from the documentation.

Spot the duplicates

1. In Ollama, list your models with its list command (see the Ollama deletion guide).
2. In LM Studio, open My Models to see its list.
3. Look for the same family and size in both: the model name, the parameter count and the quantization label.

A caution: models are often published under different names for different tools, and the same weights may be repackaged. Matching names are a strong hint. Different names do not prove the models are different. Compare sizes before you decide.

I could not confirm how the two tools relate when you import a file. Ollama’s import documentation describes creating a model from a GGUF file with a Modelfile, but it does not say whether the original file is copied or can be removed afterwards. So I cannot tell you that the original is safe to delete after an import. Check the folder sizes before and after, or ask in Ollama’s own community.

Which copy to keep

Pick by how you work:

• You mostly use one tool: keep the model there and remove it from the other.
• You switch between both: you need a copy in each. Consider whether you need both tools.
• You only tried a model once: remove it from both.

Remove each copy with its own tool, not in Finder. LM Studio’s documentation says you manage your models in the app under My Models; I could not confirm a delete step from the docs I read, so look for it in the app itself.

Safe, risky, never

Avoid creating duplicates next time

Before you download a model in a second tool, look at what the first one already has. If you want to try the same model in both, decide in advance which one stays. Remove the loser as soon as you pick, while you still remember where it came from.

What it costs to get a model back

A duplicate is only a duplicate if you can get it again. Before you remove one, check the size in the tool’s listing so you know the download you would face. If your connection is slow or metered, keep the copy in the tool you use most.

After you remove one copy

Open the tool you kept and load the model once, to be sure it works. Then measure the folder you cleaned with du -sh, and compare it with the number from before. If you changed your mind, the cost is one download into that tool.

Also check the Hugging Face cache

If you download models through Python tools, the Hugging Face cache can hold a third copy of the same weights. It has its own commands for listing and removing what is in it.

How DiskJoy fits in

DiskJoy is our product. It is pre-alpha: its scanner is designed to find duplicate files and show them for review, so two large copies of the same model would be candidates to look at. Cleanup is not built yet, so it does not remove anything. The download page shows the current status of the scan.