Developer storage guide

Where local AI models are stored

Q: How do I change where Ollama stores models?

Set the OLLAMA_MODELS environment variable to the directory you want, then restart the Ollama server. On macOS, set it for the app's launch environment, not only in your shell.

Q: Is it safe to delete ~/.cache/huggingface?

Nothing breaks — anything missing is re-downloaded on next use — but that may mean pulling tens of gigabytes again. Use huggingface-cli scan-cache to delete selectively.

Q: Why do the model folders not show up in the Finder?

Most start with a dot, which macOS hides by default. Press Cmd-Shift-period in a Finder window to show them, or use Go then Go to Folder.

Q: Can I keep models on an external drive?

Yes. Ollama uses OLLAMA_MODELS, Hugging Face uses HF_HOME, PyTorch uses TORCH_HOME, and LM Studio has a setting in the app. Expect slower first loads over USB.

Every tool keeps its weights somewhere different, none of them tell you where, and together they are the fastest-growing thing on a developer's disk. Here is the full list, per platform.

Find every model store Buy · $15

14-day free trial, no card needed. Then $15 once for 3 computers, macOS and Linux.

Bytesweep listing local AI model stores by size on a developer Mac

Why no cleaner finds these

Mainstream Mac cleaners were written before local models existed, and they work from a fixed list of known cache folders. ~/.ollama is not on it. So a tool will confidently report your Mac is clean while 120 GB of weights sit in a hidden folder in your home directory — and because they are dot-folders, the Finder does not show them either.

The paths

Ollama

macOS    ~/.ollama/models
Linux    /usr/share/ollama/.ollama/models
Windows  %USERPROFILE%\.ollama\models

List and remove with Ollama itself, so its manifest stays consistent:

ollama list
ollama rm llama3:70b

Move the store to an external drive by setting OLLAMA_MODELS before starting the server.

LM Studio

macOS    ~/.lmstudio/models
Linux    ~/.lmstudio/models
Windows  %USERPROFILE%\.lmstudio\models

Older versions used ~/.cache/lm-studio/models; if you have upgraded through several releases, check both. The directory is configurable in the app's My Models tab, which is the safe way to relocate it.

Hugging Face — transformers, diffusers, datasets

macOS/Linux  ~/.cache/huggingface/hub
Windows      %USERPROFILE%\.cache\huggingface\hub

Usually the largest and the messiest, because every library that touches the Hub writes here. Inspect and prune interactively with the official tool:

pip install "huggingface_hub[cli]"
huggingface-cli scan-cache
huggingface-cli delete-cache

Relocate it with HF_HOME. Note that the hub cache uses symlinks into a blobs folder — copying it with a tool that follows symlinks will double its size.

PyTorch

~/.cache/torch/hub
~/.cache/torch/kernels

Pretrained checkpoints downloaded by torch.hub. Relocate with TORCH_HOME.

Seven paths, and the one you have forgotten

This list is accurate today. It grows every time you try a new tool, and the store you lose track of is the expensive one. Bytesweep finds all of these plus the ones added since, reports them by size, and marks weights for review rather than sweeping — because a model you fine-tuned yourself is not a cache, and nothing else on your disk is that expensive to get wrong.

Start the free 14-day trial Buy · $15

GPT4All

macOS    ~/Library/Application Support/nomic.ai/GPT4All
Linux    ~/.local/share/nomic.ai/GPT4All
Windows  %LOCALAPPDATA%\nomic.ai\GPT4All

Whisper and whisper.cpp

~/.cache/whisper
./models  (inside a whisper.cpp checkout)

Also worth checking

~/.cache/torch/sentence_transformers
~/.keras/models
~/.cache/gguf
~/.cache/vllm

Before you delete a model

A cache is not a weight

Everything else on a developer's disk regenerates in seconds. A 40 GB model is a re-download over your connection, and a quantisation you produced locally or a model you fine-tuned yourself does not come back at all. These folders deserve review, not a sweep.

Delete through the tool where there is one

Ollama and Hugging Face both keep a manifest alongside the blobs. Removing files by hand leaves the index pointing at things that are gone, which shows up later as confusing errors. ollama rm and huggingface-cli delete-cache exist for this.

Move rather than delete

Every tool above supports relocating its store through an environment variable or a setting. If you have an external drive, moving the store beats repeatedly deleting and re-pulling the same weights.

Find every model store on your disk

Bytesweep knows all of these paths, reports them by size next to Docker, Xcode and node_modules, and marks weights as worth reviewing rather than safe to sweep — because a fine-tuned model is not a cache.

Start the free 14-day trial Get Bytesweep Pro · $15

Questions

How do I change where Ollama stores models?

Set the OLLAMA_MODELS environment variable to the directory you want, then restart the Ollama server. On macOS, set it for the app's launch environment rather than only in your shell, or the desktop app will not see it.

Is it safe to delete ~/.cache/huggingface?

Safe in the sense that nothing breaks — anything missing is re-downloaded on next use. Expensive in the sense that "next use" may mean pulling 30 GB again. Use huggingface-cli scan-cache to see what is there and delete selectively.

Why do the model folders not show up in the Finder?

Most of them start with a dot, which macOS hides by default. Press Cmd-Shift-. in a Finder window to show them, or open the path directly with Go → Go to Folder.

Can I keep models on an external drive?

Yes, and it is the better answer once you pass a hundred gigabytes. Ollama uses OLLAMA_MODELS, Hugging Face uses HF_HOME, PyTorch uses TORCH_HOME, and LM Studio has a setting in the app. Expect slower first loads over USB.

Why is my model folder bigger than the models I downloaded?

Usually partial downloads and superseded revisions. The Hugging Face hub cache keeps every revision you have ever pulled of a repository, not just the current one, which is exactly what scan-cache is designed to show you.