Where a data scientist's disk goes: environments, datasets and models
Python environments, notebooks, datasets and downloaded models each grow in their own folder. A map of every one, typical sizes, and what to clear first.
Data work stores three kinds of thing, and each is large for a different reason: environments because every project installs its own scientific stack, datasets because they are data, and models because they are downloaded weights.
The map
| What | Typical size | Where | Guide |
|---|---|---|---|
| conda environments | 1 to 5 GB each | ~/miniconda3/envs, ~/anaconda3/envs | conda |
| pip and uv caches | 1 to 10 GB | ~/Library/Caches/pip, ~/.cache/uv | uv |
| Hugging Face models and datasets | GBs per model | ~/.cache/huggingface | ML caches |
| Local LLMs | 4 to 40 GB each | ~/.ollama, LM Studio | local LLMs |
| Datasets and notebooks | varies | project folders | Jupyter and datasets |
| R packages | 1 to 3 GB | ~/Library/R | R and RStudio |
The biggest wins, in order
- Models you no longer use. One unused model can be larger than everything else here.
- Environments for finished projects. Export with
conda env exportfirst, then remove; the file recreates it. - Package caches.
conda clean --all,pip cache purge,uv cache prune. - Intermediate data in notebooks: checkpoints, cached downloads and saved outputs.
du -sh ~/.cache/huggingface ~/.ollama ~/miniconda3/envs/* ~/anaconda3/envs/* 2>/dev/null | sort -h
conda env list 2>/dev/null
conda clean --all --dry-run 2>/dev/nullKeep data off the internal disk
Datasets are the one category where an external SSD is the right home by default. They are large, read in bulk, and do not need to move with you. Pointing HF_HOME at an external drive does the same for model downloads. See external SSDs for developers.
Common questions
What takes the most space for data science on a Mac?
Downloaded models and conda environments, then datasets and package caches.
Is it safe to delete a conda environment?
Yes, if you can recreate it. Export it first with conda env export, then remove it with conda env remove.
Where does Hugging Face store models on a Mac?
In ~/.cache/huggingface by default. Setting HF_HOME moves it, for example to an external drive.
How do I clear conda's package cache?
Run conda clean --all. Add --dry-run first to see what it would remove.