Hugging Face and PyTorch caches can easily devour 50GB+ of SSD space without you even noticing. If you're running local LLMs or fine-tuning models, you know the pain of suddenly hitting a "disk full" error in the middle of a training run.
For anyone tired of manually hunting through hidden folders to find what's eating their storage, this is a much more surgical approach. It's a practical way to maintain your AI workflow without paying for a subscription service that doesn't even understand what a tensor is.
Most general cleaning apps miss these hidden directories because they aren't looking for .cache/huggingface
or specific torch binaries. I've been using a free, open-source alternative to CleanMyMac specifically designed for developers and AI engineers that actually targets these technical footprints on both macOS and Windows.
Instead of just clearing browser cookies, this tool hunts down the heavy weights:
Model Weights: Old versions of LLMs you downloaded once and forgot about.Dataset Caches: Temporary files from data pipelines.Environment Bloat: Residual files from failed conda or pip installs.
For anyone tired of manually hunting through hidden folders to find what's eating their storage, this is a much more surgical approach. It's a practical way to maintain your AI workflow without paying for a subscription service that doesn't even understand what a tensor is.
[Next GitHub Bug Bounty: Quality over Quantity →](/en/threads/3819/)
All Replies （4） #
R
I just moved my HF cache to a secondary HDD to save my boot drive.
0
M
@RayTinkerersmart move. did you notice any speed difference when models from the hdd?
0
Q
Setting the HF_HOME environment variable is the easiest way to redirect those folders permanently.
0
A
Had my OS drive crash last month because of this. Now I use a dedicated partition.
0