Hugging Face & Ollama are eating your disk — find and clear the model cache

One 7B model is ~4 GB. Pull a few LLMs, download a handful of Hugging Face checkpoints, let PyTorch cache a couple of vision backbones, and you're suddenly 50–100 GB lighter on disk — most of it models you tried once and forgot.

Where the AI caches actually live (Windows)

• Hugging Face: C:\Users\you\.cache\huggingface\hub (override with HF_HOME). Every model, dataset and tokenizer you've ever loaded via transformers, datasets or huggingface_hub lands here.
• Ollama: C:\Users\you\.ollama\models — the blobs for every model you've ollama pulled.
• PyTorch Hub / torch: C:\Users\you\.cache\torch and %LOCALAPPDATA%\torch.
• pip's download cache: big wheels (CUDA, torch) sit in %LOCALAPPDATA%\pip\Cache.

These are all re-downloadable caches, not your work. Clearing them just means the model downloads again next time you need it.

Clear them the tidy way (keep what you use)

Hugging Face — interactive, size-aware

The official CLI shows every cached revision with its size and lets you select what to delete:

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

Or scan first to see the totals: huggingface-cli scan-cache.

Ollama — remove models you no longer run

ollama list            # see what you have + sizes
ollama rm llama3:70b   # drop a specific model

pip & torch

pip cache purge

Or just sweep the whole drive at once

The CLIs above each cover one tool and assume you know it's installed. SweepDisk finds all of them in a single scan — across every drive — buckets them as regenerable AI/package caches with a safety confidence, and clears the ones you tick to the Recycle Bin. Nothing is uploaded, and nothing is deleted without your click.