Model Downloads Keep Failing or Coming Out Corrupted

A stalled pull: just re-run the same command

If a download stalls partway through - the percentage stops climbing for several minutes - the fix is simply to run the exact same pull command again:

ollama pull llama3.2

Ollama resumes from where it left off rather than starting over from zero. This covers the large majority of "download failed" situations, and it's always worth trying first before anything more drastic.

The model downloaded, but behaves strangely

This is a different, less common problem: the pull reported success, but the model gives garbage output, crashes on load, or behaves nothing like it should. This usually means the download completed but the file itself is corrupted - a dropped connection mid-transfer that didn't get caught cleanly. The fix is to remove it and pull a clean copy rather than trying to repair it:

ollama rm llama3.2
ollama pull llama3.2

Check disk space before re-pulling

A model that silently runs out of disk space mid-download can leave behind a partial file without a clear error message. Before spending time re-pulling a large model, confirm there's actually room for it:

# Mac/Linux
df -h

On Windows, check available space on the drive from File Explorer. If you're running consistently close to full, the storage guide covers sizing this properly so it doesn't keep happening.