Tuft bench
How good is a small open model at judging a nature photo? Find out on your own photos. Nothing leaves this page.
1. Add photos
Aim for 30 or more. For each quest, take photos that really complete it and photos that look similar but do not. Label honestly, including the awkward ones.
2. Run the judges
CLIP downloads about 150 MB the first time, then works offline.
Optional vision model through Ollama:
3. Results
Run a judge to see results.