5-4B, as decision models.
Open language models match or beat a proprietary decision model without specialized training
Reading output logits over answer letters and a brief LoRA fine-tuning recipe lets open models outperform Jev on decision benchmarks.
Top University
Jonathan Geuter · Andrew Cheng · David Alvarez-Melis
Harvard University · Kempner Institute
Research Digest··2 min read
Thread:Decision models (System One)
Geuter et al.
§