Orca: Progressive Learning from Complex Explanation Traces of GPT-4



Orca is a 13-billion parameter model that learns to imitate the reasoning process of large foundation models (LFMs) by using rich signals from GPT-4 including explanation traces, step-by-step thought processes, and other complex instructions. It surpasses conventional state-of-the-art instruction-tuned models and shows competitive performance in professional and academic examinations. [summary] [comments]


Source link