Zephyr: Direct Distillation of LM Alignment: Method | HackerNoonThe goal is to align a large-language model to the user's intent using distilled supervised fine-tuning (dSFT).