Emerging Tech Insider
3 minutes
Fine-tuning a pretrained language model on a diverse collection of (instruction, response) pairs so that it follows natural-language instructions across many tasks, the supervised stage that precedes preference-based alignment.