passdrill
AI & LLM Engineering · How LLMs Work: Transformers & Training · Card 007/012 medium

Ouyang et al. (2022), "Training Language Models to Follow Instructions with Human Feedback" (the InstructGPT paper), describes a three-stage pipeline for aligning a pretrained language model with human preferences. What are the three stages, in order?

  1. Pretrain on a labeled classification dataset, then deploy directly without further tuning, since pretraining alone is assumed to capture human preferences
  2. Supervised fine-tuning on human-written demonstrations, followed by training a reward model on human rankings of sampled model outputs, followed by reinforcement learning (PPO) that optimizes the policy against that reward model
  3. Train a reward model on raw internet text first, then use it to filter the pretraining corpus before any supervised fine-tuning occurs
  4. Fine-tune the model using only reinforcement learning against a fixed rule-based reward function, without using any human-labeled data at any stage
Next card → Shuffle