Reward Model

Reward model in RLHF: what does it score, and on what data does it train?

Preference pair: name the two pieces and what the label expresses.

Where do SFT and DPO sit in the classical RLHF pipeline?

DPO vs RLHF: pick the most accurate operational difference

PPO vs DPO, what's the practical difference?

What is RLHF, and why is it used after pretraining?