ConceptAI & ML
KL divergence penalty in RL
A regularization term added to the RL objective to prevent the fine-tuned policy from drifting too far from the reference SFT model where the reward model is accurate.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Lecturers
1Connections
3The lectures that define it or assume it, and the concepts next to it.