Post-training glossary
Definitions with the equation, the failure mode, and a link to where we tested it.
Terms
Direct Preference Optimization (DPO)
A preference-learning method that trains directly on chosen/rejected response pairs, optimising the RLHF objective in closed form.
Read →GlossaryGroup Relative Policy Optimization (GRPO)
A reinforcement-learning method that scores a group of sampled completions per prompt and uses their mean reward as the baseline, removing PPO's value network.
Read →GlossaryLoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning method that freezes the base weights and learns a low-rank update ΔW = BA for selected layers.
Read →GlossaryReward hacking
When a policy learns to score highly on its reward function by exploiting gaps in it rather than by solving the intended task.
Read →GlossaryKL divergence penalty
A regulariser that penalises the trained policy for moving away from a reference model, measured by KL divergence.
Read →GlossaryLLM-as-a-judge
Using a language model, prompted with a rubric, to score or compare outputs from another model.
Read →GlossaryCatastrophic forgetting
The degradation of a model's previously learned abilities after training on new, narrower data.
Read →Bring a model and a metric.
Tell us what the model should get better at. A post-training engineer replies within one business day with a plan, a compute estimate and an eval design.