Post-training

Training or adaptation performed after broad pretraining to change a model's task performance, instruction following, preferences, safety behavior, or specialization.

Post-training can include supervised fine-tuning, preference optimization, reinforcement learning, distillation, verifiable rewards, and safety-specific data or objectives. Pipelines differ by model developer and may repeat or combine stages. The term describes when the work occurs, not one algorithm or a guarantee that the result is a chat assistant.

Builder example

Models sharing a base can behave differently after post-training, but an outside observer usually cannot attribute one refusal, tone, or formatting failure to a specific stage without evidence. Select models by measured behavior and documented controls rather than a story about their training recipe.

Common confusion: Post-training is an umbrella term covering fine-tuning, reinforcement learning from human feedback (RLHF), safety training, and more. Fine-tuning is one technique within post-training. The two terms are not synonyms.