post-training

Post-training refers collectively to the training stages that follow pretraining, in which a model learns general abilities from large amounts of data. It shapes model behavior such as following instructions, conversational style, reasoning and safe responses.

Common methods include supervised fine-tuning, reinforcement learning from human preferences and reinforcement learning aimed at improving reasoning. It is also used to derive models for different purposes from the same base model.

This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.