Post-training
Training that comes after pre-training, to turn a base model into one that follows instructions and behaves as intended.
Qwen’s technical report says that post-training began with over a million supervised fine-tuning examples, followed by two preference-based stages: training on about 150,000 preference pairs, then an online method called GRPO.
See it in action
Try post-training hands-on in the lesson Instruction tuning and RLHF, part of the How AI assistants work course.