#post-training

From RLHF to GRPO/DAPO/RLVR: How Verifiable-Reward RL Replaced Human Preference Labels