Method merges multiple AI models into one without retraining

Original: Consolidating Rewarded Perturbations for LLM Post-Training

Source: arxiv.org ↗