BAR：基于混合专家的模块化后训练
摘要
扩展后训练语言模型以获取新领域能力的单体训练范式根本受限于单一训练范式：从头重新训练昂贵且规模不佳，而继续训练往往会损害现有能力。我们提出了BAR（Branch-Adapt-Route），它训练独立的领域专家，每个专家通过自己的中期训练、监督微调和强化学习管道进行训练，然后通过混合专家架构进行组合，采用轻量级路由器训练。与需要混合所有领域并为任何更新都需要完整重新处理（费用按平方增长）的重新训练方法不同，BAR enables updating individual experts independently with linear cost scaling and no degradation to existing domains. At the 7B scale, with experts for math, code, tool use, and safety, BAR achieves an overall score of 49.1 (averaged across 7 evaluation categories), matching or exceeding re-training baselines (47.8 without mid-training, 50.5 with). We further show that modular training provides a structural advantage: by isolating each domain, it avoids the catastrophic forgetting that occurs when late-stage RL degrades capabilities from earlier training stages, while significantly reducing the cost and complexity of updating or adding a domain. Together, these results suggest that decoupled, expert-based training is a scalable alternative to monolithic retraining for extending language models.
引用
@article{arxiv.2604.18473,
title = {Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts},
author = {Jacob Morrison and Sanjay Adhikesaven and Akshita Bhagia and Matei Zaharia and Noah A. Smith and Sewon Min},
journal= {arXiv preprint arXiv:2604.18473},
year = {2026}
}
备注
9 content pages, 23 pages overall, 3 figures