Rufus-Air improves large language model capabilities through staged training
Rufus-Air: An Open LLM Post-Training Recipe
Computation and LanguageArtificial IntelligenceMachine Learning
Summary
Training large AI language models is complicated and requires careful planning. The authors present Rufus-Air, a step-by-step recipe to improve an existing large model by adding skills like reasoning and coding through different training stages. Their approach uses publicly available data and open tools without needing extra human labeling or special teachers. They found that starting with strong basic training and ordering later steps based on how reliable their feedback is helps build better performing AI models. Rufus-Air produces improvements over the original model and competes well with similar open-source models.
What this means in practice
- •For ai development teams: Use Rufus-Air's reproducible pipeline to enhance large language models with diverse skills using open data and no extra human labeling.
- •For software engineering teams: Build AI systems with better reasoning, coding, and search capabilities by following the staged training approach described in Rufus-Air.
Authors
Chia-Yuan Chang, Renyuan Cheng, Rui Feng, Xiaotian Han, Yuan He, Hongye Jin, Linwei Li, Shiyang Li, Fenglin Liu, Xin Liu, Priyanka Nigam, Haoyang Wen, Zhenghao Xu, Zhuocheng Xu, Bing Yin, Qingyu Yin, Chao Zhang, Rongzhi Zhang, Zhihan Zhang, Zixuan Zhang, Zixuan Zhang, Tuo Zhao
Abstract
Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure, stage order, and stagewise results needed to reproduce the recipe. Stages progress from basic to advanced capabilities and from hard, verifiable rewards to softer judge-based signals. Training builds on open-source components and public data, much of it used as released, without new human annotation or an in-house distillation teacher. Our main findings are that (i) diverse, high-quality SFT establishes a strong capability floor; (ii) difficulty filtering keeps RL prompts within a productive learning range; (iii) reward reliability provides a practical principle for ordering stages; and (iv) infrastructure and engineering choices are part of the recipe, not just an implementation detail. Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.