RSI-Prime 数据集:让 LLM 学会训练其他模型
RSI-Prime: Training Models to Train Models
Z.ai 放出了 RSI-Prime 数据集,350 条实验轨迹专门教 LLM 怎么训练别的模型。亮点是决策级 loss masking,只让模型从高质量决策学习。在 PostTrainBench 上把 GLM-5.3 拉升了 31.95%,做自动化训练 agent 的人值得照着这个思路试试。
Z.ai 团队在 arXiv 发布 RSI-Prime 数据集,目标是提升 LLM 自主开展模型训练的能力,这是递归自我改进(RSI)的基础环节。数据集包含 350 条完整实验轨迹,覆盖七个任务类别,专门针对"模型训练模型"这一场景构建。
方法上的关键在于细粒度的决策标注:团队对每条轨迹中的实验决策进行识别和质量打分,再用决策级 loss masking 把低质量行为从监督信号中剔除,避免模型学到坏习惯。在自主后训练基准 PostTrainBench 上,用 RSI-Prime 训练后的 GLM-5.3 整体性能相对原始版本提升 31.95%。
这项工作把 LLM 的角色从"被训练者"扩展到"训练执行者",为未来模型自主迭代自身能力提供了数据与训练范式层面的路径。对做自动化实验和 agent 训练的团队来说,决策级标注思路可以直接借鉴。
RSI-Prime: Training Models to Train Models
Enabling large language models (LLMs) to autonomously conduct training and improve target models is an important foundation for recursive self-improvement (RSI). As LLMs expand their role from being trained to conducting training, their model-training capability warrants dedicated study and systematic development. To this end, we construct RSI-Prime, a dataset for improving model-training capability, comprising 350 complete experimental trajectories across seven task categories. To help LLMs learn to make high-quality experimental decisions from training practice, we perform fine-grained decision identification and quality annotation, then apply decision-level loss masking to exclude supervision on low-quality behavior. Experiments with GLM-5.3 on PostTrainBench, a benchmark for autonomous post-training, show that training on RSI-Prime yields a 31.95% relative improvement in overall performance over the original GLM-5.3, highlighting a path toward future RSI systems in which models autonomously sustain their own advancement.
This report is categorized under AITOP's "Papers" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include RSI-Prime, GLM-5.3, PostTrainBench, recursive self-improvement.
Source attribution: curated from arXiv cs.AI (@Dewu Zheng, Yang Ye, Hongyu Zhang, Jun Yu, Yanlin Wang).