one-sample-polymath-learning

【设计与创意】Demonstrate that a single strategically engineered training sample can improve reasoning across multiple domains. Polymath learning shows sample quality and multidisciplinary design matter more than quantity, enabling extreme data efficiency in RL training.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准