Language models often lack grounded reasoning capabilities in specialized domains where training data is scarce but bespoke systems excel. We introduce a general framework for distilling expert system reasoning into natural language chain-of-thought explanations, enabling compact models to acquire domain expertise and the ability to generate faithful, grounded explanations. Rather than distilling only final outputs, we capture the full reasoning process, transforming opaque expert computations into transparent, step-by-step explanations. We demonstrate this approach in chess, a canonical reasoning domain where language models continue to underperform. Our 4B parameter model, C1, advances from a near-zero baseline to 48.1\% accuracy, outperforming all open-source models and most frontier proprietary systems. Notably, C1 surpasses its distillation teacher and generates solutions in two orders of magnitude fewer tokens than baselines. Unlike prior neural chess approaches that predict only best moves, C1 generates explainable solutions revealing strategic reasoning. Our pipeline combines supervised fine-tuning and reinforcement learning with theme-balanced data sampling for comprehensive tactical coverage. Master Distillation demonstrates how to inject expert-level knowledge into compact models for under-optimized domains, offering a recipe for unlocking RLVR where LLMs lack sufficient base capabilities.
The Chess Transformer: Mastering Play using Generative Language Models
PeopleTec;Auburn University(奥本大学)
这项工作表明,自然语言Transformers可以支持更通用的战略建模,特别是对文本存档的游戏。除了学习自然语言技能之外,抽象的Transformer架构还可以在棋盘上生成有意义的移动。通过进一步的微调,Transformer通过在Portable Game Notation中对280万盘国际象棋进行训练来学习复杂的游戏玩法。经过30,000个训练步骤后,OpenAI的生成式预训练Transformer(GPT-2)优化了7.74亿个参数的权重。这款经过微调的国际象棋Transformer可以生成合理的策略,并显示可识别为经典开局的游戏阵型,例如英语或斯拉夫交换。最后,在现场比赛中,新模型展示了一个人对Transformer的界面,正确过滤非法移动,并提供了一种新的方法来挑战Transformer的象棋策略。我们预计未来的工作将建立在这个Transformer的承诺,特别是在其他战略游戏中,功能可以捕获简单但富有表现力的玩家注释的基础复杂的规则语法。
Hán Dān Xué Bù (Mimicry) or Qīng Chū Yú Lán (Mastery)? A Cognitive Perspective on Reasoning Distillation in Large Language Models
School of Philosophy, Anhui University(安徽大学哲学学院);Institute of Neuroscience, Chinese Academy of Sciences(中国科学院神经科学研究所);Faculty of Education, University of Cambridge(剑桥大学教育学院);School of Foreign Studies, South China Normal University(华南师范大学外语学院);Division of Psychology and Language Sciences, University College London(伦敦大学学院心理学与语言科学系)
Open-vocabulary Object Detection via Vision and Language Knowledge Distillation
Google Research(谷歌研究院);Nvidia(英伟达)
我们的目标是推进开放词汇目标检测,它检测任意文本输入描述的对象。根本的挑战是能否获得训练数据。进一步扩大现有目标检测数据集中包含的类的数量是昂贵的。为了克服这一挑战,我们提出了ViLD,一种通过视觉和语言知识蒸馏的训练方法。我们的方法将知识从预先训练的开放词汇图像分类模型(教师)中提取到两阶段检测器(学生)中。具体来说,我们使用教师模型来编码类别文本和对象提案的图像区域。然后,我们训练一个学生检测器,其检测到的盒子的区域嵌入与教师推断的文本和图像嵌入对齐。我们在LVIS上进行基准测试,将所有罕见类别作为训练过程中看不到的新类别。ViLD使用ResNet-50骨干网获得16.1掩码AP $_r$,甚至比有监督的对手高出3.8。当使用更强的教师模型ALIGN训练时,ViLD达到26.3 AP $_r$。该模型可以直接移植到其他数据集,无需微调,在PASCAL VOC上达到72.2 AP $_{50}$,在COCO上达到36.6 AP,在Objects 365上达到11.8 AP。在COCO上,ViLD在新型AP上的表现优于先前的最先进水平4.8,在整体AP上优于先前的最先进水平11.4。代码和演示在https://github.com/tensorflow/tpu/tree/master/models/official/detection/projects/vild上开源。