When using agent-task datasets to enhance agent capabilities for Large Language Models (LLMs), current methodologies often treat all tokens within a sample equally. However, we argue that tokens serving different roles - specifically, reasoning tokens versus boilerplate tokens (e.g., those governing output format) - differ significantly in importance and learning complexity, necessitating their disentanglement and distinct treatment. To address this, we propose a novel Shuffle-Aware Discriminator (SHAD) for adaptive token discrimination. SHAD classifies tokens by exploiting predictability differences observed after shuffling input-output combinations across samples: boilerplate tokens, due to their repetitive nature among samples, maintain predictability, whereas reasoning tokens do not. Using SHAD, we propose the Reasoning-highlighted Fine-Tuning (RFT) method, which adaptively emphasizes reasoning tokens during fine-tuning, yielding notable performance gains over common Supervised Fine-Tuning (SFT).
Reflection Steering: Disentangling Reflection from Reasoning in Activation Space for Token-Efficient Inference
School of Computing and Data Science, The University of Hong Kong(香港大学计算与数据科学学院);Department of Computing, The Hong Kong Polytechnic University(香港理工大学计算学院);School of Data Science, Lingnan University(岭南大学数据科学学院);Centre for Learning, Teaching and Technology, The Education University of Hong Kong(香港教育大学学习、教学与技术中心)
Disentangling Mathematical Reasoning in LLMs: A Methodological Investigation of Internal Mechanisms
German Research Center for Artificial Intelligence (DFKI)(德国人工智能研究中心);Saarland University(萨尔兰大学);Center for European Research in Trusted AI (CERTAIN)(可信人工智能欧洲研究中心)
Fine PT-PT Web: A High-Quality 41 Billion Tokens Data Collection of the European Portuguese Web
NOVA School of Science and Technology(NOVA科学与技术学院);NOVA LINCS;Fundação para a Ciência e Tecnologia(科学技术基金会);Instituto Superior Técnico, Universidade de Lisboa(里斯本大学高等技术学院);Instituto de Telecomunicações(电信研究院)
Entropy Gate: Entropy Quenching for Near-Lossless Token Compression in LLM Pipelines
VIA Cybersecurity Lab, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana(维亚网络安全实验室,卡布雷姆大学科学与技术学院,库马西,加纳);Quantum and Assistive Technologies Lab, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana(量子与辅助技术实验室,卡布雷姆大学科学与技术学院,库马西,加纳)
Scaling Multi-Hop Training Data via Graph-Constrained Path Selection
The Hong Kong University of Science and Technology(香港科学与技术大学);Hong Kong Generative AI Research and Development Center(香港生成式人工智能研究与发展中心);Hong Kong Baptist University(香港 Baptist 大学)
CoTJudger: A Graph-Driven Framework for Automatic Evaluation of Chain-of-Thought Efficiency and Redundancy in LRMs
University of Science and Technology of China(科学技术大学);Shenzhen University of Advanced Technology(深圳先进技术大学);Shenzhen Institutes of Advanced Technology, CAS(深圳先进技术研究所,中国科学院);Southeast University(东南大学);Nanjing University(南京大学);Beihang University(北航);University of Manchester(曼彻斯特大学)