大模型推理能力

The Energy Cost of Reasoning: Analyzing Energy Usage in LLMs with Test-time Compute

Yunho Jin, Gu-Yeon Wei, David Brooks

机构 * Harvard University（哈佛大学） 纠错

专题命中 测试时计算 ：reasoning(title,abstract)；test-time compute(title,abstract)；分类 cs.AI、cs.LG

Rank-1 LoRAs Encode Interpretable Reasoning Signals

Jake Ward, Paul Riechers, Adam Shai

机构 * MATS 纠错

专题命中 测试时计算 ：reasoning(title,abstract)；分类 cs.AI、cs.LG

Comments 39th Conference on Neural Information Processing Systems (NeurIPS 2025) Workshop: Mechanistic Interpretability Workshop

R2-KG: General-Purpose Dual-Agent Framework for Reliable Reasoning on Knowledge Graphs

Sumin Jo, Junseong Choi, Jiho Kim, Edward Choi

机构 * KAIST（韩国科学技术院） 纠错

专题命中 测试时计算 ：reasoning(title,abstract)；分类 cs.CL、cs.AI

Comments Accepted to IJCNLP-AACL 2025 Findings

Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute

Jianhao Chen, Zishuo Xun, Bocheng Zhou, Han Qi, Hangfan Zhang, Qiaosheng Zhang, Yang Chen, Wei Hu, Yuzhong Qu, Wanli Ouyang, Shuyue Hu

机构 * State Key Laboratory for Novel Software Technology（新型软件技术国家重点实验室） ； Shanghai Artificial Intelligence Laboratory（上海人工智能实验室） ； The University of Auckland（奥克兰大学） ； The Pennsylvania State University（宾夕法尼亚州立大学） 纠错

专题命中 测试时计算 ：test-time compute(title,abstract)；分类 cs.AI

Journal ref AAAI2026

Revisiting the Data Sampling in Multimodal Post-training from a Difficulty-Distinguish View

Jianyu Qi, Ding Zou, Wenrui Yan, Rui Ma, Jiaxu Li, Zhijie Zheng, Zhiguo Yang, Rongchang Zhao

专题命中 测试时计算 ：reasoning(abstract)；chain-of-thought(abstract)；CoT(abstract)；分类 cs.CL、cs.AI

Comments Accpeted by AAAI 2026

Lookahead Unmasking Elicits Accurate Decoding in Diffusion Language Models

Sanghyun Lee, Seungryong Kim, Jongho Park, Dongmin Park

机构 * KAIST（韩国科学技术院） ； University of California, Berkeley（加州大学伯克利分校） ； KRAFTON 纠错

专题命中 测试时计算 ：planning(abstract)；test-time compute(abstract)；verifier(abstract)；分类 cs.AI、cs.LG

Teaching Pretrained Language Models to Think Deeper with Retrofitted Recurrence

Sean McLeish, Ang Li, John Kirchenbauer, Dayal Singh Kalra, Brian R. Bartoldson, Bhavya Kailkhura, Avi Schwarzschild, Jonas Geiping, Tom Goldstein, Micah Goldblum

机构 * University of Maryland（马里兰大学） ； New York University（纽约大学） ； Lawrence Livermore National Laboratory（劳伦斯利弗莫尔国家实验室） ； University of North Carolina（北卡罗来纳大学） ； ELLIS Institute Tübingen, Max Planck Institute for Intelligent Systems, Tübingen AI Center（图宾根ELLIS研究所、马克斯·普朗克智能系统研究所、图宾根人工智能中心） ； Columbia University（哥伦比亚大学） 纠错

专题命中 测试时计算 ：test-time compute(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments code: https://github.com/mcleish7/retrofitting-recurrence, models: https://huggingface.co/collections/tomg-group-umd/retrofitting-recurrence

REINFORCE++: Stabilizing Critic-Free Policy Optimization with Global Advantage Normalization

Jian Hu, Jason Klein Liu, Haotian Xu, Wei Shen

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.CL、cs.LG

Comments refactor

CoPRIS: Efficient and Stable Reinforcement Learning via Concurrency-Controlled Partial Rollout with Importance Sampling

Zekai Qu, Yinxu Pan, Ao Sun, Chaojun Xiao, Xu Han

机构 * OpenBMB（开放智能实验室） ； Tsinghua University（清华大学） 纠错

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.AI、cs.LG

Comments 13 pages, 4 figures

EditGRPO: Reinforcement Learning with Post-Rollout Edits for Clinically Accurate Chest X-Ray Report Generation

Kai Zhang, Christopher Malon, Lichao Sun, Martin Renqiang Min

机构 * NEC Laboratories America（NEC美国实验室） ； Lehigh University（莱特大学） 纠错

专题命中 测试时计算 ：reasoning(abstract)；分类 cs.CL

Comments AACL 2025

F2RVLM: Boosting Fine-grained Fragment Retrieval for Multi-Modal Long-form Dialogue with Vision Language Model

Hanbo Bi, Zhiqiang Yuan, Zexi Jia, Jiapei Zhang, Chongyang Li, Peixiang Luo, Ying Deng, Xiaoyue Duan, Jinchao Zhang

专题命中 测试时计算 ：reasoning(abstract)