Cover Image

SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Wang, Yizhou, Tang, Chen, Deng, Han, Xiao, Jiabei, Liu, Jiaqi, Wu, Jianyu, Yao, Jun, Li, Pengze, Su, Encheng, Wang, Lintao, Zhuang, Guohang, Ren, Yuchen, Fei, Ben, Hu, Ming, Chen, Xin, Zhou, Dongzhan, He, Junjun, Yue, Xiangyu, Yin, Zhenfei, Wu, Jiamin, Zheng, Qihao, Zhou, Yuhao, Xu, Huihui, Ma, Chenglong, Lu, Yan, Zhang, Wenlong, Song, Chunfeng, Torr, Philip, Tang, Shixiang, Ma, Xinzhu, Ouyang, Wanli, Bai, Lei
Format:	Preprint
Published:	2025
Subjects:	Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866909959053115392
author	Wang, Yizhou Tang, Chen Deng, Han Xiao, Jiabei Liu, Jiaqi Wu, Jianyu Yao, Jun Li, Pengze Su, Encheng Wang, Lintao Zhuang, Guohang Ren, Yuchen Fei, Ben Hu, Ming Chen, Xin Zhou, Dongzhan He, Junjun Yue, Xiangyu Yin, Zhenfei Wu, Jiamin Zheng, Qihao Zhou, Yuhao Xu, Huihui Ma, Chenglong Lu, Yan Zhang, Wenlong Song, Chunfeng Torr, Philip Tang, Shixiang Ma, Xinzhu Ouyang, Wanli Bai, Lei
author_facet	Wang, Yizhou Tang, Chen Deng, Han Xiao, Jiabei Liu, Jiaqi Wu, Jianyu Yao, Jun Li, Pengze Su, Encheng Wang, Lintao Zhuang, Guohang Ren, Yuchen Fei, Ben Hu, Ming Chen, Xin Zhou, Dongzhan He, Junjun Yue, Xiangyu Yin, Zhenfei Wu, Jiamin Zheng, Qihao Zhou, Yuhao Xu, Huihui Ma, Chenglong Lu, Yan Zhang, Wenlong Song, Chunfeng Torr, Philip Tang, Shixiang Ma, Xinzhu Ouyang, Wanli Bai, Lei
contents	We present a scientific reasoning foundation model that aligns natural language with heterogeneous scientific representations. The model is pretrained on a 206B-token corpus spanning scientific text, pure sequences, and sequence-text pairs, then aligned via SFT on 40M instructions, annealed cold-start bootstrapping to elicit long-form chain-of-thought, and reinforcement learning with task-specific reward shaping, which instills deliberate scientific reasoning. It supports four capability families, covering up to 103 tasks across workflows: (i) faithful translation between text and scientific formats, (ii) text/knowledge extraction, (iii) property prediction, (iv) property classification, (v) unconditional and conditional sequence generation and design. Compared with specialist systems, our approach broadens instruction coverage, improves cross-domain generalization, and enhances fidelity. We detail data curation and training and show that cross-discipline learning strengthens transfer and downstream reliability. The model, instruct tuning datasets and the evaluation code are open-sourced at https://huggingface.co/SciReason and https://github.com/open-sciencelab/SciReason.
format	Preprint
id	arxiv_https___arxiv_org_abs_2509_21320
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines Wang, Yizhou Tang, Chen Deng, Han Xiao, Jiabei Liu, Jiaqi Wu, Jianyu Yao, Jun Li, Pengze Su, Encheng Wang, Lintao Zhuang, Guohang Ren, Yuchen Fei, Ben Hu, Ming Chen, Xin Zhou, Dongzhan He, Junjun Yue, Xiangyu Yin, Zhenfei Wu, Jiamin Zheng, Qihao Zhou, Yuhao Xu, Huihui Ma, Chenglong Lu, Yan Zhang, Wenlong Song, Chunfeng Torr, Philip Tang, Shixiang Ma, Xinzhu Ouyang, Wanli Bai, Lei Computation and Language We present a scientific reasoning foundation model that aligns natural language with heterogeneous scientific representations. The model is pretrained on a 206B-token corpus spanning scientific text, pure sequences, and sequence-text pairs, then aligned via SFT on 40M instructions, annealed cold-start bootstrapping to elicit long-form chain-of-thought, and reinforcement learning with task-specific reward shaping, which instills deliberate scientific reasoning. It supports four capability families, covering up to 103 tasks across workflows: (i) faithful translation between text and scientific formats, (ii) text/knowledge extraction, (iii) property prediction, (iv) property classification, (v) unconditional and conditional sequence generation and design. Compared with specialist systems, our approach broadens instruction coverage, improves cross-domain generalization, and enhances fidelity. We detail data curation and training and show that cross-discipline learning strengthens transfer and downstream reliability. The model, instruct tuning datasets and the evaluation code are open-sourced at https://huggingface.co/SciReason and https://github.com/open-sciencelab/SciReason.
title	SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines
topic	Computation and Language
url	https://arxiv.org/abs/2509.21320