Buchumschlag

Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Wang, Weixun, Xu, XiaoXiao, An, Wanhe, Dai, Fangwen, Gao, Wei, He, Yancheng, Huang, Ju, Ji, Qiang, Jin, Hanqi, Li, Xiaoyang, Li, Yang, Li, Zhongwen, Lin, Shirong, Liu, Jiashun, Liu, Zenan, Luo, Tao, Muhtar, Dilxat, Qu, Yuanbin, Shi, Jiaqiang, Sun, Qinghui, Tan, Yingshui, Tang, Hao, Wang, Runze, Wang, Yi, Wang, Zhaoguo, Wu, Yanan, Xiong, Shaopan, Xu, Binchen, Xu, Xander, Xu, Yuchi, Zhang, Qipeng, Zhang, Xixia, Zhao, Haizhou, Zhao, Jie, Zhao, Shuaibing, Zheng, Baihui, Zheng, Jianhui, Zheng, Suhang, Zhu, Yanni, Cai, Mengze, Cao, Kerui, Chen, Xitong, Dai, Yue, Du, Lifan, Feng, Tao, He, Tao, Hu, Jin, Hu, Yijie, Jiang, Ziyu, Li, Cheng, Li, Xiang, Liang, Jing, Lin, Xin, Liu, Chonghuan, Liu, ZhenDong, Lv, Zhiqiang, Mi, Haodong, Mo, Yanhu, Ni, Junjia, Pei, Shixin, Shen, Jingyu, Song, XiaoShuai, Wang, Cecilia, Wang, Chaofan, Wang, Kangyu, Wang, Pei, Wang, Tao, Wang, Wei, Xiao, Ke, Xu, Mingyu, Xu, Tiange, Ya, Nan, Yang, Siran, Ye, Jianan, Zang, Yaxing, Zhang, Duo, Zhang, Junbo, Zheng, Boren, Deng, Wanxi, Pan, Ling, Qu, Lin, Su, Wenbo, Wang, Jiamang, Wei, Hu, Wu, Minggang, Yu, Cheng, Zhao, Bing, Zheng, Zhicheng, Zheng, Bo
Format:	Preprint
Veröffentlicht:	2025
Schlagworte:	Artificial Intelligence Computation and Language
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

_version_	1866911508106051584
author	Wang, Weixun Xu, XiaoXiao An, Wanhe Dai, Fangwen Gao, Wei He, Yancheng Huang, Ju Ji, Qiang Jin, Hanqi Li, Xiaoyang Li, Yang Li, Zhongwen Lin, Shirong Liu, Jiashun Liu, Zenan Luo, Tao Muhtar, Dilxat Qu, Yuanbin Shi, Jiaqiang Sun, Qinghui Tan, Yingshui Tang, Hao Wang, Runze Wang, Yi Wang, Zhaoguo Wu, Yanan Xiong, Shaopan Xu, Binchen Xu, Xander Xu, Yuchi Zhang, Qipeng Zhang, Xixia Zhao, Haizhou Zhao, Jie Zhao, Shuaibing Zheng, Baihui Zheng, Jianhui Zheng, Suhang Zhu, Yanni Cai, Mengze Cao, Kerui Chen, Xitong Dai, Yue Du, Lifan Feng, Tao He, Tao Hu, Jin Hu, Yijie Jiang, Ziyu Li, Cheng Li, Xiang Liang, Jing Lin, Xin Liu, Chonghuan Liu, ZhenDong Lv, Zhiqiang Mi, Haodong Mo, Yanhu Ni, Junjia Pei, Shixin Shen, Jingyu Song, XiaoShuai Wang, Cecilia Wang, Chaofan Wang, Kangyu Wang, Pei Wang, Tao Wang, Wei Xiao, Ke Xu, Mingyu Xu, Tiange Ya, Nan Yang, Siran Ye, Jianan Zang, Yaxing Zhang, Duo Zhang, Junbo Zheng, Boren Deng, Wanxi Pan, Ling Qu, Lin Su, Wenbo Wang, Jiamang Wang, Wei Wei, Hu Wu, Minggang Yu, Cheng Zhao, Bing Zheng, Zhicheng Zheng, Bo
author_facet	Wang, Weixun Xu, XiaoXiao An, Wanhe Dai, Fangwen Gao, Wei He, Yancheng Huang, Ju Ji, Qiang Jin, Hanqi Li, Xiaoyang Li, Yang Li, Zhongwen Lin, Shirong Liu, Jiashun Liu, Zenan Luo, Tao Muhtar, Dilxat Qu, Yuanbin Shi, Jiaqiang Sun, Qinghui Tan, Yingshui Tang, Hao Wang, Runze Wang, Yi Wang, Zhaoguo Wu, Yanan Xiong, Shaopan Xu, Binchen Xu, Xander Xu, Yuchi Zhang, Qipeng Zhang, Xixia Zhao, Haizhou Zhao, Jie Zhao, Shuaibing Zheng, Baihui Zheng, Jianhui Zheng, Suhang Zhu, Yanni Cai, Mengze Cao, Kerui Chen, Xitong Dai, Yue Du, Lifan Feng, Tao He, Tao Hu, Jin Hu, Yijie Jiang, Ziyu Li, Cheng Li, Xiang Liang, Jing Lin, Xin Liu, Chonghuan Liu, ZhenDong Lv, Zhiqiang Mi, Haodong Mo, Yanhu Ni, Junjia Pei, Shixin Shen, Jingyu Song, XiaoShuai Wang, Cecilia Wang, Chaofan Wang, Kangyu Wang, Pei Wang, Tao Wang, Wei Xiao, Ke Xu, Mingyu Xu, Tiange Ya, Nan Yang, Siran Ye, Jianan Zang, Yaxing Zhang, Duo Zhang, Junbo Zheng, Boren Deng, Wanxi Pan, Ling Qu, Lin Su, Wenbo Wang, Jiamang Wang, Wei Wei, Hu Wu, Minggang Yu, Cheng Zhao, Bing Zheng, Zhicheng Zheng, Bo
contents	Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.
format	Preprint
id	arxiv_https___arxiv_org_abs_2512_24873
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem Wang, Weixun Xu, XiaoXiao An, Wanhe Dai, Fangwen Gao, Wei He, Yancheng Huang, Ju Ji, Qiang Jin, Hanqi Li, Xiaoyang Li, Yang Li, Zhongwen Lin, Shirong Liu, Jiashun Liu, Zenan Luo, Tao Muhtar, Dilxat Qu, Yuanbin Shi, Jiaqiang Sun, Qinghui Tan, Yingshui Tang, Hao Wang, Runze Wang, Yi Wang, Zhaoguo Wu, Yanan Xiong, Shaopan Xu, Binchen Xu, Xander Xu, Yuchi Zhang, Qipeng Zhang, Xixia Zhao, Haizhou Zhao, Jie Zhao, Shuaibing Zheng, Baihui Zheng, Jianhui Zheng, Suhang Zhu, Yanni Cai, Mengze Cao, Kerui Chen, Xitong Dai, Yue Du, Lifan Feng, Tao He, Tao Hu, Jin Hu, Yijie Jiang, Ziyu Li, Cheng Li, Xiang Liang, Jing Lin, Xin Liu, Chonghuan Liu, ZhenDong Lv, Zhiqiang Mi, Haodong Mo, Yanhu Ni, Junjia Pei, Shixin Shen, Jingyu Song, XiaoShuai Wang, Cecilia Wang, Chaofan Wang, Kangyu Wang, Pei Wang, Tao Wang, Wei Xiao, Ke Xu, Mingyu Xu, Tiange Ya, Nan Yang, Siran Ye, Jianan Zang, Yaxing Zhang, Duo Zhang, Junbo Zheng, Boren Deng, Wanxi Pan, Ling Qu, Lin Su, Wenbo Wang, Jiamang Wang, Wei Wei, Hu Wu, Minggang Yu, Cheng Zhao, Bing Zheng, Zhicheng Zheng, Bo Artificial Intelligence Computation and Language Agentic crafting requires LLMs to operate in real-world environments over multiple turns by taking actions, observing outcomes, and iteratively refining artifacts. Despite its importance, the open-source community lacks a principled, end-to-end ecosystem to streamline agent development. We introduce the Agentic Learning Ecosystem (ALE), a foundational infrastructure that optimizes the production pipeline for agentic model. ALE consists of three components: ROLL, a post-training framework for weight optimization; ROCK, a sandbox environment manager for trajectory generation; and iFlow CLI, an agent framework for efficient context engineering. We release ROME, an open-source agent grounded by ALE and trained on over one million trajectories. Our approach includes data composition protocols for synthesizing complex behaviors and a novel policy optimization algorithm, Interaction-Perceptive Agentic Policy Optimization (IPA), which assigns credit over semantic interaction chunks rather than individual tokens to improve long-horizon training stability. Empirically, we evaluate ROME within a structured setting and introduce Terminal Bench Pro, a benchmark with improved scale and contamination control. ROME demonstrates strong performance across benchmarks like SWE-bench Verified and Terminal Bench, proving the effectiveness of ALE.
title	Let It Flow: Agentic Crafting on Rock and Roll, Building the ROME Model within an Open Agentic Learning Ecosystem
topic	Artificial Intelligence Computation and Language
url	https://arxiv.org/abs/2512.24873