| about

w/ Jolly at the Alan Turing Memorial, Manchester, UK

I am a researcher at IQuest Research, building LLMs & Agents. I received my PhD from the lab supervised by Dr. Chenghua Lin, Professor of Natural Language Processing at The University of Manchester. I initiated the Multimodal Art Projection (M-A-P) research community with Ge Zhang, Ruibin Yuan and Yinghao Ma – which aims to drive open-source research to cutting-edge level as the industry. During my journey, I’ve also collaborated with Dr. Jie Fu and had a lot of fun.

Research

My current research focus involves:

Recursive self-improvement (RSI) – letting systems rewrite the harness they run on (ModularRSI, self-evolving terminal agents).
Generalizable coding & terminal agents, from data to model (TerminalTraj, ICML’26 Outstanding Paper; IQuest-Coder-V1, InCoder-32B).
Scalable Reinforcement Learning for LLMs (TreePO, First Return, Entropy-Eliciting Explore).

Before that, some of my research interests involve:

Music modelling – building the universal understanding (MERT) and generative (YuE, YuE-2) models for music.
LM evaluation at scale (SuperGPQA, OmniBench, Encyclo-K).
Synthetic data for vision-language models (MAmmoTH-VL).

More recent and detailed topics can be referred to my publication page or Google Scholar.

Selected Open-Source Releases

Most of my work ships as open weights, data and code – usually under IQuestLab or M-A-P on Hugging Face.

IQuest-Coder-V1 – 7B/14B/40B(-Loop) code LLMs, white-box checkpoints from pre-training to the final thinking/instruct models (weights).
TerminalTraj – 32K Docker environments and 50K verified terminal-agent trajectories.
MERT – self-supervised acoustic music understanding models, widely used as music encoders (MERT-v1-330M, ~300K monthly downloads across the family).
YuE – open foundation models for long-form music generation (YuE2-3B).
TreePO – tree-based segment-wise rollouts for scalable RL.
MAP-NEO – a fully transparent bilingual LLM series, data pipeline included (neo_7b).
OmniBench and SuperGPQA – evaluation suites for omni-language models and graduate-level knowledge across 285 disciplines.

Past Experience

Interned at J.P. Morgan Artificial Intelligence Research in 2024 summer.
I previously worked as a research assistant at Tsinghua NLP Lab with Prof. Zhiyuan Liu in 2021-2022.

Academic Service: workshop organizer at Open Science for Foundation Models, ICLR’25 and LLM4MA, ISMIR’25; shared-task organizer at NLPCC’25 (Gender Bias Mitigation); reviewer at ACL, EMNLP, CVPR, ICLR, NeurIPS and more.

news

Sep 15, 2026	Released ModularRSI, a benchmark-disjoint and modular framework for generalizable recursive harness self-improvement.
Jul 13, 2026	Four papers are accepted by the ICML’26, Seoul, and TerminalTraj received an Outstanding Paper Award.
May 20, 2026	Two papers are accepted by the ACL’26 Findings.
Mar 20, 2026	Released InCoder-32B, a code foundation model for industrial scenarios, together with the IQuest-Coder-V1 technical report.
Jan 22, 2026	Four papers are accepted by the ICLR’26, including YuE.

selected publications

arXiv

ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement

Siwei Wu, Jincheng Ren, Yizhi Li, and 11 more authors

arXiv preprint arXiv:2609.14857, 2026

arXiv Bib

@article{wu2026modularrsi,
  title = {ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement},
  author = {Wu, Siwei and Ren, Jincheng and Li, Yizhi and Li, Haau-Sing and Yang, Chengran and Zhang, Yuxuan and Gu, Weicheng and Yang, Jian and Batista-Navarro, Riza and Zhang, Chuanyi and Liu, Xianglong and Zhou, Ming and Dai, Bryan and Lin, Chenghua},
  journal = {arXiv preprint arXiv:2609.14857},
  year = {2026},
  date = {2026},
  url = {https://arxiv.org/abs/2609.14857},
}

ICML

Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments

Siwei Wu, Yizhi Li, Yuyang Song, and 8 more authors

In International Conference on Machine Learning (ICML), 2026

ICML Outstanding Paper arXiv Bib Code

Outstanding Paper Award at ICML 2026.

@inproceedings{wu2026terminaltraj,
  title = {Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments},
  author = {Wu, Siwei and Li, Yizhi and Song, Yuyang and Zhang, Wei and Wang, Yang and Batista-Navarro, Riza and Yang, Xian and Tang, Mingjie and Dai, Bryan and Yang, Jian and Lin, Chenghua},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2026},
  date = {2026},
  url = {https://arxiv.org/abs/2602.01244},
}

arXiv

IQuest-Coder-V1 Technical Report

Jian Yang, Wei Zhang, Shawn Guo, and 9 more authors

arXiv preprint arXiv:2603.16733, 2026

arXiv Bib Code

@article{yang2026iquestcoder,
  title = {IQuest-Coder-V1 Technical Report},
  author = {Yang, Jian and Zhang, Wei and Guo, Shawn and Ye, Zhengmao and Jing, Lin and Liu, Shark and Li, Yizhi and Wu, Jiajun and Liu, Cening and Song, Yuyang and Wu, Siwei and others},
  journal = {arXiv preprint arXiv:2603.16733},
  year = {2026},
  date = {2026},
  url = {https://arxiv.org/abs/2603.16733},
}

NeurIPS

OmniBench: Towards the Future of Universal Omni-Language Models

Yizhi Li, Ge Zhang, Yinghao Ma, and 8 more authors

In Advances in Neural Information Processing Systems (NeurIPS), 2025

arXiv Bib Code

@inproceedings{li2025omnibench,
  title = {OmniBench: Towards the Future of Universal Omni-Language Models},
  author = {Li, Yizhi and Zhang, Ge and Ma, Yinghao and Yuan, Ruibin and Zhu, Kang and Guo, Hangyu and Liang, Yiming and Liu, Jiaheng and Yang, Jian and Wu, Siwei and others},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2025},
  date = {2025},
  url = {https://arxiv.org/abs/2409.15272},
}

NeurIPS

SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines

M-A-P Team, Xinrun Du, Yifan Yao, and 19 more authors

In Advances in Neural Information Processing Systems (NeurIPS), 2025

arXiv Bib

@inproceedings{mapteam2025supergpqa,
  title = {SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines},
  author = {Team, M-A-P and Du, Xinrun and Yao, Yifan and Ma, Kaijing and Wang, Bingli and Zheng, Tianyu and Zhu, King and Liu, Minghao and Liang, Yiming and Jin, Xiaolong and Wei, Zhenlin and Zheng, Chujie and Deng, Kaixin and Gavin, Shawn and Jia, Shian and Jiang, Sichao and Liao, Yiyan and Li, Rui and Li, Qinrui and Li, Sirun and Li, Yizhi and others},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2025},
  date = {2025},
  url = {https://arxiv.org/abs/2502.14739},
}

arXiv

TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modeling

Yizhi Li, Qingshui Gu, Zhoufutu Wen, and 14 more authors

arXiv preprint arXiv:2508.17445, 2025

arXiv Bib Website

@article{li2025treepo,
  title = {TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modeling},
  author = {Li, Yizhi and Gu, Qingshui and Wen, Zhoufutu and Li, Ziniu and Xing, Tianshun and Guo, Shuyue and Zheng, Tianyu and Zhou, Xin and Qu, Xingwei and Zhou, Wangchunshu and Zhang, Zheng and Shen, Wei and Liu, Qian and Lin, Chenghua and Yang, Jian and Zhang, Ge and Huang, Wenhao},
  journal = {arXiv preprint arXiv:2508.17445},
  year = {2025},
  date = {2025},
  url = {https://arxiv.org/abs/2508.17445},
}

ICLR

MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training

Yizhi Li, Ruibin Yuan, Ge Zhang, and 15 more authors

In International Conference on Learning Representations (ICLR), 2024

arXiv Bib HTML Code

@inproceedings{li2024mert,
  title = {MERT: Acoustic Music Understanding Model with Large-Scale Self-supervised Training},
  author = {Li, Yizhi and Yuan, Ruibin and Zhang, Ge and Ma, Yinghao and Chen, Xingran and Yin, Hanzhi and Lin, Chenghua and Ragni, Anton and Benetos, Emmanouil and Gyenge, Norbert and Dannenberg, Roger and Liu, Ruibo and Chen, Wenhu and Xia, Gus and Shi, Yemin and Huang, Wenhao and Guo, Yike and Fu, Jie},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2024},
  date = {2024},
  url = {https://openreview.net/forum?id=w3YZ9MSlBu},
}