Yijie Jin

Yijie (Drew) is an incoming Ph.D. student at DENG Lab, Shanghai Jiao Tong University, advised by Prof. Zhijie Deng. He is also a Research Intern at Qwen, Infrastructure Team of Qwen Foundation Model. He conducts research at the intersection of Machine Learning Systems and Efficient Generative AI, with a focus on Algorithm-System Co-Design. Specifically:

Diffusion Language Models: modeling, acceleration, and inference infrastructure ( Diffulex — a flexible and efficient DLM serving engine)
DLM as Speculative Decoding Drafter: leveraging DLMs for efficient autoregressive generation
Efficient Interactive World Models: algo-sys co-optimizations for real-time efficiency

news

Mar 14, 2026	LightningRL accepted at ICML 2026 (Regular).
Feb 28, 2026	Received the Shanghai Outstanding Graduate Award.
Jan 31, 2026	D2F accepted at ICLR 2026 (Poster).
Dec 14, 2025	Invited talk on ASC story at ASC 2026 Campus Tour (Shanghai) @ SJTU.
Aug 31, 2025	Admitted as Ph.D. candidate at Qingyuan Research Institute, SJTU.
Aug 14, 2025	Invited talk on D2fEngine at vLLM Shenzhen Meetup.

selected publications

ArXiv

Multi-Block Diffusion Language Models

Yijie Jin, Jiajun Xu, Yuxuan Liu, and 8 more authors

ArXiv Preprint, 2026

arXiv Bib PDF

@article{jin2026mbd,
  title = {Multi-Block Diffusion Language Models},
  author = {Jin, Yijie and Xu, Jiajun and Liu, Yuxuan and Xu, Chenkai and Tu, Yi and Li, Jiajun and Tu, Dandan and Yan, Xiaohui and Yu, Kai and Liu, Pengfei and Deng, Zhijie},
  journal = {ArXiv Preprint},
  year = {2026},
}

ICML

LightningRL: Breaking the Accuracy–Parallelism Trade-off of Block-wise dLLMs via Reinforcement Learning

Yanzhe Hu, Yijie Jin, Pengfei Liu, and 2 more authors

International Conference on Machine Learning (ICML), 2026

arXiv Bib PDF Code

@article{hu2026lightningrl,
  title = {LightningRL: Breaking the Accuracy–Parallelism Trade-off of Block-wise dLLMs via Reinforcement Learning},
  author = {Hu, Yanzhe and Jin, Yijie and Liu, Pengfei and Yu, Kai and Deng, Zhijie},
  journal = {International Conference on Machine Learning (ICML)},
  year = {2026},
}

ArXiv

LoPA: Scaling dLLM Inference via Lookahead Parallel Decoding

Chenkai Xu^*, Yijie Jin^*, Jiajun Li, and 8 more authors

ArXiv Preprint, 2025

arXiv Bib PDF Code

@article{xu2025lopa,
  title = {LoPA: Scaling dLLM Inference via Lookahead Parallel Decoding},
  author = {Xu, Chenkai and Jin, Yijie and Li, Jiajun and Tu, Yi and Long, Guoping and Tu, Dandan and Song, Mingcong and Si, Hongjie and Hou, Tianqi and Yan, Junchi and Deng, Zhijie},
  journal = {ArXiv Preprint},
  year = {2025},
}

ICLR

Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing

Xu Wang, Chenkai Xu, Yijie Jin, and 3 more authors

International Conference on Learning Representations (ICLR), 2026

arXiv Bib PDF Code

@article{wang2025d2f,
  title = {Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing},
  author = {Wang, Xu and Xu, Chenkai and Jin, Yijie and Jin, Jiachun and Zhang, Hao and Deng, Zhijie},
  journal = {International Conference on Learning Representations (ICLR)},
  year = {2026},
}

ACL

Multimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs

Yijie Jin, Junjie Peng, Xuanchao Lin, and 3 more authors

Annual Meeting of the Association for Computational Linguistics (ACL), 2025

Bib PDF Code

@article{jin2025gsit,
  title = {Multimodal Transformers are Hierarchical Modal-wise Heterogeneous Graphs},
  author = {Jin, Yijie and Peng, Junjie and Lin, Xuanchao and Yuan, Haochen and Wang, Lan and Zheng, Cangzhi},
  journal = {Annual Meeting of the Association for Computational Linguistics (ACL)},
  year = {2025},
}