About Me
⚠️ Note: This homepage was last updated in June 2024 and will remain on pause until I produce work of significant quality and impact. As a result, regular maintenance of the site is currently unnecessary.
I am a senior undergraduate student at Tianjin University.
My research interests lie in Large Language Model and Multimodal Learning.
Selected Papers
(EMNLP 2024) Automated Evaluation Using Self-Adaptive Rubrics
Zhiyuan Fan*, Weinong Wang, Xing Wu, Debing Zhang
Under Review(NeurIPS 2024) Calibrated Self-Rewarding Vision Language Models
Yiyang Zhou, Zhiyuan Fan*, … , Huaxiu Yao
These authors contributed equally.
Under Review(ACL 2024) Exploring the Potential of Dense Information in Multimodal Alignment
Zhiyuan Fan*, Zhihong Chen, Benyou Wang
Under Review(EMNLP 2023) Efficient Data Learning for Open Information Extraction with Pre-trained Language Models
Zhiyuan Fan*, Shizhu He
[Arxiv](Electronics) Cuckoo Matrix: A High Efficient and Accurate Graph Stream Summarization on Limited Memory
Zhuo Li, Zhuoran Li, Zhiyuan Fan*, Jianli Zhao, Siming Zeng, Peng Luo and Kaihua Liu
[MDPI]
Selected Engineering Projects
- Reimplemented the tensor parallel strategy from the ground up using PyTorch and NCCL.
- Successfully trained models exceeding 100 billion parameters from scratch.
Selected Experiences
University of North Carolina at Chapel Hill
Intern research (advisor: Asst. Prof. Huaxiu Yao)
February 2024 - PresentInstitute of Neuroscience, Chinese Academy of Sciences
Intern research (advisor: Prof. Xue Li)
April 2023 - June 2023Baidu & Jina AI
Intern research (advisor: CTO, Nan Wang)
July 2022 - Jan 2023Institute of Automation, Chinese Academy of Sciences
Intern research (advisor: Assoc. Researcher Shizhu He)
April 2022 - January 2024
Selected Open Source Projects
- PaddleNLP (GitHub, 11.7k Stars)
Contributions: Submitted two PRs (7k+ lines):- Implemented DPR training on a single-GPU with a batch size of 512 using gradient cache.
- Developed a from-scratch DeBERTa pre-trained language model implementation.
- LLM Cookbook (GitHub, 10.2k Stars)
Contributions: Authored tutorials on Retrieval-Augmented Generation (RAG), agent-based systems, and techniques for processing unstructured data.
Selected Book
- Statistical Learning Method Solutions Manual (GitHub, 1.4k Stars, Online)
- First Author of Chapter 27: Pre-trained Language Model
- Reviewer for Chapters 16, 19, 20, and 21.
Selected Awards
Champion
Baidu PaddlePaddle AI Hackathon: Pretrained Language Model.
2023Gold Award
China International College Students ‘Internet+’ Innovation and Entrepreneurship Competition.
2023Silver Award
Huawei Ascend Operator Development Competition, solo.
2021