Home Publications Talks Teaching



Xiang Yue

(Pronounced as "Shiang Yoo-eh", 岳翔)

Postdoctoral Researcher
Language Technologies Institute
School of Computer Science
Carnegie Mellon University

Email: xyue2@andrew.cmu.edu

Google Scholar Github Linkedin Twitter Wechat

Bio

I am a postdoctoral researcher at CMU working with Prof. Graham Neubig on natural language processing (NLP) and large language models (LLMs). I obtained my Ph.D. from The Ohio State University (OSU-NLP-Group), where I was trained by Prof. Huan Sun and Prof. Yu Su. I also collaborate closely with Prof. Wenhu Chen at University of Waterloo. I received my B.S. in Computer Science from Wuhan University in 2018.


My research aims to understand and enhance the reasoning capabilities of LLMs across different modalities and contexts while improving their responsibility and reliability.
  • Understanding LLMs' reasoning through rigorous evaluation and benchmarking.
  • Improving LLMs' reasoning capabilities with synthetic data generation techniques. I love simple, scalable and cost-effective data solutions and enjoy pushing the boundary of state-of-the-arts while open-sourcing everything I build.
    • Representative Work: MAmmoTH and MAmmoTH2: Strong reasoning models achieving SoTA in 2023 and 2024 by scaling up reasoning rationales from different sources. Notable, MAmmoTH2's 10 million synthetic instructions achieve performance comparable to Meta Llama 3 Instruct, which relies on 10 million human-annotated examples.
    • Other Major Contributed Projects: Pangea (SoTA open-source multilingual multimodal model on both English and Multilingual evaluations), OpenCodeInterpreter (ACL 2024 Findings; 90+ accuracy on HumanEval with 7B size), MAmmoTH-VL (SoTA open-source multimodal model), MultiUI (7M multimodal GUI understanding and agent data samples)
  • Designing responsible and inclusive AI models cover different aspects (e.g., Privacy, Attribution, Multilinguality) for real-world applications (e.g., Healthcare).

What's New

Last Updated: 12/2024