Research Scientist @ NVIDIA

Yang Chen

Yang Chen

I am an AI Research Scientist @ NVIDIA (ADLR Group) building Large Language Models. I received my Ph.D. from Georgia Tech in August 2024.

My current research interests lie in AI reasoning and scaling reinforcement learning for LLM reasoning.

Connect with me

Announcements

AceReason: Math and Code Reasoning through RL

TLDR: A math and code reasoning model trained entirely through RL. We found math-RL improves code reasoning and use it as a warm-up for code-RL to improve training stability and efficiency.

  • AIME 2024: 78.6% (+8.9%)
  • AIME 2025: 67.4% (+17.2%)
  • LiveCodeBench v5: 61.1% (+8.1%)
  • Codeforce: 2024 ELO (+543)

May 22, 2025

AceReason Overview

Announcements

AceMath-RL: Training Math Reasoning Model with RL

TLDR: A math reasoning model trained entirely through RL, starting from the Deepseek-R1-Distilled-Qwen-7B.

  • AIME 2024: 69.0% (+13.5%)
  • AIME 2025: 53.6% (+14.4%)
  • LiveCodeBench v5: 44.4% (+6.8%)

We share our training recipe, training logs, and data curation details in the blog.

April 22, 2025

AceMath Overview

Announcements

AceMath: Advancing Frontier Math Reasoning with Post-Training and Reward Modeling

Hugging Face Logo Hugginface (Models & Datasets) Technical Report Project Page

Dec 19th, 2024

Past Work

Research ยท 2022 - 2024

Multimodal LLM

Building AI that understand visual world

Research ยท 2019 - 2024

Multilingual LLM

Bridging representation across global languages