H2Lab @ UW

We explore fundamental research in generative AI and NLP to create solutions with real-world impact, advancing a future of AI that is open, transparent, and scientifically grounded. Our team's research spans:

• Large language models — developing pioneering methods, data technologies, and architectures to make these systems more powerful, efficient, and effective.
• Reasoning and Agents — developing Reinforcement Learning methods, reasoning algorithms, tool use, and agents.
• Open AI initiatives — building some of the first truly open models, such as OLMo and Tulu. We release state-of-the-art systems, datasets, training resources, and tools to the public to ensure traceability, reproducibility, and accessibility.
• Science of AI and evaluation — systematically examining the capabilities, limitations, and behaviors of AI systems to better understand how large models learn, reason, and make decisions.
• AI for science — creating systems that interpret scientific text and data, with a focus on integrating tools and intelligent agents to accelerate discovery.

Our work is regularly published at leading AI, ML, and NLP conferences (NeurIPS, ICLR, ICML, ACL, EMNLP, NAACL, COLM). Our open-source data, code, and models are hosted on Hugging Face and GitHub. As of 2025, our models have been downloaded over 10 million times. We are affiliated with UW-NLP and UW-AI.

• July 14, 2025:

  Congratulations to Prof. Hanna Hajishirzi for being promoted as full professor!

• June 13, 2025:

  Molmo received best paper honorable mention at CVPR 2025.

• June 4, 2025:

  Sewon Min received ACM dissertation award honorable mention, and ACL dissertation award.

• June 3, 2025:

  Congratulations to Dr. Akari Asai for successfully defending her Ph.D.

• View all news