University of Washington
academicThe University of Washington's Paul G. Allen School of Computer Science & Engineering is one of the field's most productive open-model and open-data groups, working closely with the co-located Allen Institute for AI (AI2) — a relationship so tight that much flagship output (OLMo, Tulu, Molmo, OpenScholar) ships under Ai2 branding and is tracked there, not here. UW's own defining contribution is open data at scale: Ludwig Schmidt's mlfoundations group produced DataComp-LM, whose from-scratch DCLM-7B and 240T-token CommonCrawl testbed set the open-data standard, alongside MINT-1T and Magpie.
Systems and tokenization run deep: Zihao Ye's FlashInfer (MLSys 2025 Best Paper) now powers vLLM, SGLang, TensorRT-LLM and TGI; SuperBPE and infini-gram reshape tokenization and n-gram retrieval; and Aditya Kusupati's Matryoshka Representation Learning became the industry standard for variable-dimension embeddings. Ranjay Krishna's RAIVN lab drives the vision line (TrajViT, Generate Any Scene, Superposed Decoding).
Industry and AI2 affiliations are pervasive: Hajishirzi, Smith and Krishna hold senior AI2 roles, Zettlemoyer splits with Meta, Farhadi is AI2's CEO, and Schmidt left for Stanford/Anthropic in November 2024 — so 2024 mlfoundations datasets are UW-era while later output (e.g. Gelato) attributes to Stanford. The entry keeps only UW-branded artifacts.
People
- Hannaneh Hajishirzi Google Scholar — Professor, Allen School; Senior Director of NLP, AI2
- Luke Zettlemoyer Google Scholar — Professor, Allen School; Research Scientist, Meta AI
- Noah A. Smith Google Scholar — Professor, Allen School; Senior Director of NLP Research, AI2
- Ranjay Krishna Google Scholar — Assistant Professor, Allen School (RAIVN Lab); Research Lead, AI2 (PRIOR)
- Ali Farhadi Google Scholar — Professor, Allen School; CEO, AI2