Stanford Center for Research on Foundation Models
The Center for Research on Foundation Models (CRFM) is an interdisciplinary initiative at Stanford HAI that makes fundamental advances in the study, development, and deployment of foundation models. It combines technical research, beneficial applications, societal-impact work, and AI policy, with an emphasis on open scientific study and accessible tools.
At a Glance
- Academic researchers and students
- AI and foundation-model developers
- Model and application developers
- Healthcare, biomedical, legal, education, finance, and enterprise practitioners
- +2 more
AI Tools by Stanford Center for Research on Foundation Models
(1)
HELM
Open Source LLM Evaluation Framework
Discussions
No discussions yet
Be the first to start a discussion about Stanford Center for Research on Foundation Models
Latest News
CRFM launched HELM Arabic Enterprise, a transparent and reproducible Arabic-language enterprise evaluation leaderboard developed with Arabic.AI.
CRFM launched HELM Arabic, evaluating LLMs on seven Arabic-language benchmarks in collaboration with Arabic.AI.
CRFM introduced the HELM Long Context leaderboard for transparent, comparable, and reproducible evaluation of long-context models, in collaboration with LVMH.
CRFM published Reliable and Efficient Amortized Model-Based Evaluation, presenting an approach to reduce the cost of repeated model evaluation.
Products & Services
Open-source Python framework and web leaderboards for holistic, reproducible, and transparent evaluation of language and multimodal models. It standardizes datasets and benchmarks, provides a unified model interface, tracks metrics such as accuracy, efficiency, bias, and toxicity, and exposes prompt-level results.
Biomedical language model, previously called PubMedGPT, with public model access and code for pre-training and fine-tuning.
A research instruction-following model fine-tuned from LLaMA 7B using 52K instruction-following demonstrations.
JAX-based framework and codebase for transparent, accessible large-scale foundation-model training, including distributed training and evaluation workflows.
Market Position
CRFM is positioned as an academic, interdisciplinary, and open-science alternative to proprietary model-evaluation and transparency work. Rather than selling a hosted model API, it publishes research, open model/training artifacts, benchmarks, reproducible leaderboards, and policy analysis; HELM differentiates through holistic multi-metric evaluation, prompt-level transparency, and reproducibility across many model providers and modalities.
Leadership
Founders
Percy Liang
Stanford associate professor of computer science, faculty member of HAI, and CRFM director. He leads research on machine learning and foundation models.
Rishi Bommasani
Stanford computer-science researcher and lead author of CRFM's inaugural 'On the Opportunities and Risks of Foundation Models' report; his work focuses on foundation models, evaluation, and societal impact.
Executive Team
Percy Liang
Director
Stanford associate professor of computer science and HAI faculty member; leads CRFM's interdisciplinary research agenda.
Founding Story
CRFM grew from an informal Stanford community created in March 2021 in response to the rapid emergence of large, broadly trained models such as GPT-3, BERT, and CLIP and the need to understand their capabilities, risks, and societal consequences. Stanford established the center to convene researchers across disciplines, conduct open scientific study, and make the engineering expertise, compute, data, code, and evaluation resources needed for foundation-model research more accessible.
Business Model
Revenue Model
CRFM is a Stanford research center rather than a commercial product company. Its support model includes corporate support through the Stanford HAI Membership program, research supporters, sponsored collaborations, and in-kind API credits or compute.
Target Markets
- Academic researchers and students
- AI and foundation-model developers
- Model and application developers
- Healthcare, biomedical, legal, education, finance, and enterprise practitioners
- Policymakers and regulators
- The broader public and AI community
- Evaluating and comparing foundation models and LLMs
- Reproducing model-benchmark results and inspecting individual prompts and responses
- Training and fine-tuning open foundation models
- Biomedical and healthcare language-model research
- Long-context, safety, instruction-following, multilingual, legal, financial, and enterprise evaluation
- Research and policy analysis of open-model benefits, risks, and transparency