Summary
Elias Frantar is a researcher and Member of Technical Staff at OpenAI with nine years of experience making massive ML models far more efficient. He completed a PhD at IST Austria under Dan Alistarh, where he created breakthrough compression tools—GPTQ and SparseGPT—for low-bit quantization and sparsification of 100B+ models and built Marlin, an INT4xFP16 inference kernel achieving near-ideal speedups. During internships at Google Brain and DeepMind he characterized scaling laws for sparse foundation models and developed QMoE to compress and run trillion-parameter Mixture-of-Experts models in resource-constrained settings. His work spans theory, algorithm design, and highly optimized CUDA kernels, delivering state-of-the-art speed-vs-accuracy trade-offs across vision and language models. Based in San Francisco, he pairs deep research with hands-on engineering and unusual hobbyist chops—designing record-setting, super-fast Lego Rubik’s Cube robots that have attracted millions of views.
9 years of coding experience
2 years of employment as a software developer
Vienna University of Technology
Doctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at Institute of Science and Technology Austria
High School Diploma, Informatics, GPA: 4.00/4.00, High School Diploma, Informatics, GPA: 4.00/4.00 at Technologisches Gewerbemuseum