Sunny Guha

I'm a machine learning researcher working on pretraining and architecture research for large language models — optimizers, parametrizations, normalization, and the empirics of what actually makes scale work. Most recently I've contributed to training frontier-scale foundation models.

Before ML, I was a theoretical physicist. I did my PhD at Texas A&M on string theory and conformal field theory — superspace formulations of 11-dimensional supergravity and M-theory, and analytic structure of CFT correlators. A lot of the instincts carry over: dimensional analysis, scaling arguments, and a stubborn preference for understanding why a thing works before trusting that it does.

I write about the things I'm reading and building on the blog. A list of things I've made is on projects, and academic work — both ML and physics — is on publications.

Reach me at sunyguha91@hotmail.com. Find me on GitHub and LinkedIn.