Jose Javier Gonzalez Ortiz

  • Senior Research Scientist, Databricks AI

I am a Senior Research Scientist on the Databricks AI LLM Scaling team. I was part of the core team behind DBRX, a frontier LLM model developed by Databricks. I am currently working on improving Large Language Model (LLM) training and inference efficiency, with an emphasis on low-precision and quantization techniques. Our recent paper FlashOptim reduces model training memory by more than 50% without sacrificing quality.
I received my PhD in Computer Science from MIT CSAIL, where I worked with John Guttag and Adrian Dalca. During my PhD, I worked on efficient deep learning methods, and for my thesis I focused on learning reconfigurable vision models, where multiple tasks are learned jointly in an amortized way. We achieved this using techniques such as hypernetworks and in-context learning. During my PhD, I also worked on asynchronous large-scale distributed training and neural network pruning methods.
At MIT, I co-organized the Missing Semester course, a lecture series on programmer tools often left untaught in modern CS curricula. During my PhD, I interned at Facebook AI Research and Microsoft Research. Previously, I received a BSc in Telematics Engineering from ICAI in Spain, and interned at CERN Openlab.

Deep Learning Libraries

Publications

Projects

Awards & Honors