Jiawei Zhao

I am a Research Scientist at Meta FAIR, where I work on topics related to optimization, reasoning and efficiency. I received my Ph.D. at Caltech. My research seeks to uncover the statistical principles related to Large Language Models (LLMs), with the aim of developing algorithms that are theoretically grounded, scalable to large-scale models, and efficient in practice. This includes

• modern optimization algorithms for model training (GaLore, GaLore 2, signSGD-MV)
• efficient LLM reasoning and large-scale reinforcement learning (DeepConf, GRESO, M2PO)
• foundations of deep learning, quantization, and efficient inference (check out my research for details)

Reach me at jwzhao at meta dot com (work) or jwzzhao at gmail dot com (personal).