Susan Zhang
Co-first author of Meta AI's OPT; built LLM training infrastructure at FAIR and RL systems for OpenAI Five
Susan Zhang's homepage describes work building large-scale systems for data and models, including a reinforcement learning system at OpenAI for OpenAI Five and the LLM training infrastructure at Meta AI / FAIR that created OPT-175B. Zhang is one of three equal-contribution first authors of OPT: Open Pre-trained Transformer Language Models (2022), a suite of decoder-only models from 125M to 175B parameters; the release included a logbook of the training run and the full training codebase, metaseq, which is published under the MIT License. Zhang is also a co-author of the OpenAI paper Dota 2 with Large Scale Deep Reinforcement Learning (2019).1234
Work
OPT and metaseq (external site: github.com)Open-source code
One of three equal-contribution first authors of OPT (2022), with Stephen Roller and Naman Goyal. Zhang's homepage says the OPT-175B release included the 175B research model, a 114-page training logbook, the full training codebase, and smaller models from 125M to 66B parameters; the metaseq repository is published under the MIT License.213
OpenAI Five (external site: arxiv.org)Published research
Co-author of Dota 2 with Large Scale Deep Reinforcement Learning (2019), which describes the distributed training system used to train OpenAI Five.41
Other documented roles
Roles the sources document but that are not confirmed as current.
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.