Distributed Training Engineer, Sora

About the Team The Sora team is working on making video a key capability of OpenAI’s foundation models. We are a hybrid research and product team that seeks to understand and expand the capabilities of our video models, while ensuring their reliability and safety. We accomplish this both through directly studying and experimenting with the models, as well as deploying them into the real-world to distribute their benefits widely.   About the Role As a Distributed Systems/ML engineer, you will work on improving the training throughput for our internal training framework and enable researchers to experiment with new ideas. This requires good engineering (for example designing, implementing, and optimizing state-of-the-art AI models), writing bug-free machine learning code (surprisingly difficult!), and acquiring deep knowledge of the performance of supercomputers. We’re looking for people who love optimizing performance, understanding distributed systems, and who cannot stand having bugs in their code.  This role is based in San Francisco, CA. We use a hybrid work model of 3 days in the office per week and offer relocation assistance to new employees. In this role, you will: - Collaborate with researchers to enable them to develop systems-efficient video models and architectures - Apply the latest techniques to our internal training framework to achieve impressive hardware efficiency for our training runs - Profile and optimize our training framework You might thrive in this role if you: - Have experience working with multi-modal ML pipelines - Love diving deep into systems implementations and understanding their fundamentals in order to improve their performance and maintainability - Have strong software engineering skills and are proficient in Python. - Have experience understanding and optimizing training kernels - Are passionate about understanding stable training dynamics About OpenAI OpenAI is an AI research and deployment company dedicated to en