Graph Compiler Engineer

About the Team

The Platform ML team builds the ML side of our state-of-the-art internal training framework used to train our cutting-edge models.  We work on distributed model execution as well as the interfaces and implementation for model code, training, and inference.  Our priorities are to maximize training throughput (how quickly we can train a new model) and researcher throughput (how quickly we can develop new models) with the goal of accelerating progress towards AGI.  We frequently collaborate with other teams to speed up the development of new capabilities.

About the Role

As a Graph Compiler Engineer you will work on the capture and optimization of model graphs for our training runs.  Seamless graph capture is important for keeping researcher throughput high, while graph-level optimizations will allow us to get state-of-the-art training throughput on current and future accelerators.

We’re looking for engineers who are excited to work on our training framework and with researchers at the cutting edge of ML research.  Given the scale of our training, small improvements to hardware utilization for our largest runs can produce large cost savings.  Enabling researchers to iterate quickly on their ideas will produce large improvements in model capabilities and compute efficiency.

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:

• Work on our internal training framework to enable easy graph capture of our research experiments
• Apply advanced optimizations to the graph to maximally utilize hardware resources

You might thrive in this role if you:

• Have strong software engineering skills and are proficient in Python
• Have experience running machine learning experiments
• Have experience working on a machine learning compiler
• Have deep knowledge of GPU performance and systems-level optimization