The role, as published
About the company
Our client is a fast-growing technology company.
The role
Raydar is recruiting for this role on behalf of our client. Own speed and efficiency for machine learning training and serving systems, working from low-level GPU code up to multi-machine deployments. You will profile, tune and harden these systems alongside a small, senior technical group.
What you'll do
- Improve GPU and system performance for model training and serving.
- Find and resolve slowdowns across the hardware, memory, software and cluster layers.
- Write low-level GPU optimizations using CUDA and Triton.
- Design and improve distributed training and serving engines.
- Own communication performance between GPUs and machines, including high-speed networking and disaggregated serving.
- Build benchmarking and regression tooling so performance improvements hold up in production.
- Co-design kernels, runtimes and models with the underlying hardware in mind.
What we're looking for
- Hands-on experience writing custom GPU kernels for NVIDIA hardware.
- Has built core capabilities inside a widely used model training or serving framework, rather than only deploying or integrating it.
- Degree in computer science or a related quantitative field; recent graduates should come from a top program.
- A year or more of current, professional experience in machine learning infrastructure or GPU tuning.
- Track record of speeding up model execution at the hardware, runtime and multi-machine levels.
- Strong grounding in machine learning and computer science fundamentals.
Bonus points
- Background tuning generative models that work with visual or other non-text data.
Location and work model
- California, United States
- On-site, 5 days a week in the office
- Full-time
Compensation, as the posting describes it
- Equity