Senior Research Engineer - Performance Optimization

We are looking for engineers with significant problem solving experience in PyTorch, CUDA and distributed systems. You will work with Research Scientists to build & train cutting edge foundation models on thousands of GPUs. 

Responsibilities

• Ensure efficient implementation of models & systems for data processing, training, inference and deployment

• Identify and implement optimization techniques for massively parallel and distributed systems

• Identify and remedy efficiency bottlenecks (memory, speed, utilization) by profiling and implementing high-performance CUDA, Triton, C++ and PyTorch code

• Work closely together with the research team to ensure systems are planned to be as efficient as possible from start to finish

• Build tools to visualize, evaluate and filter datasets

• Implement cutting-edge product prototypes based on multimodal generative AI

Experience

• Experience training large models using Python & Pytorch, including practical experience working with the entire development pipeline from data processing, preparation & data loading to training and inference.

• Experience optimizing and deploying inference workloads for throughput and latency across the stack (inputs, model inference, outputs, parallel processing etc.)

• Experience with profiling CPU & GPU code in PyTorch, including Nvidia Nsight or similar.

• Experience writing & improving highly parallel & distributed PyTorch code, with familiarity in DDP, FSDP, Tensor Parallel, etc.

• Experience writing high-performance parallel C++. Bonus if done within an ML context with PyTorch, like for data loading, data processing, inference code.

• Experience with high-performance Triton / CUDA and writing custom PyTorch kernels. Top candidates will be able to utilize tensor cores; optimize performance with CUDA memory and other similar skills.

• Good to have experience working with Deep learning concepts such as Transformers & Multimodal Generative models such as Diffusion Models and GANs.

• Good to have experience building inference / demo prototype code (incl. Gradio, Docker etc.)

• 
  

Compensation

• The pay range for this position in California is $180,000 - $250,000yr; however, base pay offered may vary depending on job-related knowledge, skills, candidate location, and experience. We also offer competitive equity packages in the form of stock options and a comprehensive benefits plan. 

Your applications are reviewed by real people.

At Luma AI, we believe that multimodality is critical for intelligence. To go beyond language models and build more aware, capable and useful systems, the next step function change will come from vision. So, we are working on training and scaling up multimodal foundation models for systems that can see and understand, show and explain, and eventually interact with our world to effect change.

We will deploy these systems to make a new kind of intelligent creative partner that can imagine with us. Free and away from the pressure of being creative. It's for all of us whose imaginations have been constrained, who've had to channel vivid dreams through broken words, hoping others will see what we see in our mind's eye. A partner that can help us show — not just tell.

Dream Machine is an early step to building that. Try it here

Why you should join us:

• Luma is bringing together the best team in the world to achieve our goal, from researchers to engineers and designers to growth operators

• Luma is not just a lab - we are deeply product focused and our vision merging AI models and delightful products is unique in the industry

• We build. We ship. Our early products have been wildly successful