Job Description

About Luma AI

Luma's mission is to build multimodal AI to expand human imagination and capabilities. 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.

About the Role

The Performance Optimization team at Luma is dedicated to maximizing the efficiency and performance of our AI models. Working closely with both research and engineering teams, this group ensures that our cutting-edge multimodal models can be trained efficiently and deployed at scale while maintaining the highest quality standards.

Responsibilities

• Profile and optimize GPU/CPU/Accelerator code for maximum utilization and minimal latency
• Write high-performance PyTorch, Triton, CUDA, deferring to custom PyTorch operations if necessary
• Develop fused kernels and leverage tensor cores and modern hardware features for optimal hardware utilization on different hardware platforms
• Optimize model architectures and implementations for distributed multi-node production deployment
• Build performance monitoring and analysis tools and automation
• Research and implement cutting-edge optimization techniques for transformer model

Experience

• Expert-level proficiency in Triton/CUDA programming and GPU optimization
• Strong PyTorch skills
• Experience with PyTorch kernel development and custom operations
• Proficiency with profiling tools (NVIDIA Nsight, torch profiler, custom tooling)
• Deep understanding of transformer architectures and attention mechanisms
• (Preferred) Experience with compilers/exporters such as torch.compile, TensorRT, ONNX, XLA
• (Preferred) Experience optimizing inference workloads for latency and throughput
• (Preferred) Experience with Triton compiler and kernel fusion techniques
• (Preferred) Knowledge of warp-level intrinsics and advanced CUDA optimization

Your applications are reviewed by real people.