SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Luma AI is seeking a Research Scientist / Engineer to optimize the company's multimodal models for performance at scale. You will profile and optimize GPU, CPU, and accelerator code to ensure efficient training and deployment without sacrificing model quality. This is deep performance engineering work focused on fused kernels, tensor cores, Triton and CUDA, and distributed multi-node deployment.
Key Responsibilities:
- Profile and optimize GPU/CPU/accelerator code for maximum utilization and minimal latency
- Write high-performance PyTorch, Triton, and CUDA code, developing custom operations as needed
- Develop fused kernels and leverage tensor cores and modern hardware features across 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 models
First 90 Days:
Days 1–30: Profile current training and inference paths to identify the biggest performance wins. Days 30–60: Ship and validate a kernel or architecture optimization that measurably improves utilization or latency. Days 60–90: Build monitoring and automation to prevent performance regression.
About Luma: Luma's mission is to build unified general intelligence that can generate, understand, and operate in the physical world. The company believes multimodality is critical for intelligence, with vision as the next step beyond language models.
Requirements:
- Expert-level Triton/CUDA programming and GPU optimization
- Strong PyTorch skills, including kernel development and custom operations
- Proficiency with profiling tools (NVIDIA Nsight, torch profiler, custom tooling)
- Deep understanding of transformer architectures and attention mechanisms
Nice to Have:
- Experience with compilers and exporters (torch.compile, TensorRT, ONNX, XLA)
- Experience optimizing inference workloads for latency and throughput
- Triton compiler and kernel fusion techniques
- Knowledge of warp-level intrinsics and advanced CUDA optimization
