GPU Performance Architect for Multimodal Transformers

Luma; United Kingdom; On-site; GBP 110,000 - 150,000; Full time

Job summary

Luma is seeking a seasoned GPU optimization expert to accelerate multimodal transformer training and inference. You’ll write kernels in CUDA/Triton, optimize PyTorch ops, and push hardware to the limit across distributed multi-node deployments.

You'll own profiling, kernel fusion, and performance monitoring, delivering tangible latency and throughput gains while advancing cutting-edge transformer research.

Qualifications

• Expert-level Triton/CUDA programming required.
• Strong PyTorch kernel development experience.
• Proficiency with profiling tools (Nsight, torch profiler).
• Deep understanding of transformer architectures and attention mechanisms.

Responsibilities

• Profile and optimize GPU/CPU/accelerator code for maximum utilization and minimal latency.
• Write high-performance PyTorch, Triton, and CUDA, dropping to custom operations when 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 automation.
• Research and implement cutting-edge optimization techniques for transformer models.

Skills

Expertise in Triton/CUDA

PyTorch kernel development

Profiling tools

Transformer architectures knowledge

Job description

Luma is seeking a seasoned GPU optimization expert to accelerate multimodal transformer training and inference. You’ll write kernels in CUDA/Triton, optimize PyTorch ops, and push hardware to the limit across distributed multi-node deployments.

You'll own profiling, kernel fusion, and performance monitoring, delivering tangible latency and throughput gains while advancing cutting-edge transformer research.