GPU Performance Engineer for Multimodal Transformers

AItoolnavio; Greater London; Hybrid; GBP 90,000 - 130,000; Full time

Job summary

Luma is seeking an expert GPU optimization engineer to accelerate multimodal transformer workloads. You will profile, tune, and deploy kernel- and operator-level improvements across GPUs, CPUs, and accelerators to maximize training throughput and minimize latency.

You will implement fused kernels, leverage tensor cores, and push Triton/CUDA capabilities, while collaborating on distributed multi-node deployment and performance instrumentation.

Qualifications

• 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.

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 analysis tools and automation.
• Research and implement cutting-edge optimization techniques for transformer models.
• Scale and systemize performance improvements for long-term reliability.

Skills

Triton/CUDA programming

PyTorch kernel development

GPU optimization

Profiling tools

Tools

NVIDIA Nsight

Torch profiler

Custom tooling

Job description

Luma is seeking an expert GPU optimization engineer to accelerate multimodal transformer workloads. You will profile, tune, and deploy kernel- and operator-level improvements across GPUs, CPUs, and accelerators to maximize training throughput and minimize latency.

You will implement fused kernels, leverage tensor cores, and push Triton/CUDA capabilities, while collaborating on distributed multi-node deployment and performance instrumentation.