Staff ML Performance Engineer, Performance Architecture

Job summary

Responsibilities

You'll identify, quantify and deliver optimisations across training and cloud inference workloads. You'll profile workloads to find bottlenecks, build optimisations that work across targets rather than one-off fixes, and track the gains with clear benchmarks.

• Profiling ML workloads across training and cloud inference to identify bottlenecks, using system and kernel level profilers
• Designing and implementing efficiency improvements to maximise MFU, throughput and utilisation, e.g. parallelism, compilation, mixed precision, caching
• Building reusable, cross-target optimisations (kernels, data loaders, frameworks such as Triton) rather than one-off, per-workflow fixes
• Designing and implementing benchmarking tools to track efficiency gains and catch regressions on priority training and cloud inference workloads
• Informing cloud GPU hardware strategy and readiness in partnership with platform teams

Tech stack

• Triton
• Python
• CUDA