LLM Inference Performance Engineer

G-Research; Greater London; On-site; GBP 90,000 - 150,000; Full time

Benefits offered by this job

Discretionary bonus

35 days leave

Pension contributions

Healthcare and life assurance

Cycle-to-work

Job summary

G-Research in London is seeking an exceptional NLP Performance Engineer to own large-scale LLM inference performance within the NLP Engineering team. You’ll optimise the inference stack for cost-efficiency and speed, collaborating with researchers to bring ideas to life.

This hands-on role involves profiling workloads, improving deployment across GPU architectures, and building reliable tooling for ML workloads, while communicating across research, infra and engineering teams.

Qualifications

• Bachelor’s, Master’s or PhD in computer science or equivalent experience.
• Proven experience profiling, benchmarking and optimising large-scale LLM inference workloads.
• Scientific, evidence-led approach to performance optimisation with rigorous benchmarking.

Responsibilities

• Profiling, benchmarking and optimising large-scale LLM inference workloads across compute infrastructure.
• Ensuring efficient deployment of latest models across GPU architectures; adapt inference stack as hardware evolves.
• Designing and implementing inference optimisations while maintaining output quality.
• Developing reference implementations, libraries and tooling to improve NLP workloads.
• Collaborating with researchers, senior stakeholders and engineers to design optimised solutions.
• Evolve the compute stack with systems, architecture and platform teams for long-term decisions.

Skills

LLM inferences

Performance benchmarking

Python

CUDA

PyTorch

Software engineering

Communication

Transformer inference

Education

Bachelor/Master/PhD in Computer Science

Tools

vLLM

SGLang

TensorRT-LLM

TGI

CUDA toolkit

Job description

G-Research in London is seeking an exceptional NLP Performance Engineer to own large-scale LLM inference performance within the NLP Engineering team. You’ll optimise the inference stack for cost-efficiency and speed, collaborating with researchers to bring ideas to life.

This hands-on role involves profiling workloads, improving deployment across GPU architectures, and building reliable tooling for ML workloads, while communicating across research, infra and engineering teams.