Edge AI Software Engineer: Optimise Models & Runtimes

Arm Limited; Lunds kommun; Hybrid; SEK 700,000 - 900,000; Full time

Benefits offered by this job

Hybrid working

Open-source collaboration

Job summary

Arm Limited in Lund is seeking experienced software engineers to help develop ExecuTorch, PyTorch’s framework for efficient AI on edge devices. You’ll contribute across AI model optimisation, compilers, runtimes and hardware acceleration, spanning Python tooling to portable C++ runtimes and hardware backends.

You’ll move between high-level model transformations and low-level runtime and hardware behaviour to tackle performance challenges, improving latency, memory use and power efficiency.

Qualifications

• Experience with C++ and Python in production-quality software.
• Knowledge of compilers, runtimes or performance optimisation.
• Interest in how software executes on hardware and improving efficiency.
• Good communication and ability to work across teams.

Responsibilities

• Develop and optimise ExecuTorch compiler, runtime and tooling, from Python APIs to C/C++ components.
• Analyse and profile AI workloads to reduce latency, memory use, cycle count and power usage.
• Develop optimised backends for Mali GPUs, Cortex-A/M CPUs and Ethos-U NPUs and other edge platforms.
• Collaborate with Arm and PyTorch community to deliver solutions across the stack.
• Share knowledge and mentor other engineers as appropriate.

Skills

C++

Python

Performance optimisation

Compiler knowledge

Tools

PyTorch

Ahead-of-time compilers

Job description

Arm Limited in Lund is seeking experienced software engineers to help develop ExecuTorch, PyTorch’s framework for efficient AI on edge devices. You’ll contribute across AI model optimisation, compilers, runtimes and hardware acceleration, spanning Python tooling to portable C++ runtimes and hardware backends.

You’ll move between high-level model transformations and low-level runtime and hardware behaviour to tackle performance challenges, improving latency, memory use and power efficiency.