Software Engineer - AI/HPC Specialist

Description

We are looking for software engineers to help scale and improve the efficiency of large AI training and inference for HW accelerators. A core part of this is optimising collective operations to enable optimised network utilisation for data sharing. This is an opportunity to work within a highly skilled team, collaborating with a large set of cross-functional partners and help bringing next generation large cluster architectures to life.

Responsibilities

Work on collective communications stacks to optimise networking operations, leading to improved AI inference and training model performance Drive implementation of latency and bandwidth critical networking operations, as well as out-of-band signalling Debug custom and third party multi-host, accelerator enabled AI platforms Software development using C++/C and Python Work closely with other teams to deliver impact develop & improve features and innovations Extend and optimize large scale learning collective operations

Qualifications

3+ years of experience developing in C++/C and Python Experience with High Performance Computing/Networking or AI systems applications frameworks Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience Specialized experience in one or more of the following machine learning/deep learning domains: Hardware accelerators, AI Infrastructure, or high performance networking Solid experience in debugging of distributed systems, revision control systems, testing, and CI pipelines Experience and understanding of AI/HPC systems Deep understanding of the transport stack (e.g. RDMA/RoCE, Infiniband, TCP/IP), its constraints and performance measures and how transport considerations enable the collective communications stack Experience in one or more of the following machine learning/deep learning domains: hardware accelerators, AI Infrastructure, and/or high performance computing (HPC), particularly pertaining to interconnect and collective communications stacks Familiarity with relevant tools, libraries, and frameworks (like PyTorch, NCCL, MPI, CUDA)