Machine Learning Performance Engineer

Location: Hong Kong

Company

hirify.global is a financial services firm developing technology and quantitative systems for global trading and research.

What you will do

• Optimise machine learning model training and inference for large-scale, real-time, and research workloads.
• Improve CUDA-based systems through a whole-systems approach spanning storage, networking, hosts, and GPUs.
• Debug training performance end to end and distinguish useful throughput from measured throughput.
• Analyse low-level behaviour including cache access, tensor core arithmetic, warp synchronisation, and asynchronous memory loads.
• Connect GPU clusters and support distributed training using high-performance networking and collective algorithms.

Requirements

• Experience with low-level systems programming, optimisation, and modern machine learning techniques and toolsets.
• Low-level GPU knowledge covering PTX, SASS, warps, cooperative groups, Tensor Cores, and memory hierarchy.
• Experience debugging and optimising with CUDA GDB, Nsight Systems, and Nsight Compute.
• Knowledge of Triton, CUTLASS, CUB, Thrust, cuDNN, and cuBLAS.
• Background with InfiniBand, RoCE, GPUDirect, PXN, rail optimisation, NVLink, NCCL, or MPI.
• Fluent English and an inventive approach to solving complex systems problems.

Culture & Benefits

• Work in a machine learning team supporting a global trading business.
• Use a rapid-feedback trading environment for machine learning experimentation.
• Collaborate on technically challenging performance and systems problems.