Machine Learning Performance Engineer

TL;DR

Machine Learning Performance Engineer: Optimizing performance of ML models for training and inference with an accent on low-level GPU programming, system-wide optimization, and high-throughput real-time inference. Focus on debugging and enhancing CUDA performance, distributed GPU training, and networking technologies for GPU clusters.

Company

hirify.global is a global trading firm leveraging machine learning and technology to innovate in finance.

What you will do

• Optimize performance of ML models during training and inference across systems.
• Improve CUDA code and GPU-level operations including memory hierarchy and tensor cores.
• Analyze and enhance throughput and latency in real-time and research inference systems.
• Work with distributed GPU training technologies like NCCL and MPI.
• Utilize networking technologies such as Infiniband, RoCE, and NVLink to link GPU clusters.
• Debug and profile performance using CUDA GDB, NSight Systems, and related tools.

Requirements

• Fluency in English
• Experience with modern ML techniques and toolsets.
• Strong low-level GPU knowledge including PTX, SASS, warps, and cooperative groups.
• Proficiency with CUDA debugging and optimization tools.
• Understanding of distributed GPU training and networking technologies.
• Inventive mindset with willingness to question approaches and tools.