Job Details

Senior AI/ML Tooling Engineer

Role: We are looking for an ML tooling engineer to build tools to analyze and optimize distillation, training, and inference of ML models. You will develop and enhance GM's internal ML tooling for high performance software by leveraging state of the art tools like Nsight Systems, PyTorch, etc. The Autonomous Vehicle (AV) software stack heavily relies on machine learning models to perform critical driving tasks. These cutting-edge custom ML models require an ecosystem of in-house tooling to analyze and improve them. In this role, you will collaborate closely with engineers and researchers from different AV Engineering teams (e.g., Computer Vision, Perception, Behavioral Prediction) to scope out system requirements, while engaging with AV hardware teams to understand the target hardware platform and its constraints.

What You'll Do

• Identify new opportunities to improve both training and inference efficiency
• Build workflows for correctness and performance analysis on physical in-car compute and sensors
• Build tooling to predict model performance based on architecture and data shape
• Build tooling to trace actual performance on large distributed training and distillation jobs, running on the world's most powerful GPUs, and analyze the results
• Continually evolve the toolchain and stack, to leverage the latest advancements in AI
• Influence model architecture decisions and strategy within GM

Your Skills & Abilities

• 5+ years of experience in the field of AI/ML
• Experience with ML frameworks (e.g., PyTorch, TensorFlow) and NVIDIA developer ecosystem (TensorRT, Nsight-systems, Nsight-compute)
• Expertise in writing production quality Python/C++ code
• Expertise in the software development life-cycle - coding, debugging, optimization, testing, integration
• BS, or higher degree, in CS/CE/EE, or equivalent

What will give you a competitive edge

• Experience developing and deploying machine learning models
• GPU programming (CUDA) and familiarity with ML SW stack (e.g., cuDNN, cuBLAS)
• Experience with ML accelerators and hardware architecture