Software Engineer - SI Inference Codesign

Job Description

Our team productionizes and experiments with novel ML models for our custom HW- we train and deploy large neural networks for efficient inference on compute-constrained edge devices (CPU / GPU / SI ASIC). The nature of this role is multi-disciplinary - you will work at the intersection of machine learning and systems by building the co-designing ML models and their developing the associated deployment infrastructure that enables the seamless training, deployment, and inference of all neural networks that run at Tesla SI. 

Job Responsibilities

• Build robust SI frameworks to lower neural networks to edge devices
• Build robust SI infrastructure to train and fine-tune networks for Autopilot and Optimus on large GPU clusters
• Deploy state-of-the-art neural networks on heterogenous compute, including Tesla’s in-house SI ASIC, with an aim to maximize network performance while minimizing latency
• Collaborate with SI scientists and compiler engineers to effectively compress large models to run in low precision
• Design and implement custom GPU kernels (CUDA / OpenCL) for efficient training and post-processing of network outputs 

Job Requirements

• Proficiency with Python and C++, including modern C++ (14/17/20)
• Proficiency with PyTorch or another machine learning framework
• Proficiency with training and deploying neural networks for real-world SI
• Proficiency with computer systems and computer architecture
• Experience with CUDA