Senior/Staff ML Performance Engineer, Nuro

• 3+ years of professional or research experience in ML infrastructure, distributed training, or ML systems engineering
• Hands-on experience with quantization methods, including Activation-Aware Weight Quantization (AWQ), Accurate Quantized Training (AQT), FP-8 training, or related methods
• Experience building or maintaining quantization libraries (e.g., AQT, bitsandbytes, NVIDIA Transformer Engine, DeepSpeed Compression)
• Understanding of calibration and scaling strategies for quantized models to minimize accuracy loss

Desirable

• Nuro is seeking an experienced ML Performance Engineer with deep expertise in quantized training to join our ML Infrastructure team
• In this role, you will drive the adoption of state-of-the-art quantization techniques, enabling training and deployment of highly-efficient models that power the Nuro Driver™
• You will help to shape the technical strategy and partner closely with research and product groups to ensure our ML infrastructure is optimized for both cutting-edge research and real-time deployment on autonomous vehicles
• As an ML Performance Engineer for Nuro's ML Training Infrastructure you will improve model training efficiency and drive the adoption of state-of-the-art quantization and low-precision training techniques. This will include:
• Staying ahead of emerging research and evaluating new methods
• Implementing quantized training methods (e.g., AWQ, AQT, GPTQ) for new and existing self-driving models
• Leading the design and implementation of efficiency initiatives for model training, including low-bit quantization, and pruning for both research and production workloads
• Championing and implementing tools and approaches to pinpoint root-causes for possible model quality and accuracy regressions when training at lower precisions
• Collaborating cross-functionally with research, infrastructure, and product teams balancing accuracy, latency, and resource constraints