2 months ago
Base Salary
$251k - $310k/yr
Responsibilities
- Analyze model architectures to identify training and inference bottlenecks involving memory bandwidth, compute, and communication.
- Develop and apply quantization, pruning, knowledge distillation, and efficient attention techniques.
- Optimize model code for TPU and GPU accelerators using compiler features and low-level libraries.
- Evaluate data, tensor, pipeline, and expert parallelism strategies to improve model scalability and efficiency.
- Design and implement low-latency, high-throughput serving solutions for generative models and optimize training pipelines.
- Build and maintain performance analysis, profiling, and debugging tools for machine learning models.
Requirements
- MS or PhD in Computer Science, Machine Learning, Robotics, or a related field.
- At least five years of experience with deep learning architectures, particularly Transformers, Diffusion Models, and MoEs, along with related algorithms and optimization techniques.
- Proficiency in JAX and Flax, with potential experience in TensorFlow or PyTorch.
- Expertise using XProf, Perfetto, or NVIDIA Nsight for diagnosing machine learning workload performance issues.
- Hands-on experience with quantization, pruning, distillation, and other model compression methods.
- Strong Python programming skills and potentially C++ experience, along with software development best practices.
- Preferred knowledge of TPU and GPU architectures and how to optimize code for them.
- Preferred familiarity with ML compilers such as XLA and how they translate high-level code into efficient hardware instructions.
- Preferred understanding of training and serving models across multiple devices and machines.
- Preferred experience contributing to frameworks and libraries that improve training speed and scalability, such as JAX, Gemax, or XManager.
Benefits
- Full-time position with eligibility for a discretionary annual bonus program, equity incentive plan, and company benefits program.
- Salary range varies by U.S. location and may vary for remote work based on location.
Categories
About Waymo
On the journey to be the world's most trusted driver. With the Waymo Driver, we can improve mobility while saving thousands of lives. Download the Waymo One app and ride today. Waymo reaches out to candidates from official channels only (e.g. directly from @waymo.com email addresses, or through our recruiters or sourcers who are noted as such on LinkedIn). We do not contact candidates about career opportunities through instant messaging apps like Telegram, email addresses from domains other than waymo.com (such as Gmail addresses), direct messages on Twitter, Facebook, and Instagram, or text messages. Visit waymo.com to check out our official job listings. Need support? Send us a message – we're here to help.