Staff Software Engineer, Machine Learning, TPU Workload Optimization

About the job

Google's software engineers develop the next-generation technologies that change how billions of users connect, explore, and interact with information and one another. Our products need to handle information at massive scale, and extend well beyond web search. We're looking for engineers who bring fresh ideas from all areas, including information retrieval, distributed computing, large-scale system design, networking and data storage, security, artificial intelligence, natural language processing, UI design and mobile; the list goes on and is growing every day. As a software engineer, you will work on a specific project critical to Google’s needs with opportunities to switch teams and projects as you and our fast-paced business grow and evolve. We need our engineers to be versatile, display leadership qualities and be enthusiastic to take on new problems across the full-stack as we continue to push technology forward.

Our mission is to provide the best possible cloud-based ML pre-training, post-training, and inference solutions to our customers. The team is part of the AI Engine and focuses on the training/inference workloads and the infrastructure. Our team’s mission is to provide the infrastructure and the framework support to enable serving of ML models on Cloud GPUs and TPUs. In particular, we support external customers for their ML training/inference onboarding and optimizations. You will join a team working on an emergent product with potential to change how customers use Google infrastructure for machine learning training/inference. It is a dynamic and exciting environment that is open to change and is full of growth opportunities.

The AI and Infrastructure team is redefining what’s possible. We empower Google customers with breakthrough capabilities and insights by delivering AI and Infrastructure at unparalleled scale, efficiency, reliability and velocity. Our customers include Googlers, Google Cloud customers, and billions of Google users worldwide.

We're the driving force behind Google's groundbreaking innovations, empowering the development of our cutting-edge AI models, delivering unparalleled computing power to global services, and providing the essential platforms that enable developers to build the future. From software to hardware our teams are shaping the future of world-leading hyperscale computing, with key teams working on the development of our TPUs, Vertex AI for Google Cloud, Google Global Networking, Data Center operations, systems research, and much more.

Responsibilities

• Act as ML software engineer leading the onboarding for stable stack, and optimization for training and serving solutions on the latest TPU NPIs.
• Set the technical direction and architect the training/inference onboarding and optimization frameworks for our customer needs. Work closely with customers to onboard their workloads to production and optimization.
• Collaborate with the ML research, ML performance, model optimization tooling, and other optimization teams.
• Ship stable stack to our customers for training/inference optimization solutions on TPU.
• Prepare and optimize very large reference models, demonstrating state of the art single-host and multi-host inference solutions, especially at large scale.

Qualifications

Minimum qualifications:

• Bachelor’s degree or equivalent practical experience.
• 8 years of experience in software development.
• 4 years of experience leading ML design and optimizing ML infrastructure (e.g., model deployment, model evaluation, data processing, debugging, fine tuning).
• 2 years of experience with state of the art training (e.g. Megatron-LM, DeepSpeed) and inference techniques (e.g. TensorRT-LLM, vLLM, SGLang).

Preferred qualifications:

• Master’s degree or PhD in Engineering, Computer Science, or a related technical field.
• Experience in optimizing machine learning models for large scale training/inference workloads.
• Experience in different large scale ML optimizations techniques for improving latency and throughput.
• Experience with accelerators (TPUs or GPUs), or HPC.