Research Engineer, Large Scale Pre-Training Performance

We are seeking a research engineer to define, drive, and critically contribute to the next generation of the state-of-the-art ML models on TPU. As part of the Pre-Training team you will co-design the model, and implement critical components across Model architecture, ML frameworks, custom kernels and platform, to deliver frontier models with maximum efficiency.

Artificial Intelligence could be one of humanity’s most useful inventions. At Google DeepMind, we’re a team of scientists, engineers, machine learning experts and more, working together to advance the state of the art in artificial intelligence. We use our technologies for widespread public benefit and scientific discovery, and collaborate with others on critical challenges, ensuring safety and ethics are the highest priority.

We’re looking for a Research Engineer to re-define efficient training of frontier LLMs at massive scale. This role offers an opportunity to influence the design of frontier LLM models, and drive an effort to ensure efficient training and inference.

• Being responsible for Pre-Training efficiency and optimising the performance of the latest models on Google’s fleet of hardware accelerators - throughout the entire LLM research, training and deployment lifecycle.
• Being responsible for guiding model design to ensure inference-efficiency.
• Greatly improving the performance of LLM models on hardware accelerators by optimizing at all levels, including developing custom kernels when necessary.
• Collaborating with the compiler, framework, and platform teams. And ensure efficient training at industry-largest scale.
• Profile models to identify performance bottlenecks and opportunities for optimization.
• Develop low-level custom kernels for maximum performance of the most critical operators.
• Collaborating with research teams by enabling new critical operators in advance of their availability in frameworks and compilers.

In order to set you up for success as a Research Engineer at Google DeepMind,  we look for the following skills and experience:

• A proven track record of critical contributions to the distributed training of LLMs at 1e25 FLOPs scale on modern GPU/TPU clusters  
• Experience in programming hardware accelerators GPU/TPUs via ML frameworks (e.g. JAX, PyTorch) and low-level programming models (e.g. CUDA, OpenCL)
• Experience in leveraging custom kernels and compiler infrastructure to improve performance on hardware
• Experience with Python and neural network training (publications, open-source projects, relevant work experience, etc.)

At Google DeepMind, we value diversity of experience, knowledge, backgrounds and perspectives and harness these qualities to create extraordinary impact. We are committed to equal employment opportunity regardless of sex, race, religion or belief, ethnic or national origin, disability, age, citizenship, marital, domestic or civil partnership status, sexual orientation, gender identity, pregnancy, or related condition (including breastfeeding) or any other basis as protected by applicable law. If you have a disability or additional need that requires accommodation, please do not hesitate to let us know.