Research Scientist • Intern

Who you are

You are pursuing a PhD in a relevant field and have a strong foundation in machine learning frameworks, particularly PyTorch. Your academic background has equipped you with the skills to tackle complex problems in model performance and optimization. You have experience with GPU programming and understand the intricacies of distributed systems, which will be essential in your role. You are passionate about research and eager to contribute to open-source projects, demonstrating your commitment to the community. Your analytical skills allow you to explore novel approaches to improve model efficiency and performance.

What you'll do

As a Research Scientist Intern at Meta, you will focus on enhancing the performance of PyTorch models by implementing cutting-edge techniques in mixture-of-experts systems. You will work on optimizing end-to-end training and inference throughput on modern accelerators, such as NVIDIA Hopper. Your responsibilities will include exploring communication-aware distributed training methods and kernel optimizations to unlock new performance regimes for large-scale sparse models. You will collaborate with a team of experts to improve the stability and extensibility of the PyTorch framework, ensuring it remains user-friendly while achieving high performance. This internship will provide you with the opportunity to contribute to impactful research and potentially publish your findings.

What we offer

Meta offers a dynamic internship experience where you can work alongside leading researchers in the field. You will have access to state-of-the-art hardware and software tools to facilitate your research. The internship duration ranges from twelve to twenty-four weeks, with various start dates throughout the year. You will gain valuable insights into the industry and have the chance to make significant contributions to the PyTorch community. We encourage you to apply even if your experience doesn't match every requirement, as we value diverse perspectives and backgrounds.