Research Scientist Intern, PyTorch Core (PhD)
Meta

Job Info

Our team makes PyTorch run faster and more resource-efficient without sacrificing the flexibility and ease of use of PyTorch.We are looking for a talented intern to join our team on an ambitious open-source, public initiative. The project is aimed at designing an LLM-based Triton and CUDA kernel generator through modern AI techniquesTeam scope:-Advance PyTorch 2.0 technologies that bring torch.compile() to the heart of PyTorch -Advance PyTorch Distributed through torch.compile()-Improve PyTorch out-of-the-box performance on GPU, CPU, accelerators-Vertical performance optimization for models for training and inference-Contribute to the research direction of AI for GPU programming, https://gpu-mode.github.io/popcorn/Our team at Meta AI offers twelve (12) to sixteen (16) weeks long internships and we have various start dates throughout the year. To learn more about our research, visit https://ai.facebook.com.

Research Scientist Intern, PyTorch Core (PhD) Responsibilities:

• Contribute to the development of the LLM for generating GPU kernels using the PyTorch ecosystem stack
• Research and explore techniques to improve the performance of the LLM, such as multiagent systems, planning (Monte Carlo Tree search and similar), reasoning, Reinforcement Learning, and AI Agents
• Collaborate with the open-source community to gather feedback and iterate on the project
• Contribute to PyTorch (Dynamo, Inductor, Core, Distributed) within the scope of the project
• Improve PyTorch performance in general
• Explore the intersection of PyTorch compiler and PyTorch distributed
• Optimize Generative AI models across the stack (pre-training, fine-tuning, and inference)
• Conduct cutting-edge research on ML compiler and ML distributed technologies

Minimum Qualifications:

• Currently has, or is in the process of obtaining, a PhD degree in the field of Computer Science or a related STEM field
• Experience in ML compiler, Distributed Training, ML systems, or similar
• Proficient in Python
• Must obtain work authorization in the country of employment at the time of hire, and maintain ongoing work authorization during employment

Preferred Qualifications:

• Experience working on other ML compiler stack, especially on PT2 stack or Triton
• Cuda programming, C++
• Experience with planning algorithms, MCTS, AI Agents, LLM performance, Reinforcement learning
• Research or software engineer experience demonstrated via grants, fellowships, patents, internship, work experience, and/or coding competitions
• Experience doing performance optimization on machine learning models
• First-authored publications at peer-reviewed conferences (e.g. NeurIPS, MLSys, ASPLOS, PLDI, CGO, PACT, ICML, or similar)
• Intent to return to degree-program after the completion of the internship/co-op

About Meta: