About the job
Join us at the forefront of AI compiler technology and help shape the future of accelerated computing. NVIDIA is seeking passionate engineers to build the next generation of tools used by AI developers and researchers worldwide. Our team is developing Thunder, an ambitious, source-to-source compiler built to unlock outstanding performance for PyTorch models on NVIDIA GPUs. This is a unique opportunity to contribute to a project that enhances the PyTorch ecosystem, working with modern compiler stacks like PyTorch 2.0's TorchDynamo and TorchInductor to create powerful, open-source solutions that benefit the entire community. If you are driven to solve complex problems and want to make a foundational impact on the AI ecosystem, apply to join our collaborative and innovative team.
Responsibilities
lead the design, implementation, optimization, and maintenance of the core compiler technologies that accelerate massive deep learning workloads; dive deep into performance analysis, scrutinizing workloads running on thousands of GPUs to find optimization opportunities that will shape the future design of Thunder; work closely with leading compiler, library, and systems teams—including experts behind nvFuser, TVM, XLA, and CUDA—to translate the latest research into practical, high-impact solutions for the open-source community
Qualifications
Minimum
A Bachelor's, Master's, or Ph.D. in Computer Science or a related technical field (or equivalent experience); 8+ years of relevant work experience; a strong command of Python and experience building complex, well-tested software systems; hands-on experience with deep learning frameworks like PyTorch or JAX; a solid foundation in compiler concepts such as abstract syntax trees (ASTs), intermediate representations (e.g., SSA form), program analysis, and code generation; excellent communication and collaboration skills, essential for working effectively in a distributed, open-source environment
Preferred
Previous contributions to deep learning compiler projects (e.g., TVM, MLIR, IREE) or deep learning frameworks themselves; deep expertise in the internals of PyTorch, particularly its compiler stack (TorchDynamo, TorchInductor); experience with JAX-like functional transformations and their application in a compiler context; familiarity with parallel programming, distributed systems, and writing high-performance CUDA code; a track record of impactful participation in open-source communities, such as through code contributions, design discussions, or mentorship