Senior Staff Engineer – PyTorch Accelerator

Summary:

We are seeking an experienced PyTorch maintainer to bring new AI accelerators to PyTorch and get that work accepted into the official upstream project.

Responsibilities:

Upstream new hardware support to PyTorch: RFCs, pull requests, code reviews, issue triage, and ongoing maintenance.

Work with PyTorch maintainers and working groups to align integration with PyTorch architecture, testing standards, and release requirements. Design and implement backend integration (PrivateUse1, dispatch keys, operator registration, memory management, streams/events, autograd), partnering with compiler, runtime, kernel, driver, and hardware teams. Improve ATen operators, torch.compile, distributed training, mixed precision, and profiling; build and maintain tests, CI workflows, and benchmarks for correctness and performance. Debug correctness, performance, memory, build, and compatibility issues across the PyTorch stack; write clear documentation and mentor other engineers.

Skills:

Basic Qualifications:

• 5+ years of experience in open-source ecosystem development or related roles.

• Strong hands-on experience with PyTorch internals: ATen, the dispatcher, operator registration, autograd, tensor/device abstractions, and Python/C++ extensions.

• Strong C++ and Python skills, with a track record of contributing clean, well-tested code to large open-source projects (design discussions, code reviews, pull requests, CI debugging).

• Experience integrating ML frameworks with hardware accelerators, runtimes, compilers, kernels, or device drivers, and a good understanding of GPU/AI accelerator architecture (memory hierarchy, asynchronous execution, streams/events, profiling).

• Comfortable with PyTorch build/test workflows (Linux, CMake, GitHub, CI) and debugging across Python, C++, runtime, and hardware layers.

• Bachelor's degree in Computer Science, Computer Engineering, Electrical Engineering, or a related field, or equivalent experience.

Preferred Qualifications:

• Recognized PyTorch contributor or maintainer, with changes accepted into the main project.

• Experience with PyTorch accelerator integration: PrivateUse1 backend development, OpenReg-style reference implementations, or out-of-tree backend maintenance.

• Experience with torch.compile (TorchDynamo, TorchInductor), MLIR/LLVM, Triton, or XLA, or optimizing large-scale LLM training (e.g. VeRL) and inference (e.g., vLLM, SGLang).

• Experience building benchmarking, profiling, and performance regression systems; familiarity with distributed backends such as NCCL, Client, HCCL, oneCCL, or custom ProcessGroup implementations.

• Experience working with hardware vendors or open-source foundations.