Principal Software Engineer – PyTorch Training Frameworks
Core
Principal PyTorch training framework expert driving performance, scalability, and correctness of large-scale AI training on AMD Instinct accelerators.
Role type
Principal Software Engineer (AI Training Frameworks)
Builds
High-performance distributed training workloads (LLMs/foundation models) for AMD Instinct hardware
Domain
AI/ML Infrastructure, High-Performance Computing, GPU Accelerators
Deliverable
production ML models
Required skills
PyTorch internals, distributed training (DDP/FSDP), performance engineering, Python, C/C++, memory optimization, multi-node debugging
Preferred skills
ROCm stack, TorchInductor, Triton, CUDA/HIP programming, Linux systems, upstream open-source contribution
Technologies
PyTorch, ROCm, NCCL, RCCL, TorchInductor, Triton, CUDA, HIP, Linux
Responsibilities
Set technical direction for PyTorch training performance and reliability; Optimize training workloads across single and multi-node systems; Debug performance bottlenecks and correctness issues in the full stack; Contribute to upstream PyTorch development; Develop benchmarks and profiling workflows; Mentor engineers on performance practices; Engage with customers on training enablement.
Seniority
Principal, hands-on IC with architectural influence
