Principal / Senior GPU Software Performance Engineer, Post-Training

What you'll do

• Lead performance optimization for finetuning and RL training on AMD GPUs.
• Enhance throughput, memory efficiency, and stability in multi-GPU/multi-node setups.
• Develop efficient kernels and graph-level optimizations for deep learning frameworks.
• Profile and resolve bottlenecks using standard tooling to prevent CI regressions.
• Ship reproducible pipelines and documentation for internal and external use.

What we're looking for

• Proven GPU performance engineering for deep learning (ROCm/HIP, Triton, etc.)
• Hands-on experience with SFT, LoRA, and RL-based training at scale
• Strong PyTorch expertise including torch.distributed, FSDP/ZeRO or equivalent
• Proficient in Python and C++; capable of reading/writing kernels
• Experience optimizing multi-GPU/multi-node training and communication patterns
• Track record of profiling, diagnosing, and resolving performance bottlenecks