Design and build a comprehensive performance testing platform for evaluating LLM inference workloads across GPU clusters
Define and implement the benchmarking methodology, metrics, and test suites that measure latency, throughput, memory utilization, power consumption, and model accuracy
Establish baseline performance for unoptimized models (Llama 3.2 70B, DeepSeek, etc.) and validate post-optimization improvements
Develop automated testing pipelines for continuous performance validation across compiler releases and model updates
Investigate performance bottlenecks using profiling tools (ROCm profilers, GPU traces, system-level monitoring) and work with the compiler team to drive optimizations
Create dashboards and reporting that provide clear visibility into performance trends, regressions, and wins
Collaborate cross-functionally with compiler engineers, ML engineers, and DevOps to ensure performance testing is integrated into our development workflow
Document best practices for performance testing and optimization of ML workloads on GPU hardware

7+ years of experience in performance engineering, benchmarking, or systems engineering roles
Deep understanding of ML inference workloads, particularly transformer-based models and LLMs
Hands-on experience with GPU programming and optimization (CUDA, ROCm, or similar)
Strong programming skills in Python and C/C++
Proven track record of building performance testing infrastructure or benchmarking platforms from scratch
Experience with ML frameworks (PyTorch, TensorFlow, ONNX Runtime, vLLM, TensorRT-LLM, etc.)
Proficiency with profiling and debugging tools for GPU workloads
Strong analytical skills with the ability to design experiments, analyze results, and communicate findings clearly
Experience with CI/CD systems and test automation frameworks

Senior ML Performance Engineer

Key skills