Member of Technical Staff - Pretraining / Inference Optimization

Familiarity with effective techniques in optimizing inference and training workloads. Knowledge in optimizing for both memory-bound and compute-bound operations. Understanding of GPU memory hierarchy and computation capabilities. Deep understanding of efficient attention algorithms. Experience implementing forward and backward Triton kernels with a focus on correctness and floating-point errors. Ability to integrate custom-written kernels into a PyTorch framework using tools like pybind.

Finding ideal training strategies for various model sizes and compute loads. Profiling, debugging, and optimizing single and multi-GPU operations using tools such as Nsight. Reasoning about speed and quality trade-offs of quantization for model inference. Developing and improving low-level kernel optimizations for state-of-the-art inference and training. Innovating new ideas to maximize GPU performance.