Что будет проверяться
Training Performance Engineers maximize the efficiency, speed, and hardware utilization of OpenAI's large-scale training runs, profiling and eliminating bottlenecks across compute, memory, and the network fabric. They work hands-on with communication libraries (NCCL, MPI, UCX), checkpointing, and large-scale data loading on multi-thousand-GPU clusters. A technical interview would probe GPU architecture and the memory hierarchy, profiling and roofline analysis, collective-communication patterns, and how to diagnose why a distributed run is achieving low Model FLOPs Utilization.
ExoForm не аффилирован с OpenAI. Это независимая тренировочная страница.