Что будет проверяться
The Pre-Training team builds and scales the systems and methods that train Grok's foundation models, optimizing multi-GPU training efficiency and experimenting with architecture and data at frontier scale. The work requires deep familiarity with distributed, large-scale neural network training. A technical interview would probe distributed training parallelism strategies, optimizing Model FLOPs Utilization on large clusters, debugging unstable or diverging training runs, and tradeoffs in scaling data, model size, and compute under a fixed budget.
ExoForm не аффилирован с xAI. Это независимая тренировочная страница.