フリー問題

NVIDIA-Certified Associate: AI Infrastructure and Operations のフリー問題 5 / 20 問目

問題文

A platform team is sizing two environments. One will train a large language model from a checkpoint over several weeks. The other will answer user requests from the finished model. The team wants to know which resource becomes the distinguishing requirement of the training environment. Which resource is it?

選択肢

  1. Single-request response time, since each training step must return within a fixed time budget to keep the schedule.
  2. Accelerator memory and inter-accelerator bandwidth, because the backward pass keeps gradients and optimizer state alongside the weights.
  3. Host processor clock speed, because the optimizer update is executed on the host between steps.
  4. Storage capacity for the finished weights, because the trained artifact is the largest object the whole run produces.

解答・解説を確認するには

正解と解説の確認、回答の記録には無料登録が必要です。登録すると演習モードでフリー問題に回答し、正誤と解説をその場で確認できます。