Skip to content
Discussion options

You must be logged in to vote

目前 OpenCompass 进行任务切分的默认逻辑有更新:

  • 传参 --max-num-workers 的值是 n,假设待测试的有 M 个模型和 D 个数据集
  • 所有的 (模型、数据集) 对均会被拆分成 n 份,每一份都只有 1 / n 份的数据。然后每个模型对应的切分任务会被均匀地分成 n 个 task,共计 n * M 个 task,然后用 n 个 worker 去并发地跑。
  • 一般来说一个 worker 对应一块 GPU,该对应关系可以用 --max-workers-per-gpu 控制。
  • 如果 GPU 数目小于 --max-num-workers,那么就会有排队。

Replies: 1 comment 2 replies

Comment options

You must be logged in to vote
2 replies
@airaria
Comment options

@disperaller
Comment options

Answer selected by airaria
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
3 participants