Skip to content

Improve GPU utilization in ray mode #670

Open
@pan-x-c

Description

@pan-x-c

Search before continuing 先搜索,再继续

  • I have searched the Data-Juicer issues and found no similar feature requests. 我已经搜索了 Data-Juicer 的 issue 列表但是没有发现类似的功能需求。

Description 描述

In the current version of Data-Juicer, operators in ray mode cannot fully utilize the GPU resources of the cluster. Specifically, GPU resources are shared by all operators, resulting in each operator being unable to utilize the resources of the entire cluster during operation.

We can try to collocate the models required by multiple operators on the same GPU through Ray PlacementGroup to improve parallelism.

Use case 使用场景

No response

Additional 额外信息

No response

Are you willing to submit a PR for this feature? 您是否乐意为此功能提交一个 PR?

  • Yes I'd like to help by submitting a PR! 是的!我愿意提供帮助并提交一个PR!

Metadata

Metadata

Assignees

Labels

dj:distissues/PRs about distributed data processingenhancementNew feature or request

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions