Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Semi-Auto] Add split infer_backward rule #56508

Merged
merged 2 commits into from
Sep 12, 2023

Conversation

pkuzyc
Copy link
Contributor

@pkuzyc pkuzyc commented Aug 21, 2023

PR types

Function optimization

PR changes

Others

Description

Pcard-70448

Add infer_backward rule for split to infer inputs' dims mappings from outputs':

  1. the splitted axis cannot be sharded.
  2. for the other axes set the input's dims mapping equal to the corresponding outputs' axes.

@paddle-bot
Copy link

paddle-bot bot commented Aug 21, 2023

你的PR提交成功,感谢你对开源项目的贡献!
请关注后续CI自动化测试结果,详情请参考Paddle-CI手册
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@paddle-ci-bot
Copy link

paddle-ci-bot bot commented Sep 1, 2023

Sorry to inform you that 5873a5b's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.


std::vector<TensorDistAttr> output_dist_attrs;
for (int64_t i = 0; i < noutputs; i++) {
output_dist_attrs.emplace_back(output_specs[i].dist_attr());
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

problematic, if output tensor split_axis is sharded, it should be recover to replicated.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done


# num_or_sections = 2, axis = 1
# [0, -1, -1], [0, -1, -1] --> [0, -1, -1], [0, -1, -1], [0, -1, -1]
# (outputs --> input, outputs)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lack case where one or multiple output tensors are sharded on split_axis.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

@pkuzyc pkuzyc marked this pull request as draft September 7, 2023 13:41
@pkuzyc pkuzyc marked this pull request as ready for review September 7, 2023 13:41
Copy link
Contributor

@JZ-LIANG JZ-LIANG left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@JZ-LIANG JZ-LIANG merged commit 2d17239 into PaddlePaddle:develop Sep 12, 2023
@pkuzyc pkuzyc deleted the split_backward branch September 27, 2023 10:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants