Skip to content

[NPU] Add label_smooth_op #34828

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 3 commits into from
Sep 2, 2021
Merged

[NPU] Add label_smooth_op #34828

merged 3 commits into from
Sep 2, 2021

Conversation

@Ray2020BD
Copy link
Contributor

@Ray2020BD Ray2020BD commented Aug 11, 2021 •
edited
Loading

PR types

New features

PR changes

OPs

Describe

[NPU] Add label_smooth op

image
image

@paddle-bot-old
Copy link

paddle-bot-old bot commented Aug 11, 2021

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

qili93
qili93 reviewed Aug 12, 2021
View reviewed changes
Copy link
Contributor

@qili93 qili93 left a comment •
edited
Loading

Choose a reason for hiding this comment

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

关于小OP组合的影响，能帮忙给一个分析报告吗？需要以下信息来分析小OP组合对整体模型的性能和功能上的影响：

(1) 这个OP在大数据输入下对比CPU OP的性能 (考虑到NPU Kernel Launch 的消耗，存在OP性能不如CPU OP的可能)
(2) 这个OP在大数据输入下的显存占用，可以对比huber_loss一个OP实现的显存占用（之前有同学反馈小OP组合会导致显存占用是原来的8倍）

基于以上分析结果，再考虑我们是否能够接受小OP组合的方案，或者需要采用TBE算子开发的方案。

@Ray2020BD
Copy link
Contributor Author

Ray2020BD commented Aug 13, 2021 •
edited
Loading

关于小OP组合的影响，能帮忙给一个分析报告吗？需要以下信息来分析小OP组合对整体模型的性能和功能上的影响：

(1) 这个OP在大数据输入下对比CPU OP的性能 (考虑到NPU Kernel Launch 的消耗，存在OP性能不如CPU OP的可能)
(2) 这个OP在大数据输入下的显存占用，可以对比huber_loss一个OP实现的显存占用（之前有同学反馈小OP组合会导致显存占用是原来的8倍）

基于以上分析结果，再考虑我们是否能够接受小OP组合的方案，或者需要采用TBE算子开发的方案。

已补充测试，目前有cpu vs npu的测试，测试结果显示在shape增大时候npu运算更有优势，显存开销也不大，小op组合的方案性能尚可。

@Ray2020BD Ray2020BD closed this Aug 13, 2021
@Ray2020BD Ray2020BD reopened this Aug 13, 2021
@qili93
Copy link
Contributor

qili93 commented Aug 16, 2021

关于小OP组合的影响，能帮忙给一个分析报告吗？需要以下信息来分析小OP组合对整体模型的性能和功能上的影响：
(1) 这个OP在大数据输入下对比CPU OP的性能 (考虑到NPU Kernel Launch 的消耗，存在OP性能不如CPU OP的可能)
(2) 这个OP在大数据输入下的显存占用，可以对比huber_loss一个OP实现的显存占用（之前有同学反馈小OP组合会导致显存占用是原来的8倍）
基于以上分析结果，再考虑我们是否能够接受小OP组合的方案，或者需要采用TBE算子开发的方案。

已补充测试，目前有cpu vs npu的测试，测试结果显示在shape增大时候npu运算更有优势，显存开销也不大，小op组合的方案性能尚可。

这个分析结果“shape = [2000,4000]时，占用显存最高28061MB。”有点问题呢，NPU单卡最高显存也就15307MB，为什么这里显存可以占用到28061MB呢？

qili93
qili93 previously approved these changes Aug 31, 2021
View reviewed changes
Copy link
Contributor

@qili93 qili93 left a comment

Choose a reason for hiding this comment

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

LGTM

@qili93 qili93 merged commit e57a88b into PaddlePaddle:develop Sep 2, 2021
@Ray2020BD Ray2020BD deleted the npu_label_smooth_2 branch September 2, 2021 11:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@qili93 qili93 qili93 approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@Ray2020BD @qili93