Skip to content

Conversation

@zhiqiu
Copy link
Contributor

@zhiqiu zhiqiu commented Apr 28, 2024

PR Category

Auto Parallel

PR Types

Performance

Description

ATT
In some case, transpose can be simplified to reshape, to reduce kernel time.
For example, Tensor a.shape=[1, 1024, 4096], transpose a with perm [1, 0, 2], which is a common case of LLM hidden_states with shape=[b,s,h]
Pcard-73145

@paddle-bot
Copy link

paddle-bot bot commented Apr 28, 2024

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.

@zhiqiu zhiqiu force-pushed the dev/eliminate_transpose branch from 177d35d to a9b0aae Compare April 28, 2024 09:44
zhangbo9674
zhangbo9674 previously approved these changes Apr 28, 2024
@zhangbo9674 zhangbo9674 merged commit 2a2f635 into PaddlePaddle:develop Apr 29, 2024
runzhech pushed a commit to runzhech/Paddle that referenced this pull request Apr 30, 2024
* replace transpose by reshape in some cases

* support transpose backward

* add to engine

* remove program_guard

* add ut
runzhech pushed a commit to runzhech/Paddle that referenced this pull request Apr 30, 2024
* replace transpose by reshape in some cases

* support transpose backward

* add to engine

* remove program_guard

* add ut
co63oc pushed a commit to co63oc/Paddle that referenced this pull request May 6, 2024
* replace transpose by reshape in some cases

* support transpose backward

* add to engine

* remove program_guard

* add ut
hxzd5568 pushed a commit to hxzd5568/Paddle that referenced this pull request May 8, 2024
* replace transpose by reshape in some cases

* support transpose backward

* add to engine

* remove program_guard

* add ut
co63oc pushed a commit to co63oc/Paddle that referenced this pull request May 10, 2024
* replace transpose by reshape in some cases

* support transpose backward

* add to engine

* remove program_guard

* add ut
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