Skip to content

Conversation

@miaobyte
Copy link
Contributor

No description provided.

@miaobyte
Copy link
Contributor Author

miaobyte commented Apr 17, 2025

deepx改为

py侧的函数,尽可能前向传播inplace化,从而让推理,训练的前向最省内存

autograd交给中层去做,autograder先接受init序列,再接收forward的整组IR序列

和pytorch不同,deepx相信autograder能把forward的IR序列,先修改为非inplace的IR序列,以此根据autograder内IRmap注册的反向算子序列,生成反向计算图及IR序列,再根据IR序列的关系,尽可能尝试inplace化,从而达到反向传播也最省内存

@miaobyte miaobyte merged commit 4a223b9 into array2d:main Apr 17, 2025
3 checks passed
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.

1 participant