-
Notifications
You must be signed in to change notification settings - Fork 5.8k
[Auto Parallel] Add spmd_rule about sharding on the same tensor dim by many mesh dim for reshape #74352
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
[Auto Parallel] Add spmd_rule about sharding on the same tensor dim by many mesh dim for reshape #74352
Conversation
你的PR提交成功,感谢你对开源项目的贡献! |
512ab0f
to
e3a415f
Compare
e3a415f
to
096843e
Compare
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #74352 +/- ##
==========================================
Coverage ? 95.15%
==========================================
Files ? 3
Lines ? 165
Branches ? 0
==========================================
Hits ? 157
Misses ? 8
Partials ? 0 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
463d98f
to
92c910e
Compare
/re-run all-failed |
/re-run all-failed |
…y many mesh dim for reshape (PaddlePaddle#74352) * Add spmd_rule about sharding on the same tensor dim by many mesh dim for reshape * Fix order * refine code && tests case * fix typos * fix typos * Fix paddle::get usage
PR Category
Auto Parallel
PR Types
New features
Description
目前在合并的时候,是把最左边的相邻切分合并到第一个维度进行切分,如果第一个维度不够切,就后续就进行复制,现在的处理是遇到输入切多刀的时候,不够切对应 mesh shape 的乘积,就整个维度的切分不保留,进行复制,就算够切多刀其中的一个维度,{{0},{1,2}} 的时候, 0,1 够切最左边的维度,0,1,2 切不了,现在就会是 {{0}, {}}。
拆分和合并的逻辑相同,都是在多刀在最左边的维度,其余复制,还需要考虑 split 切成的维度够不够切和是否能整除。