[AutoParallel]Fix get_group method of processmesh #73099
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
|
本地单测测试结果如图所示： |
|
Sorry to inform you that f056bd8's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
… fix_get_group
Codecov Report❌ Patch coverage is
❌ Your patch status has failed because the patch coverage (57.14%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage. Additional details and impacted files@@ Coverage Diff @@
## develop #73099 +/- ##
==========================================
Coverage ? 57.14%
==========================================
Files ? 1
Lines ? 14
Branches ? 0
==========================================
Hits ? 8
Misses ? 6
Partials ? 0 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
| ) | ||
|
|
||
| return parallel_group_map[dim_name]() | ||
| existing_group = None |
There was a problem hiding this comment.
冗余变量，在if set(group.ranks) == set(self._process_ids)分支下直接返回 group就好
|
|
||
|
|
||
| if __name__ == "__main__": | ||
| test_dp_parallel() |
There was a problem hiding this comment.
fleet_test_xx，都是类似的文件，是否可以想办法合并成一个文件
|
/re-run all-failed |
|
补充本地测试覆盖效果：
|
PR Category
Auto Parallel
PR Types
Bug fixes
Description
ProcessMesh 的 get_group 方法，在实际使用时会重复创建通信组，导致显存爆炸，或者通信过程中会存在非预期的错误。因此在通过 ProcessMesh 转换为动手使用的通信组是得慎重，若此时已经存在与mesh.get_group方法相同mesh的group，使用get_group 应该获取该group，而不是直接创建新的通信组。