[BE][FSDP] Fix that MP config not being passed to FSDP #80869
Conversation
[ghstack-poisoned]
awgu
requested review from
mrshenli,
pritamdamania87,
zhaojuanmao,
rohan-varma and
H-Huang
as code owners
🔗 Helpful links
✅ No Failures (0 Pending)As of commit 6a59cfc (more details on the Dr. CI page): Expand to see more💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
facebook-github-bot
added
the
oncall: distributed
|
@pytorchbot rebase |
|
@pytorchbot successfully started a rebase job. Check the current status here |
The mixed precision config was being passed to the wrapped module instead of propagating to the FSDP constructor. This fix will become obsolete with the `FSDPTestModel` PR higher in the stack, but it is still worth fixing by itself. [ghstack-poisoned]
|
Successfully rebased |
The mixed precision config was being passed to the wrapped module instead of propagating to the FSDP constructor. This fix will become obsolete with the `FSDPTestModel` PR higher in the stack, but it is still worth fixing by itself. [ghstack-poisoned]
|
@pytorchbot merge |
|
@pytorchbot successfully started a merge job. Check the current status here |
|
Merge failed due to Refusing to merge as mandatory check(s) pull failed for rule superuser |
The mixed precision config was being passed to the wrapped module instead of propagating to the FSDP constructor. This fix will become obsolete with the `FSDPTestModel` PR higher in the stack, but it is still worth fixing by itself. [ghstack-poisoned]
|
@pytorchbot merge |
|
@pytorchbot successfully started a merge job. Check the current status here |
|
Hey @awgu. |
Summary: The mixed precision config was being passed to the wrapped module instead of propagating to the FSDP constructor. This fix will become obsolete with the `FSDPTestModel` PR higher in the stack, but it is still worth fixing by itself. Pull Request resolved: #80869 Approved by: https://github.com/rohan-varma Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/f583f81b6e7c9e61151dd263e487ce65af54eab3 Reviewed By: mehtanirav Differential Revision: D37749398 Pulled By: awgu fbshipit-source-id: be0a655e0676707904f68e1abc630e107146355f
Stack from ghstack:
test_mixed_precision_e2e_full_shard()#80915 [BE][FSDP] Subtest prefetching intest_mixed_precision_e2e_full_shard()test_fsdp_core.py#80908 [BE][FSDP] Subtest prefetching intest_fsdp_core.pyFSDPTestModelinterface #80873 [BE][FSDP] IntroduceFSDPTestModelinterface_get_full_detached_param()#80871 [BE][FSDP] Retire_get_full_detached_param()common_fsdp.pyimports #80870 [BE][FSDP] Sortcommon_fsdp.pyimportstorch.cuda.synchronize()#80868 [BE][FSDP] Remove unneededtorch.cuda.synchronize()The mixed precision config was being passed to the wrapped module instead of propagating to the FSDP constructor. This fix will become obsolete with the
FSDPTestModelPR higher in the stack, but it is still worth fixing by itself.