[DCP] Avoid in-place update and deepcopy during dudpe#149320
Closed
saumishr wants to merge 1 commit intopytorch:mainfrom
Closed
[DCP] Avoid in-place update and deepcopy during dudpe#149320saumishr wants to merge 1 commit intopytorch:mainfrom
saumishr wants to merge 1 commit intopytorch:mainfrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/149320
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit 35c4f4a with merge base 6c7d841 ( UNSTABLE - The following jobs are marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
meetv18
approved these changes
Mar 17, 2025
Contributor
There was a problem hiding this comment.
nit:
Suggested change
| plan_to_item_indexes: list[set[MetadataIndex]] = [set(item.index for item in plan.items) for plan in all_plans] | |
| plan_to_item_indices: list[set[MetadataIndex]] = [set(item.index for item in plan.items) for plan in all_plans] |
4f984ef to
d199cad
Compare
d199cad to
f21e63d
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
1 similar comment
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
saumishr
added a commit
to saumishr/pytorch
that referenced
this pull request
Mar 17, 2025
Summary: Pull Request resolved: pytorch#149320 Avoid in-place update and deepcopy during dudpe. Deepcopy becomes prohibitively expensive with models having a huge number of FQNs. This was manifestd in the Ads 2K experiment as well. Here are the results from the TextRay model in Mitra: #### Control job with deepcopy regression: - https://fburl.com/scuba/dai_modelstore/ac5so0i0 First save ~24.8s Global step latency is ~7-8s https://fburl.com/scuba/pytorch_dcp_logging/kdmwiemk Test job with the new fix to avoid deepcopy: - https://fburl.com/scuba/dai_modelstore/aqi93x3a First save is ~21s - global step latency ~2s https://fburl.com/scuba/pytorch_dcp_logging/w7muzr84 Test Plan: ``` buck test 'fbcode//mode/dev-nosan' fbcode//caffe2/test/distributed/checkpoint:test_planner ``` https://www.internalfb.com/intern/testinfra/testrun/3940649945104822 Reviewed By: MeetVadakkanchery Differential Revision: D71245218
407ae55 to
072a3f8
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
072a3f8 to
bb30ade
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
bb30ade to
008915c
Compare
008915c to
7c3abf3
Compare
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
7c3abf3 to
621a7ec
Compare
Summary: Pull Request resolved: pytorch#149320 Avoid in-place update and deepcopy during dudpe. Deepcopy becomes prohibitively expensive with models having a huge number of FQNs. This was manifestd in the Ads 2K experiment as well. Here are the results from the TextRay model in Mitra: #### Control job with deepcopy regression: - https://fburl.com/scuba/dai_modelstore/ac5so0i0 First save ~24.8s Global step latency is ~7-8s https://fburl.com/scuba/pytorch_dcp_logging/kdmwiemk Test job with the new fix to avoid deepcopy: - https://fburl.com/scuba/dai_modelstore/aqi93x3a First save is ~21s - global step latency ~2s https://fburl.com/scuba/pytorch_dcp_logging/w7muzr84 Test Plan: ``` buck test 'fbcode//mode/dev-nosan' fbcode//caffe2/test/distributed/checkpoint:test_planner ``` https://www.internalfb.com/intern/testinfra/testrun/3940649945104822 Reviewed By: MeetVadakkanchery Differential Revision: D71245218
Contributor
|
This pull request was exported from Phabricator. Differential Revision: D71245218 |
621a7ec to
35c4f4a
Compare
Contributor
|
@pytorchbot merge -i (Initiating merge automatically since Phabricator Diff has merged, merging with -i because oss signals were bypassed internally) |
Collaborator
Merge startedYour change will be merged while ignoring the following 2 checks: pull / cuda12.4-py3.10-gcc9-sm75 / test (pr_time_benchmarks, 1, 1, linux.g4dn.metal.nvidia.gpu), pull / linux-jammy-py3-clang12-executorch / test (executorch, 1, 1, linux.2xlarge) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary:
Avoid in-place update and deepcopy during dudpe. Deepcopy becomes prohibitively expensive with models having a huge number of FQNs. This was manifestd in the Ads 2K experiment as well. Here are the results from the TextRay model in Mitra:
Control job with deepcopy regression:
First save ~24.8s
Global step latency is ~7-8s
Test job with the new fix to avoid deepcopy:
First save is ~21s
global step latency ~2s
Test Plan:
https://www.internalfb.com/intern/testinfra/testrun/3940649945104822
Differential Revision: D71245218
cc @LucasLLC @pradeepfn @H-Huang @awgu @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o