An improved heuristic for operator reordering for peak memory + debugging logs#161810
Closed
xuanzhang816 wants to merge 2 commits intogh/xuanzhang816/25/basefrom
Closed
An improved heuristic for operator reordering for peak memory + debugging logs#161810xuanzhang816 wants to merge 2 commits intogh/xuanzhang816/25/basefrom
xuanzhang816 wants to merge 2 commits intogh/xuanzhang816/25/basefrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/161810
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit f54b3b3 with merge base 65d642d ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Contributor
Author
|
@pytorchbot label "topic: not user facing" |
yf225
approved these changes
Sep 12, 2025
Contributor
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Collaborator
Merge failedReason: Command Details for Dev Infra teamRaised by workflow job |
Contributor
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
…ging logs (pytorch#161810) Revisiting the idea in pytorch#140195 For the lpmf algorithm in the memory reorder pass, in some cases, when all the nodes that can be scheduled are quite large, it is beneficial to switch the scheduling strategy. So instead of using size as the criterion, we choose a node that can unlock more nodes to become schedulable by analyzing their successor nodes. For an internal use case, we observe up to 20 GiB memory difference and here are the before and after memory snapshot. More information can be found in [D81270682](https://www.internalfb.com/diff/D81270682) (internal only). <img width="348" height="227" alt="image" src="https://github.com/user-attachments/assets/fb71e840-1508-44ed-bc9d-5eb4d364607d" /> In addition, add the functionality to upload the graph to tlparse for offline debugging. The format of the json is in consistency with the simulator [here](https://fburl.com/code/3l3d3qi4) (internal only). Pull Request resolved: pytorch#161810 Approved by: https://github.com/yf225
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…ging logs (pytorch#161810) Revisiting the idea in pytorch#140195 For the lpmf algorithm in the memory reorder pass, in some cases, when all the nodes that can be scheduled are quite large, it is beneficial to switch the scheduling strategy. So instead of using size as the criterion, we choose a node that can unlock more nodes to become schedulable by analyzing their successor nodes. For an internal use case, we observe up to 20 GiB memory difference and here are the before and after memory snapshot. More information can be found in [D81270682](https://www.internalfb.com/diff/D81270682) (internal only). <img width="348" height="227" alt="image" src="https://github.com/user-attachments/assets/fb71e840-1508-44ed-bc9d-5eb4d364607d" /> In addition, add the functionality to upload the graph to tlparse for offline debugging. The format of the json is in consistency with the simulator [here](https://fburl.com/code/3l3d3qi4) (internal only). Pull Request resolved: pytorch#161810 Approved by: https://github.com/yf225
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…ging logs (pytorch#161810) Revisiting the idea in pytorch#140195 For the lpmf algorithm in the memory reorder pass, in some cases, when all the nodes that can be scheduled are quite large, it is beneficial to switch the scheduling strategy. So instead of using size as the criterion, we choose a node that can unlock more nodes to become schedulable by analyzing their successor nodes. For an internal use case, we observe up to 20 GiB memory difference and here are the before and after memory snapshot. More information can be found in [D81270682](https://www.internalfb.com/diff/D81270682) (internal only). <img width="348" height="227" alt="image" src="https://github.com/user-attachments/assets/fb71e840-1508-44ed-bc9d-5eb4d364607d" /> In addition, add the functionality to upload the graph to tlparse for offline debugging. The format of the json is in consistency with the simulator [here](https://fburl.com/code/3l3d3qi4) (internal only). Pull Request resolved: pytorch#161810 Approved by: https://github.com/yf225
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…ging logs (pytorch#161810) Revisiting the idea in pytorch#140195 For the lpmf algorithm in the memory reorder pass, in some cases, when all the nodes that can be scheduled are quite large, it is beneficial to switch the scheduling strategy. So instead of using size as the criterion, we choose a node that can unlock more nodes to become schedulable by analyzing their successor nodes. For an internal use case, we observe up to 20 GiB memory difference and here are the before and after memory snapshot. More information can be found in [D81270682](https://www.internalfb.com/diff/D81270682) (internal only). <img width="348" height="227" alt="image" src="https://github.com/user-attachments/assets/fb71e840-1508-44ed-bc9d-5eb4d364607d" /> In addition, add the functionality to upload the graph to tlparse for offline debugging. The format of the json is in consistency with the simulator [here](https://fburl.com/code/3l3d3qi4) (internal only). Pull Request resolved: pytorch#161810 Approved by: https://github.com/yf225
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
3 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.
Revisiting the idea in #140195
For the lpmf algorithm in the memory reorder pass, in some cases, when all the nodes that can be scheduled are quite large, it is beneficial to switch the scheduling strategy. So instead of using size as the criterion, we choose a node that can unlock more nodes to become schedulable by analyzing their successor nodes.
For an internal use case, we observe up to 20 GiB memory difference and here are the before and after memory snapshot. More information can be found in D81270682 (internal only).
In addition, add the functionality to upload the graph to tlparse for offline debugging. The format of the json is in consistency with the simulator here (internal only).
Stack from ghstack (oldest at bottom):
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben