[DRAFT][Reshape] Guard-free reshape for contiguous tensors to avoid data dependent errors. #148742
Conversation
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/148742
Note: Links to docs will display an error until the docs builds have been completed. ❌ 10 New Failures, 4 Unrelated FailuresAs of commit 8bb976d with merge base d363913 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This PR needs a
|
[ghstack-poisoned]
Main reason for refactor is to avoid data dependent error that the default path have. this is because this new path have no checks on sizes [ghstack-poisoned]
Main reason for refactor is to avoid data dependent error that the default path have. this is because this new path have no checks on sizes [ghstack-poisoned]
Main reason for refactor is to avoid data dependent error that the default path have. this is because this new path have no checks on sizes [ghstack-poisoned]
ColinPeppler
reviewed
Mar 7, 2025
| return prims.view_of(a) | ||
|
|
||
| strides = [1] | ||
| for x in reversed(shape[1:]): |
There was a problem hiding this comment.
should this be shape[:-1]?
There was a problem hiding this comment.
haha maybe i am new ish to python no fancy slicing in c++.
I will try it
There was a problem hiding this comment.
you mean why not
[1:][::-1]
I want to remove first element the reverse
Main reason for refactor is to avoid data dependent error that the default path have. this is because this new path have no checks on sizes [ghstack-poisoned]
Main reason for refactor is to avoid data dependent error that the default path have this is because this new path have no checks on sizes. Does this deviate from previous behaviors/or from torch eager? specially when it comes to strides? I need to dig deep on this, I do not have a clear answer. There was in situation where this used to diverge from previous behavior when we reshape the input to itself but i addressed that. In general we have three choices here: 1. use the new logic only for unbacked. 2. use it for all compile 3. use it for eager and compile I do not want to spend time fixing the failing tests if we are going to go with (1) or if the idea is not accepted thats why i would like to have this discussed first. the failures does seems not risky (change expected string or runtime assert ..etc) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]
This was referenced Mar 8, 2025
… to avoid data dependent errors." Main reason for refactor is to avoid data dependent error that the default path have this is because this new path have no checks on sizes. Does this deviate from previous behaviors/or from torch eager? specially when it comes to strides? I need to dig deep on this, I do not have a clear answer. There was in situation where this used to diverge from previous behavior when we reshape the input to itself but i addressed that. In general we have three choices here: 1. use the new logic only for unbacked. 2. use it for all compile 3. use it for eager and compile I do not want to spend time fixing the failing tests if we are going to go with (1) or if the idea is not accepted thats why i would like to have this discussed first. the failures does seems not risky (change expected string or runtime assert ..etc) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]
This was referenced Mar 10, 2025
… to avoid data dependent errors." Main reason for refactor is to avoid data dependent error that the default path have this is because this new path have no checks on sizes. Does this deviate from previous behaviors/or from torch eager? specially when it comes to strides? I need to dig deep on this, I do not have a clear answer. There was in situation where this used to diverge from previous behavior when we reshape the input to itself but i addressed that. In general we have three choices here: 1. use the new logic only for unbacked. 2. use it for all compile 3. use it for eager and compile I do not want to spend time fixing the failing tests if we are going to go with (1) or if the idea is not accepted thats why i would like to have this discussed first. the failures does seems not risky (change expected string or runtime assert ..etc) cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames [ghstack-poisoned]
Divigroup-RAP
pushed a commit
to Divigroup-RAP/PYTORCH
that referenced
this pull request
Apr 22, 2025
ghstack-source-id: e8baa07 Pull Request resolved: pytorch/pytorch#148742
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Stack from ghstack (oldest at bottom):
Main reason for refactor is to avoid data dependent error that the default path have
this is because this new path have no checks on sizes.
Does this deviate from previous behaviors/or from torch eager? specially when it comes to strides?
I need to dig deep on this, I do not have a clear answer. There was in situation where this used to diverge
from previous behavior when we reshape the input to itself but i addressed that.
In general we have three choices here:
I do not want to spend time fixing the failing tests if we are going to go with (1) or if the idea is not accepted
thats why i would like to have this discussed first. the failures does seems not risky (change expected string or runtime assert ..etc)
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames