Conversation
[ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit 085640e (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions to the (internal) Dr. CI Users group. |
[ghstack-poisoned]
[ghstack-poisoned]
soulitzer
left a comment
There was a problem hiding this comment.
Looks pretty good overall! Added some minor comments.
|
In terms of further testing, what if you had a wrapper function that unpacked the returned dual tensor from the original function, perturbs the tangent by some eps, and then does make_dual again before returning? |
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
[ghstack-poisoned]
|
Sorry for the important update but implementing the OpInfo showed some limitations in the implementation that was here and that it wasn't working for most complex workloads. |
ghstack-source-id: 9ed86ab Pull Request resolved: pytorch#57633
[ghstack-poisoned]
|
This is ready for final review |
[ghstack-poisoned]
soulitzer
left a comment
There was a problem hiding this comment.
Added a couple more comments. Might be good to make a quick PR testing that CI passes with slow_mode=False if you haven't already.
[ghstack-poisoned]
|
I tested locally with slow gradcheck flag. How do you do a PR to test with slow? Modify the global setting to False by default and push that? Or there is another way? |
[ghstack-poisoned]
ghstack-source-id: ad301a5 Pull Request resolved: pytorch#57633
|
@soulitzer Running the slow gradcheck version in #57952 |
[ghstack-poisoned]
ghstack-source-id: 68f15a0 Pull Request resolved: pytorch#57633
[ghstack-poisoned]
|
@albanD has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
soulitzer
left a comment
There was a problem hiding this comment.
LGTM! Thanks for the addition!
|
Unlanding this stack as it appears to have hit a logical merge conflict and is breaking multiple builds on the HUD. Sample tests broken: test_forward_mode_AD_narrow_cpu_complex128 |
|
This pull request has been reverted by 2d7d692. |
Summary: Pull Request resolved: pytorch#57633 Test Plan: Imported from OSS Reviewed By: agolynski Differential Revision: D28387765 Pulled By: albanD fbshipit-source-id: ed15049b5bdacca54f775b50ef166d540ba0b847
Stack from ghstack:
Differential Revision: D28387765