compiled autograd: default fakeify backward inputs with static shapes instead of duck sizing#133581
Closed
bdhirsh wants to merge 1 commit intogh/bdhirsh/607/basefrom
Closed
compiled autograd: default fakeify backward inputs with static shapes instead of duck sizing#133581bdhirsh wants to merge 1 commit intogh/bdhirsh/607/basefrom
bdhirsh wants to merge 1 commit intogh/bdhirsh/607/basefrom
Conversation
… instead of duck sizing [ghstack-poisoned]
This was referenced Aug 15, 2024
Closed
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/133581
Note: Links to docs will display an error until the docs builds have been completed. ❌ 40 New Failures, 1 Cancelled Job, 1 Unrelated FailureAs of commit 64d034e with merge base 454713f ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following job failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Member
|
Still don't fully understand the change, changing the fakification logic shouldn't do anything:
|
Contributor
|
@bdhirsh I have actually completely forgotten what exactly we talked about in our 1:1, so you had better it write it down here :P |
Member
|
The compiled autograd failures are both due to the GmWrapper inputs not being unflattened properly in the previous PR. Tests passed for me if you do something like If the traced graph is always the same between tracing with static FakeTensor vs dynamic ones (is this is true?), then this approach should be okay |
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
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.
This fixes the problem described here, by tweaking compiled autograd to default to fakifying with static shapes instead of defaulting to duck sizing.
I think there are some (independent) dynamic shape + compiled autograd issues though, detailed here #133575 (cc @ezyang @penguinwu @bobrenjc93 @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames @xmfan @yf225 @rec )
Stack from ghstack (oldest at bottom):
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames