[compiled autograd] Always proxy autograd.Function nodes; handle AOT backwards#143405
Closed
zou3519 wants to merge 20 commits intogh/zou3519/1109/basefrom
Closed
[compiled autograd] Always proxy autograd.Function nodes; handle AOT backwards#143405zou3519 wants to merge 20 commits intogh/zou3519/1109/basefrom
zou3519 wants to merge 20 commits intogh/zou3519/1109/basefrom
Conversation
…backwards WIP [ghstack-poisoned]
This was referenced Dec 17, 2024
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/143405
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit db79cf9 with merge base 54e2f4b ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…handle AOT backwards" WIP cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
…handle AOT backwards" WIP cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
…handle AOT backwards" WIP cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
zou3519
commented
Dec 20, 2024
…handle AOT backwards" WIP cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
…handle AOT backwards" WIP cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
This was referenced Jan 3, 2025
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests [ghstack-poisoned]
Collaborator
|
@zou3519 your PR has been successfully reverted. |
zou3519
added a commit
that referenced
this pull request
Jan 23, 2025
…backwards (#143405) We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests Pull Request resolved: #143405 Approved by: https://github.com/jansel, https://github.com/xmfan ghstack dependencies: #143296, #143304, #143387
zou3519
added a commit
that referenced
this pull request
Jan 23, 2025
#143417) The previous PRs built up to this. We change compiled autograd's initial trace to stop baking in metadata. While tracing, we allocate some weirdly shaped tensors that we can put proxies on. The initial trace should not be accessing any metadata of these tensors (it will likely error out if it does because of how weird the shapes are). This involved fixing some various sites where we do specialize on the metadata, like: - we change CopySlices's apply_with_saved to proxy some calls into the graph (this change is fairly hard to split out by itself). - we stop calling InputBuffer::add - we delete the weird metadata from the graph so that no graph passes can make use of it. Test Plan: - tests Pull Request resolved: #143417 Approved by: https://github.com/jansel, https://github.com/xmfan ghstack dependencies: #143296, #143304, #143387, #143405
zou3519
added a commit
that referenced
this pull request
Jan 23, 2025
Compiled autograd's initial trace traces through the AOTBackward epilogue. The Tensor Subclass code is not traceable. This PR changes it so that when we see Tensor Subclass constructors, we proxy nodes for their construction into the graph. Test Plan: - New basic test with TwoTensor - Existing tests Pull Request resolved: #144115 Approved by: https://github.com/jansel, https://github.com/xmfan, https://github.com/bdhirsh ghstack dependencies: #143296, #143304, #143387, #143405, #143417
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
zou3519
added a commit
that referenced
this pull request
Jan 24, 2025
…backwards (#143405) We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests Pull Request resolved: #143405 Approved by: https://github.com/jansel, https://github.com/xmfan ghstack dependencies: #143296, #143304, #143387 ghstack-source-id: 82c8362
zou3519
added a commit
that referenced
this pull request
Jan 24, 2025
#143417) The previous PRs built up to this. We change compiled autograd's initial trace to stop baking in metadata. While tracing, we allocate some weirdly shaped tensors that we can put proxies on. The initial trace should not be accessing any metadata of these tensors (it will likely error out if it does because of how weird the shapes are). This involved fixing some various sites where we do specialize on the metadata, like: - we change CopySlices's apply_with_saved to proxy some calls into the graph (this change is fairly hard to split out by itself). - we stop calling InputBuffer::add - we delete the weird metadata from the graph so that no graph passes can make use of it. Test Plan: - tests Pull Request resolved: #143417 Approved by: https://github.com/jansel, https://github.com/xmfan ghstack dependencies: #143296, #143304, #143387, #143405 ghstack-source-id: f0bdd63
zou3519
added a commit
that referenced
this pull request
Jan 24, 2025
Compiled autograd's initial trace traces through the AOTBackward epilogue. The Tensor Subclass code is not traceable. This PR changes it so that when we see Tensor Subclass constructors, we proxy nodes for their construction into the graph. Test Plan: - New basic test with TwoTensor - Existing tests Pull Request resolved: #144115 Approved by: https://github.com/jansel, https://github.com/xmfan, https://github.com/bdhirsh ghstack dependencies: #143296, #143304, #143387, #143405, #143417 ghstack-source-id: b79847c
…handle AOT backwards" We will always proxy autograd.Function nodes in compiled autograd's initial graph capture (previously there was an option to proxy vs trace into the autograd.Function) We have some requirements for the AOTBackward. Compiled Autograd runs accumulate grad reordering passes on the AOTBackward graph directly after the initial graph capture, so we can't just proxy a single node for it. Instead, we: - proxy the AOTBackward prologue function into the CA graph - copy-paste the AOTBackward graph into the CA graph - trace directly through the epilogue (the traced nodes go into the CA graph). Tracing through the epilogue is safe (assuming no Tensor subclasses) because the only thing the epilogue does is drop some outputs. The Tensor subclass situation was already broken so this doesn't regress anything but this PR sets it up to be fixed (in a followup, where we will proxy "make_subclass" calls into the graph from the epilogue). Test Plan: - existing tests cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan [ghstack-poisoned]
zou3519
added a commit
that referenced
this pull request
Jan 24, 2025
Compiled autograd's initial trace traces through the AOTBackward epilogue. The Tensor Subclass code is not traceable. This PR changes it so that when we see Tensor Subclass constructors, we proxy nodes for their construction into the graph. Test Plan: - New basic test with TwoTensor - Existing tests Pull Request resolved: #144115 Approved by: https://github.com/jansel, https://github.com/xmfan, https://github.com/bdhirsh ghstack dependencies: #143296, #143304, #143387, #143405, #143417 ghstack-source-id: 8876f26
facebook-github-bot
pushed a commit
that referenced
this pull request
Jan 24, 2025
Summary: This PR squashes together the following commits: #144115 #143417 #143405 #143387 #143304 #143296 This is a refactor of compiled autograd to use "functional autograd". The end goal is that it gets compiled autograd's initial capture to stop specializing on Tensor metadata, therefore allowing compiled autograd to better handle Tensor subclasses. For more information, please read the commit messages for each PR. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan bypass-github-export-checks failing CI test is bypassable on OSS but cannot be retried. Reviewed By: jansel, xmfan, bdhirsh Differential Revision: D68120850 Pulled By: zou3519
zou3519
added a commit
that referenced
this pull request
Jan 26, 2025
Summary: This PR squashes together the following commits: #144115 #143417 #143405 #143387 #143304 #143296 This is a refactor of compiled autograd to use "functional autograd". The end goal is that it gets compiled autograd's initial capture to stop specializing on Tensor metadata, therefore allowing compiled autograd to better handle Tensor subclasses. For more information, please read the commit messages for each PR. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov xmfan bypass-github-export-checks failing CI test is bypassable on OSS but cannot be retried. Reviewed By: jansel, xmfan, bdhirsh Differential Revision: D68120850 Pulled By: zou3519
pytorchmergebot
pushed a commit
that referenced
this pull request
Jan 27, 2025
This PR squashes together the following commits: #144115 #143417 #143405 #143387 #143304 #143296 This is a refactor of compiled autograd to use "functional autograd". The end goal is that it gets compiled autograd's initial capture to stop specializing on Tensor metadata, therefore allowing compiled autograd to better handle Tensor subclasses. For more information, please read the commit messages for each PR. Pull Request resolved: #144707 Approved by: https://github.com/bdhirsh, https://github.com/xmfan, https://github.com/jansel
nWEIdia
pushed a commit
to nWEIdia/pytorch
that referenced
this pull request
Jan 27, 2025
This PR squashes together the following commits: pytorch#144115 pytorch#143417 pytorch#143405 pytorch#143387 pytorch#143304 pytorch#143296 This is a refactor of compiled autograd to use "functional autograd". The end goal is that it gets compiled autograd's initial capture to stop specializing on Tensor metadata, therefore allowing compiled autograd to better handle Tensor subclasses. For more information, please read the commit messages for each PR. Pull Request resolved: pytorch#144707 Approved by: https://github.com/bdhirsh, https://github.com/xmfan, https://github.com/jansel
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
5 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.
Stack from ghstack (oldest at bottom):
We will always proxy autograd.Function nodes in compiled autograd's
initial graph capture (previously there was an
option to proxy vs trace into the autograd.Function)
We have some requirements for the AOTBackward. Compiled Autograd runs
accumulate grad reordering passes on the AOTBackward graph directly
after the initial graph capture, so we can't just proxy a single node for it.
Instead, we:
graph).
Tracing through the epilogue is safe (assuming no Tensor subclasses)
because the only thing the epilogue does is drop some outputs. The
Tensor subclass situation was already broken so this doesn't regress
anything but this PR sets it up to be fixed (in a followup, where we
will proxy "make_subclass" calls into the graph from the epilogue).
Test Plan:
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang @aakhundov @xmfan