[ao][sparsity] comsability for sparsity and QAT convert #74848
Closed
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
Summary: WIP Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
This was referenced Mar 28, 2022
Contributor
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 0cc5bdc (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).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
Summary: WIP Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
HDCharles
changed the title
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
HDCharles
commented
Mar 28, 2022
HDCharles
requested review from
andrewor14,
dzdang,
jerryzh168,
terrychenism,
vkuzo and
z-a-f
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
HDCharles
added a commit
that referenced
this pull request
Mar 29, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 2153364 Pull Request resolved: #74848
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
HDCharles
added a commit
that referenced
this pull request
Apr 7, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations also added comments to the test code Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 57f9e42 Pull Request resolved: #74848
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
lezcano
reviewed
Apr 7, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations also added comments to the test code Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations also added comments to the test code Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
HDCharles
added a commit
that referenced
this pull request
Apr 7, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations also added comments to the test code Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 53b5776 Pull Request resolved: #74848
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
HDCharles
added a commit
that referenced
this pull request
Apr 8, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 3bf0f42 Pull Request resolved: #74848
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
lezcano
reviewed
Apr 8, 2022
Collaborator
lezcano
left a comment
lezcano
left a comment
There was a problem hiding this comment.
Just one more thing I forgot! Parametrisations may be many-to-one if the right-inverse returns more than one tensor. You can find examples of these in test_multiple_inputs_parametrization. Could you add a test for this case?
torch/nn/utils/parametrize.py
Outdated
|
|
||
| # need to initialize the param in to_module if it doesn't exist already | ||
| if not hasattr(to_module, parameter_name): | ||
| setattr(to_module, parameter_name, deepcopy(from_module.parametrizations[parameter_name].original)) |
Collaborator
There was a problem hiding this comment.
I believe this branch is not tested, is it? Furthermore, I believe it is incorrect. I think it should be
Suggested change
| setattr(to_module, parameter_name, deepcopy(from_module.parametrizations[parameter_name].original)) | |
| setattr(to_module, parameter_name, getattr(from_module, parameter_name)) |
Note that the original parameter may be a tuple if the parametrization is many-to-one, so setting it as an attribute would not be of much use.
Contributor
Author
There was a problem hiding this comment.
yeah, its tested in line 3282 in test_nn.py within test_transfer_parametrizations_and_params, I can split it into another test if you like, it looked like some of the other parametriation tests were grouped in some similar cases so I wasn't sure.
I added your suggested change, and added a test for the many-to-one case which works after some changes.
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
lezcano
approved these changes
Apr 8, 2022
Collaborator
lezcano
left a comment
lezcano
left a comment
There was a problem hiding this comment.
LGTM CI abiding. Thank you for the many improvements and the testing on the parametrisations end!
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240272](https://our.internmc.facebook.com/intern/diff/D35240272) [ghstack-poisoned]
HDCharles
added a commit
that referenced
this pull request
Apr 8, 2022
Summary: The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to move the weight and bias and any associated paramerizations to the new module. This PR also adds tests for transfer_parametrizations_and_params and type_before_parametrizations to test_nn.py and also added comments to the test code for composability. Test Plan: python test/test_ao_sparsity.py TestComposability python test/test_nn.py TestNN Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 0cd91d0 Pull Request resolved: #74848
Contributor
Author
|
@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
Contributor
|
@pytorchbot merge this (Initiating merge automatically since Phabricator Diff has merged) |
Contributor
|
Hey @HDCharles. |
facebook-github-bot
pushed a commit
that referenced
this pull request
Apr 11, 2022
Summary: Pull Request resolved: #74848 The primary issue for enabling sparsity to work with QAT convert (unlike normal quantization convert) is that when the parametrized module undergoes the QAT convert, the parametrizations need to be maintained. If the parametrizations don't get transfered during the convert, the sparsifier would lose its connection to the model. In practice this was handled using the transfer_parametrizations_and_params function to identify all parametrizations on the original module and then move them (and their associated parameters) to the new module. Test Plan: python test/test_ao_sparsity.py TestComposability Imported from OSS Reviewed By: malfet Differential Revision: D35240272 fbshipit-source-id: 08d6a938d5919ba2dfd8490b1c768fafc5b179dd
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):
Summary: The primary issue for enabling sparsity to work with QAT
convert (unlike normal quantization convert) is that when the
parametrized module undergoes the QAT convert, the parametrizations need
to be maintained. If the parametrizations don't
get transfered during the convert, the sparsifier would lose its
connection to the model. In practice this was handled using the
transfer_parametrizations_and_params function to move the weight and
bias and any associated paramerizations to the new module. This PR also adds
tests for transfer_parametrizations_and_params and type_before_parametrizations
to test_nn.py and also added comments to the test code for
composability.
Test Plan: python test/test_ao_sparsity.py TestComposability
python test/test_nn.py TestNN
Reviewers:
Subscribers:
Tasks:
Tags:
Differential Revision: D35240272