[ao][sparsity] make sparsity compose with PTQ convert #74846

HDCharles · 2022-03-28T18:33:10Z

Stack from ghstack (oldest at bottom):

Summary: This PR primarily allows the PTQ convert function to work with
parametrized modules. Given that the parametrized weight is what is used
by default in convert, as long as sparsifier.step() has already been
called, the converted model will use the sparisified weights. There is
currently no way to handle things if sparsifier.step() has not been
called. Lastly, added the has_no_children_ignoring_parametrizations function because
parametrized modules no longer look like leaves due to the
parametrizations module attached to them

Test Plan: python test/test_ao_sparsity.py TestComposability

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: D35240275

Summary: making sure it works for PTQ even if squash_mask isn't applied Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

facebook-github-bot · 2022-03-28T18:33:18Z

🔗 Helpful links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/74846
📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
↩️ [fb-only] Re-run with SSH instructions
❓Need help or want to give feedback on the CI? Visit our office hours

💊 CI failures summary and remediations

As of commit 78ab3d0 (more details on the Dr. CI page):

1/1 failures introduced in this PR

🕵️‍♀️ 1 failure not recognized by patterns:

The following CI failures may be due to changes from the PR

Job	Step	Action
^{pull / linux-bionic-rocm5.0-py3.7 / test (default, 2, 2, linux.rocm.gpu)}	^{Set up job}	🔁 rerun

This comment was automatically generated by Dr. CI (expand for details).

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

Summary: making sure it works for PTQ even if squash_mask isn't applied Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the is_leaf_or_only_parametrized function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

HDCharles · 2022-03-30T03:47:58Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

vkuzo · 2022-03-30T12:58:46Z

torch/nn/quantized/modules/linear.py

 from torch.nn.utils.fusion import fuse_linear_bn_weights
 from typing import Optional

+from torch.nn.utils.parametrize import is_parametrized


nit: fix lint

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-03-30T15:24:38Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-03-30T16:46:35Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-03-31T18:00:33Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

jerryzh168 · 2022-04-01T21:29:06Z

test/ao/sparsity/test_composability.py

        self.assertTrue(isinstance(mod[5], torch.nn.quantized.Linear))
        self.assertEqual(mod(torch.randn(1, 4, 4, 4)).shape, torch.Size([1, 4, 4, 4]))
+
+    def test_convert_without_squash_mask(self):


nit: same more this, I feel these tests are a bit complicated so we should have more comments on the context, also for future tests in this file as well

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-04-04T16:18:01Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-04-04T18:05:10Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

HDCharles · 2022-04-04T19:08:47Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-04-05T17:51:27Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D35240275](https://our.internmc.facebook.com/intern/diff/D35240275) [ghstack-poisoned]

HDCharles · 2022-04-05T22:37:31Z

@HDCharles has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

Summary: Pull Request resolved: #74846 This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the is_leaf_or_only_parametrized function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Imported from OSS Reviewed By: vkuzo Differential Revision: D35240275 fbshipit-source-id: 48529f2a83edfe6d8a2d2dff8ca3d08a3fb0d553

github-actions · 2022-04-06T04:27:52Z

Hey @HDCharles.
You've committed this PR, but it does not have both a 'release notes: ...' and 'topics: ...' label. Please add one of each to the PR. The 'release notes: ...' label should represent the part of PyTorch that this PR changes (fx, autograd, distributed, etc) and the 'topics: ...' label should represent the kind of PR it is (not user facing, new feature, bug fix, perf improvement, etc). The list of valid labels can be found here for the 'release notes: ...' and here for the 'topics: ...'.
For changes that are 'topic: not user facing' there is no need for a release notes label.

Summary: This PR primarily allows the PTQ convert function to work with parametrized modules. Given that the parametrized weight is what is used by default in convert, as long as sparsifier.step() has already been called, the converted model will use the sparisified weights. There is currently no way to handle things if sparsifier.step() has not been called. Lastly, added the has_no_children_ignoring_parametrizations function because parametrized modules no longer look like leaves due to the parametrizations module attached to them Test Plan: python test/test_ao_sparsity.py TestComposability Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: ce30add Pull Request resolved: pytorch/pytorch#74846

Fixing sparsity and composability convert

7976ec6

Summary: making sure it works for PTQ even if squash_mask isn't applied Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

HDCharles requested review from albanD and jbschlosser as code owners March 28, 2022 18:33

facebook-github-bot added the cla signed label Mar 28, 2022

This was referenced Mar 28, 2022

[ao][sparsity] make sparsity and PTQ compose #74845

Closed

[ao][sparsity] Composability of fusion and sparsity #74847

Closed

[ao][sparsity] comsability for sparsity and QAT convert #74848

Closed

HDCharles added 2 commits March 28, 2022 16:08

Update on "Fixing sparsity and composability convert"

c496f35

Summary: making sure it works for PTQ even if squash_mask isn't applied Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]

HDCharles changed the title ~~Fixing sparsity and composability convert~~ [ao][sparsity] make sparsity compose with PTQ convert Mar 28, 2022

HDCharles added 3 commits March 28, 2022 16:45

HDCharles requested review from andrewor14, dzdang, jerryzh168, terrychenism, vkuzo and z-a-f March 28, 2022 23:52

HDCharles added 3 commits March 28, 2022 17:01

albanD removed their request for review March 29, 2022 18:09

vkuzo approved these changes Mar 30, 2022

View reviewed changes

jerryzh168 reviewed Apr 1, 2022

View reviewed changes

pytorchmergebot closed this in 9bb21fa Apr 6, 2022

facebook-github-bot deleted the gh/HDCharles/63/head branch April 9, 2022 14:16

WBobby mentioned this pull request Aug 17, 2022

Add ROCm5.2.3/AMDGPU support for PyTorch WBobby/pytorch#2

Closed

[ao][sparsity] make sparsity compose with PTQ convert #74846

[ao][sparsity] make sparsity compose with PTQ convert #74846

Uh oh!

Conversation

HDCharles commented Mar 28, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

facebook-github-bot commented Mar 28, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful links

💊 CI failures summary and remediations

🕵️‍♀️ 1 failure not recognized by patterns:

Uh oh!

HDCharles commented Mar 30, 2022

Uh oh!

vkuzo Mar 30, 2022

Choose a reason for hiding this comment

Uh oh!

HDCharles commented Mar 30, 2022

Uh oh!

HDCharles commented Mar 30, 2022

Uh oh!

HDCharles commented Mar 31, 2022

Uh oh!

jerryzh168 Apr 1, 2022 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HDCharles commented Apr 4, 2022

Uh oh!

HDCharles commented Apr 4, 2022

Uh oh!

HDCharles commented Apr 4, 2022

Uh oh!

HDCharles commented Apr 5, 2022

Uh oh!

HDCharles commented Apr 5, 2022

Uh oh!

github-actions bot commented Apr 6, 2022

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

HDCharles commented Mar 28, 2022 •

edited

Loading

facebook-github-bot commented Mar 28, 2022 •

edited

Loading

jerryzh168 Apr 1, 2022 •

edited

Loading