prepare removal of deprecated functionality in torch.testing #87969

pmeier · 2022-10-28T07:14:24Z

Stack from ghstack (oldest at bottom):

Redo of #86586 with all BC breaking changes granularly placed into separate commits.

Per title. Deprecation happened on Feb 25, 2022 in c6f1bbc, which made it into the 1.12 release. Since it is now 245 days later and the next release will be 1.14, the removals later in the stack comply with the BC policy.

[ghstack-poisoned]

pytorch-bot · 2022-10-28T07:14:26Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/87969

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

❌ 3 Failures

As of commit 90869bd:

NEW FAILURES - The following jobs have failed:

linux-bionic-cuda11.6-py3.9-gcc7 / test (multigpu, 1, 1, linux.16xlarge.nvidia.gpu) (gh)
linux-focal-py3.7-clang7-asan / test (default, 4, 5, linux.2xlarge) (gh)
linux-focal-rocm5.2-py3.8-distributed / test (distributed, 1, 2, linux.rocm.gpu) (gh)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

pmeier

There were two things that needed manual handling:

torch.testing.make_non_contiguous needed to be replaced with one of the suggested options as was applicable:

pytorch/torch/testing/_deprecated.py

Lines 106 to 114 in 166b5d3

    
           @warn_deprecated( 
        
               "Depending on the use case there a different replacement options:\n\n" 
        
               "- If you are using `make_non_contiguous` in combination with a creation function to create a noncontiguous tensor " 
        
               "with random values, use `torch.testing.make_tensor(..., noncontiguous=True)` instead.\n" 
        
               "- If you are using `make_non_contiguous` with a specific tensor, you can replace this call with " 
        
               "`torch.repeat_interleave(input, 2, dim=-1)[..., ::2]`.\n" 
        
               "- If you are using `make_non_contiguous` in the PyTorch test suite, use " 
        
               "`torch.testing._internal.common_utils.noncontiguous_like` instead." 
        
           )

torch.testing.assert_allclose couldn't always 1-to-1 replaced with the new-ish (stable since Feb 2022) torch.testing.assert_close. See #61844 for a detailed analysis of the differences.

I've highlighted all places where I did more than a simple replace of deprecated functionality below.

pmeier · 2022-10-28T07:33:06Z

caffe2/python/operator_test/_utils.py

+# The function below is a faithful replica of the former `torch.testing.assert_allclose`. This is only here,
+# because it is used extensively throughout the tests in this package while needing one feature that
+# the new `torch.testing.assert_close` does not offer: comparison between numpy arrays and torch tensors. See
+# https://github.com/pytorch/pytorch/issues/61844 for the reasoning why this feature was removed.


Per comment. This only applies to the tests located in caffe2/python/operator_test/*.

pmeier · 2022-10-28T07:33:45Z

test/distributions/test_distributions.py

        dist = Poisson(rate_zero)
        dist.log_prob(torch.ones_like(rate_zero)).backward()
-        torch.testing.assert_allclose(rate_zero.grad, torch.inf)
+        self.assertEqual(rate_zero.grad, torch.inf)


Comparing scalar tensors to Python scalars is not supported by torch.testing.assert_close. I've opted to use self.assertEqual here since that still includes this type wrangling for BC.

pmeier · 2022-10-28T07:35:13Z

test/test_jit_fuser_te.py

            trace = torch.jit.trace(fn, args)
            self.assertAllFused(trace.graph_for(*args))
-            torch.testing.assert_allclose(fn(*args), trace(*args))
+            torch.testing.assert_close(fn(*args), trace(*args), equal_nan=True)


In torch.testing.assert_close, equal_nan is False by default. Are NaN's actually ok here or did this pass silently before?

pmeier · 2022-10-28T07:36:01Z

test/test_mps.py

        py_relu_cpu = py_relu.to("cpu")

-        torch.testing.assert_allclose(np_relu, py_relu_cpu)
+        self.assertEqual(np_relu, py_relu_cpu)


Comparing numpy.ndarray's to torch.Tensor's is not supported in torch.testing.assert_close. self.assertEqual still allows it.

pmeier · 2022-10-28T07:36:43Z

test/test_torch.py

+                    idx = torch.testing.make_tensor(
+                        num_src, low=0, high=num_dest, dtype=idx_dtype, device=device, noncontiguous=index_noncontig
+                    )


If we want to create a noncontiguous tensor right away, we can use torch.testing.make_tensor directly.

pmeier · 2022-10-28T07:36:56Z

test/test_torch.py

                        src = torch.randn(num_copy, *other_sizes, device=device)
                        if not src_contig:
-                            src = torch.testing.make_non_contiguous(src)
+                            src = noncontiguous_like(src)


Not using torch.testing.make_tensor here since src samples from a normal distribution and the former would sample from a uniform one. If the intent is to just sample positive as well as negative values, LMK.

pmeier · 2022-10-28T07:37:09Z

test/test_torch.py

                        idx = torch.randperm(num_dest, dtype=dtype, device=device).narrow(0, 0, num_copy)
                        if not index_contig:
-                            idx = torch.testing.make_non_contiguous(idx)
+                            idx = noncontiguous_like(idx)


Same as above but for a random permutation.

pmeier · 2022-10-28T07:37:50Z

torch/testing/_internal/common_fsdp.py

            fsdp_loss = fsdp_loss.cuda()
        fsdp_unsharded_params = get_full_params(fsdp_model)
-        torch.testing.assert_allclose(ref_loss, fsdp_loss)
+        torch.testing.assert_close(ref_loss, fsdp_loss, check_dtype=False)


Strict dtype checking is the default for torch.testing.assert_close. Are mismatching dtypes ok here?

_Redo of #86586 with all BC breaking changes granularly placed into separate commits._ --- Per title. Deprecation happened on Feb 25, 2022 in c6f1bbc, which made it into the 1.12 release. Since it is now 245 days later and the next release will be 1.14, the removals later in the stack comply with the [BC policy](https://github.com/pytorch/pytorch/wiki/PyTorch's-Python-Frontend-Backward-and-Forward-Compatibility-Policy#minimizing-the-disruption-of-bc-breaking-changes). [ghstack-poisoned]

mruberry

LGTM!

There are several test failures, but I can't imagine they're related to this PR

See #87969 or #86586 for the reasoning. Pull Request resolved: #87970 Approved by: https://github.com/mruberry

See #87969 or #86586 for the reasoning. Pull Request resolved: #87971 Approved by: https://github.com/mruberry

See #87969 or #86586 for the reasoning. Pull Request resolved: #87972 Approved by: https://github.com/mruberry

See #87969 or #86586 for the reasoning. Pull Request resolved: #87973 Approved by: https://github.com/mruberry

See #87969 or #86586 for the reasoning. Pull Request resolved: #87974 Approved by: https://github.com/mruberry

…#87969) _Redo of pytorch#86586 with all BC breaking changes granularly placed into separate commits._ --- Per title. Deprecation happened on Feb 25, 2022 in c6f1bbc, which made it into the 1.12 release. Since it is now 245 days later and the next release will be 1.14, the removals later in the stack comply with the [BC policy](https://github.com/pytorch/pytorch/wiki/PyTorch's-Python-Frontend-Backward-and-Forward-Compatibility-Policy#minimizing-the-disruption-of-bc-breaking-changes). Pull Request resolved: pytorch#87969 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87970 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87971 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87972 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87973 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87974 Approved by: https://github.com/mruberry

…#87969) _Redo of pytorch#86586 with all BC breaking changes granularly placed into separate commits._ --- Per title. Deprecation happened on Feb 25, 2022 in c6f1bbc, which made it into the 1.12 release. Since it is now 245 days later and the next release will be 1.14, the removals later in the stack comply with the [BC policy](https://github.com/pytorch/pytorch/wiki/PyTorch's-Python-Frontend-Backward-and-Forward-Compatibility-Policy#minimizing-the-disruption-of-bc-breaking-changes). Pull Request resolved: pytorch#87969 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87970 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87971 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87972 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87973 Approved by: https://github.com/mruberry

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87974 Approved by: https://github.com/mruberry

prepare removal of deprecated functionality in torch.testing

845eaa7

[ghstack-poisoned]

pytorch-bot bot added ciflow/mps Run MPS tests (subset of trunk) release notes: distributed (ddp) release notes category labels Oct 28, 2022

pytorch-bot bot added the ciflow/mps Run MPS tests (subset of trunk) label Oct 28, 2022

pmeier commented Oct 28, 2022

View reviewed changes

pmeier added 2 commits October 28, 2022 10:31

pytorchbot added the open source label Oct 28, 2022

pmeier added 2 commits October 28, 2022 15:04

pmeier marked this pull request as ready for review November 2, 2022 10:27

pmeier requested review from H-Huang, awgu, kulinseth, kwen2501, mrshenli, pritamdamania87, rohan-varma and zhaojuanmao as code owners November 2, 2022 10:27

pmeier requested review from jerryzh168 and z-a-f as code owners November 2, 2022 10:27

pmeier mentioned this pull request Nov 2, 2022

remove deprecated torch.testing functionality #86586

Closed

mruberry approved these changes Nov 2, 2022

View reviewed changes

pytorchmergebot closed this in bc73aff Nov 2, 2022

pytorchmergebot pushed a commit that referenced this pull request Nov 2, 2022

remove deprecated rand and randn from torch.testing (#87970)

554cdc9

See #87969 or #86586 for the reasoning. Pull Request resolved: #87970 Approved by: https://github.com/mruberry

pytorchmergebot pushed a commit that referenced this pull request Nov 2, 2022

remove deprecated device getter from torch.testing (#87971)

a360be5

See #87969 or #86586 for the reasoning. Pull Request resolved: #87971 Approved by: https://github.com/mruberry

pytorchmergebot pushed a commit that referenced this pull request Nov 2, 2022

remove deprecated dtype getters from torch.testing (#87972)

8893c6c

See #87969 or #86586 for the reasoning. Pull Request resolved: #87972 Approved by: https://github.com/mruberry

pytorchmergebot pushed a commit that referenced this pull request Nov 2, 2022

remove make_non_contiguous from torch.testing (#87973)

b9c6178

See #87969 or #86586 for the reasoning. Pull Request resolved: #87973 Approved by: https://github.com/mruberry

pytorchmergebot pushed a commit that referenced this pull request Nov 2, 2022

remove assert_allclose from torch.testing (#87974)

5669e10

See #87969 or #86586 for the reasoning. Pull Request resolved: #87974 Approved by: https://github.com/mruberry

kulinseth pushed a commit to kulinseth/pytorch that referenced this pull request Nov 5, 2022

remove assert_allclose from torch.testing (pytorch#87974)

357794c

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87974 Approved by: https://github.com/mruberry

kulinseth pushed a commit to kulinseth/pytorch that referenced this pull request Dec 10, 2022

remove assert_allclose from torch.testing (pytorch#87974)

522fc1c

See pytorch#87969 or pytorch#86586 for the reasoning. Pull Request resolved: pytorch#87974 Approved by: https://github.com/mruberry

kit1980 added the Merged label Mar 24, 2023

facebook-github-bot deleted the gh/pmeier/35/head branch June 8, 2023 18:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

prepare removal of deprecated functionality in torch.testing #87969

prepare removal of deprecated functionality in torch.testing #87969

Uh oh!

pmeier commented Oct 28, 2022 •

edited

Loading

Uh oh!

pytorch-bot bot commented Oct 28, 2022 •

edited

Loading

Uh oh!

pmeier left a comment

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

pmeier Oct 28, 2022

Uh oh!

mruberry left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

	@warn_deprecated(
	"Depending on the use case there a different replacement options:\n\n"
	"- If you are using `make_non_contiguous` in combination with a creation function to create a noncontiguous tensor "
	"with random values, use `torch.testing.make_tensor(..., noncontiguous=True)` instead.\n"
	"- If you are using `make_non_contiguous` with a specific tensor, you can replace this call with "
	"`torch.repeat_interleave(input, 2, dim=-1)[..., ::2]`.\n"
	"- If you are using `make_non_contiguous` in the PyTorch test suite, use "
	"`torch.testing._internal.common_utils.noncontiguous_like` instead."
	)

prepare removal of deprecated functionality in torch.testing #87969

prepare removal of deprecated functionality in torch.testing #87969

Uh oh!

Conversation

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

Uh oh!

Uh oh!

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

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/87969

❌ 3 Failures

Uh oh!

pmeier left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mruberry left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

pmeier commented Oct 28, 2022 •

edited

Loading

pytorch-bot bot commented Oct 28, 2022 •

edited

Loading