Foreach Test Refactor: Pointwise, Min/Max-imum #61327
Closed
+173
−233
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
- rewrite pointwise unittests using `ops` decorator - rewrite minimum&maximum unittests using `ops` decorator - enable minimum/maximum fastpath for BFloat16 - remove _test_data method
💊 CI failures summary and remediationsAs of commit 48b2cd4 (more details on the Dr. CI page and at hud.pytorch.org/pr/61327):
🕵️ 4 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
28 tasks
ezyang
added
the
triaged
ngimel
reviewed
Jul 19, 2021
test/test_foreach.py
Outdated
| opinfo.sample_inputs(device, dtype, N, noncontiguous=not is_fastpath), | ||
| opinfo.sample_inputs(device, dtype, N, noncontiguous=not is_fastpath), | ||
| ] | ||
| self._pointwise_test(dtype, op, ref, inputs, is_fastpath, is_inplace=False, values=None) |
There was a problem hiding this comment.
are there any tests where values is not None?
There was a problem hiding this comment.
oh, I forgot it at all. I'm adding tests with values.
test/test_foreach.py
Outdated
| def test_minmax_inf_nan(self, device, dtype, op): | ||
| inputs = ( | ||
| [ | ||
| torch.tensor([inf], device=device, dtype=dtype), |
There was a problem hiding this comment.
can you please use float('inf') etc here, and remove torch._six imports? Also, how is this test different from the previous one?
There was a problem hiding this comment.
can you please use
float('inf')etc here, and removetorch._siximports?
Does merge of test_minmax_inf_nan and test_minmax_float_inf_nan sound good to you?
Also, how is this test different from the previous one?
The only difference is the use of @ops decorator.
There was a problem hiding this comment.
I must be missing something, both tests have @ops(foreach_minmax_op_db), but nm, combining these 2 tests is good.
test/test_foreach.py
Outdated
| try: | ||
| actual = foreach_op(tensors1, tensors2, tensors3) | ||
| except RuntimeError as e: | ||
| with self.assertRaisesRegex(type(e), re.escape(str(e))): |
There was a problem hiding this comment.
when do errors happen here? It seems like inputs corresponding to the same op are always on the same device, e.g. the below translates into
[native_op(*_cuda_tensors), native_op(*_cpu_tensors)]
which should work. Also, no harm in writing it like this, it's a bit cleaner than zip of previously zipped tensors.
There was a problem hiding this comment.
when do errors happen here?
cpu kernels don't support bfloat16 and half, but @ops(foreach_pointwiese_op_db) creates the test cases of these two dtypes. Do you think using @dtypes decorator and limiting the used dtypes to, e.g., all_fp_dtypes?
It seems like inputs corresponding to the same op are always on the same device,...
that sounds way better, thank you.
There was a problem hiding this comment.
Yeah, if the purpose of this test is to test behavior on the different devices, it's better to do it for a single datatype (e.g. float). But you probably want to actually test that inputs on the different devices through an error, and now you are not doing that.
ngimel
approved these changes
Jul 21, 2021
| complex(1.0 - random.random(), 1.0 - random.random()), | ||
| ) | ||
| for N, scalar in itertools.product(N_values, scalars): | ||
| for N, scalar in itertools.product(N_values, Scalars): |
|
@ngimel has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
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.
opsdecoratoropsdecorator#58833
cc: @ptrblck @ngimel