[numpy] torch.sqrt : promote integer inputs to float
#47293
Conversation
facebook-github-bot
added
cla signed
oncall: jit
💊 CI failures summary and remediationsAs of commit 91b9a3a (more details on the Dr. CI page): ✅ None of the CI failures appear to be your fault 💚
🚧 2 ongoing upstream failures:These were probably caused by upstream breakages that are not fixed yet:
This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 18 times. |
|
Reason for disabling >>> import torch
>>> torch.tensor(4980736.).sqrt()
tensor(2231.7563)
>>> torch.tensor(4980736.).sqrt().to(torch.bfloat16)
tensor(2224., dtype=torch.bfloat16) # Seems like bfloat16 can't represent tensor(2231.7563) |
|
Reason for disabling |
ejguan
added
module: numpy
Codecov Report
@@ Coverage Diff @@
## master #47293 +/- ##
==========================================
- Coverage 81.23% 80.82% -0.42%
==========================================
Files 1836 1806 -30
Lines 197684 189822 -7862
==========================================
- Hits 160586 153416 -7170
+ Misses 37098 36406 -692 |
| @@ -419,7 +419,36 @@ def sample_inputs(self, device, dtype, requires_grad=False): | |||
| ref=np.nan_to_num, | |||
| dtypes=all_types_and(torch.half, torch.bool), | |||
| dtypesIfCPU=None, | |||
| dtypesIfCUDA=None) | |||
| dtypesIfCUDA=None), | |||
| UnaryUfuncInfo('sqrt', | |||
There was a problem hiding this comment.
It's great to see another OpInfo! With this ported this PR can also remove the TorchMathTestMeta for sqrt in test_torch.py (see torch_op_tests, I can't link the line because the file is too big to render on Github).
| dtypesIfCUDA=None), | ||
| UnaryUfuncInfo('sqrt', | ||
| ref=np.sqrt, | ||
| domain=(0, float('inf')), |
There was a problem hiding this comment.
Metanote: it's a little unfortunate that we don't have dtype-specific domains so we can't test taking the sqrt of negative complex values easily.
| SkipInfo('TestUnaryUfuncs', 'test_reference_numerics', | ||
| dtypes=[torch.bfloat16]), | ||
| # RuntimeError: sqrt does not support automatic differentiation for outputs with complex dtype. | ||
| SkipInfo('TestGradients', 'test_fn_grad', |
There was a problem hiding this comment.
Metanote: if this happens again we should add a property for whether a function supports complex autograd or not, and, if complex autograd isn't supported, skip the complex autograd tests.
|
Hey @kshitij12345, this looks great, as usual! Would you update this to remove the redundant test for sqrt in test_torch.py, too? |
|
@mruberry Have removed the redundant test. PTAL :) |
There was a problem hiding this comment.
@mruberry has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
There was a problem hiding this comment.
@mruberry has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
|
@mruberry Gentle Ping :) |
I had to re-run some internal tests that were flaky. This should land soon. |
Summary: Reference pytorch#42515 Pull Request resolved: pytorch#47293 Reviewed By: malfet Differential Revision: D24855994 Pulled By: mruberry fbshipit-source-id: 1e6752f2eeba6d638dea0bdea0c650cf722718c9
Reference #42515