-
Notifications
You must be signed in to change notification settings - Fork 21.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Support dtype
kwarg in _foreach_norm
#125665
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/125665
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 1e300d6 with merge base 5fb11cd (): NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This comment was marked as outdated.
This comment was marked as outdated.
9d0bb68
to
ae98aff
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Took a look at the changes, they seem reasonable but the meta test failures are real.
Typically I think we've skipped when the original OpInfos are broken, but I have not dug into the problem here so it may be a different concern.
yes, they are real failures but not that opinfos are broken. The cause is that |
Could you elaborate what you mean by this? Or will you be pushing a fix for this? |
pushed a fix🙂 |
for num_tensors, ord, out_dtype in product( | ||
num_input_tensors, | ||
(0, 1, 2, -1, -2, float('inf'), float('-inf')), | ||
[None] + get_all_dtypes(), | ||
(None,) + (torch.complex128,) if dtype in complex_types() else (torch.float64,), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
what coverage does this remove?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
integers, complex64, float32, float16, and bfloat16.
integers wouldn't be supported anyway, i.e., test_foreach's test_parity checks the error messages are identical.
some lower prec's could cause no direct cast or something
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'd rather still test more + just have expected skips vs silently testing less
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I found it not really straightforward to give skips/xfail to some of the sample inputs generated which I'm speculating test_meta needs (thought the failures would be that broken opinfo ilk).
Would you happen to have some reference?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm sure my previous response is misleading. DecorateInfo to cover all the dtypes here would void the valuee of tests. The dtypes with if-else here doesn’t eliminate any of the existing test cases as it’s aimed to test the new kwarg. Inappropriate dtype to the kwarg just leads to an error, and unfortunately for meta test, all dtypes (meaning parameterization of inputs) end up being decorated as xfail if the dtype kwarg parameterization covers all the possible ones
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah this parametrizes out_dtype, which is new.
dypte
kwarg in _foreach_norm
dtype
kwarg in _foreach_norm
@pytorchbot rebase |
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Signed-off-by: Masaki Kozuki <mkozuki@nvidia.com>
Successfully rebased |
546421e
to
1e300d6
Compare
for num_tensors, ord, out_dtype in product( | ||
num_input_tensors, | ||
(0, 1, 2, -1, -2, float('inf'), float('-inf')), | ||
[None] + get_all_dtypes(), | ||
(None,) + (torch.complex128,) if dtype in complex_types() else (torch.float64,), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah this parametrizes out_dtype, which is new.
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 1 jobs have failed, first few of them are: trunk / linux-focal-rocm6.1-py3.8 / test (distributed, 1, 1, linux.rocm.gpu) Details for Dev Infra teamRaised by workflow job |
@pytorchbot merge -i ROCm dist failures seem not related. |
Merge startedYour change will be merged while ignoring the following 1 checks: trunk / linux-focal-rocm6.1-py3.8 / test (distributed, 1, 1, linux.rocm.gpu) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Fixes pytorch#125040 Pull Request resolved: pytorch#125665 Approved by: https://github.com/janeyx99
is this still being checked...? The checks look okay |
@pytorchbot merge -i |
Merge startedYour change will be merged while ignoring the following 1 checks: trunk / linux-focal-rocm6.1-py3.8 / test (distributed, 1, 1, linux.rocm.gpu) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: Command Fixes #125040 Pull Request resolved: #125665
Details for Dev Infra teamRaised by workflow job |
Ah this was merged but just not closed on Wednesday around 4pm. Closing as this commit has been merged. |
Fixes #125040
cc @awgu