trying to make pow work for tensor raised to the power of a scalar #46185
Conversation
|
unit test? |
aten/src/ATen/native/Pow.cpp
Outdated
| @@ -62,6 +62,9 @@ Tensor& pow_(Tensor& base, Scalar alpha) { | |||
| } | |||
|
|
|||
| Tensor pow(const Tensor& base, const Tensor& exp) { | |||
| if (exp.dim() == 0) { | |||
There was a problem hiding this comment.
can you add a test in test/test_torch.py?
💊 CI failures summary and remediationsAs of commit f3765e9 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures 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 32 times. |
There was a problem hiding this comment.
@janeyx99 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
I'm guessing there's already a test for this?
I see you added a test for this 👍
There's no test for this, right? Should one be added? |
There was a problem hiding this comment.
@janeyx99 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
|
@samestep: |
There was a problem hiding this comment.
@janeyx99 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Fixes #46037
I'm not sure this is the most performant solution, but this works:
torch.pow(cuda_tensor, 5) should work and worked before.
torch.pow(cuda_tensor, torch.tensor(5)), should work and works now!
torch.pow(cuda_tensor, torch.tensor((5,))), should NOT work and complain the tensors are on different devices and indeed continues to complain.