Add torch.linalg.norm
#42749
Add torch.linalg.norm
#42749
Conversation
💊 CI failures summary and remediationsAs of commit d069246 (more details on the Dr. CI page):
🕵️ 2 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
|
Hey @kurtamohler, let me know when this is done being drafted and ready for review. |
9e0eb53 to
0f0d85d
Compare
|
Will do, @mruberry. I believe I just need to update the documentation, and perhaps add a few more test cases, and then I'll let you know it's ready. |
1ec8b6c to
0c88fc8
Compare
866dcf3 to
896319a
Compare
| } | ||
| } | ||
| resize_output(result, result_.sizes()); | ||
| result.copy_(result_); |
There was a problem hiding this comment.
Note: I don't think we can do the set_ pattern used above here because it would replace result's storage. This might be OK, but our current thinking is to preserve given storages where possible.
Sure. Definitely couldn't hurt. |
* Avoid unecessary clone * Use torch.empty rather than torch.tensor in unit test
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.
|
Looks like |
|
@kurtamohler xla tests are usually failing with discontiguous tensors - do your test produce those? If that's the case, those tests just have to be disabled on the xla side (either decorate with @onlyCPUandCUDA or something like that, or ask @ailzhang to disable them on the xla side). |
|
@ngimel none of the inputs or outputs of nuclear_norm in that test are discontiguous. But it's possible that something inside |
Just skip the test by decorating it with https://github.com/pytorch/xla/blob/19fc8124d5cbefd18796b2439ddf3fded5c7a9ba/test/pytorc
A373
h_test_base.py#L51 If you really wanted to you could run the XLA tests by installing PyTorch/XLA and using their test runner, but I don't think it's worthwhile in this case. The reason this test is failing is because XLA doesn't know to skip it and it's added to TestTorchDeviceType, which spawns "child" per-device classes (TestTorchDeviceTypeCPU, TestTorchDeviceTypeCUDA, TestTorchDeviceTypeXLA) depending on what software and hardware is available on the platform. Decorating the test with |
|
Lint failures are in base, not your PR. |
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.
|
This has landed and tests look good. Nice work, @kurtamohler. It's great to have this centerpiece for torch.linalg in! |
* Add cross signature * Add det specification * Add diagonal specification * Add inv specification * Add norm specification * Add outer product specification * Add outer specification * Add trace specification * Add transpose * Update index * Fix type annotation * Update norm behavior for multi-dimensional arrays * Support all of NumPy's norms Further support for supporting all of NumPy's norms comes from pending updates to Torch (see pytorch/pytorch#42749). * Switch order * Split into separate tables * Escape markup * Escape markup * Add matrix_transpose interface Interface inspired by TensorFlow (see https://www.tensorflow.org/api_docs/python/tf/linalg/matrix_transpose) and Torch (see https://pytorch.org/docs/stable/generated/torch.transpose.html). Allows transposing a stack of matrices. * Rename parameters * Remove matrix_transpose signature
Adds
torch.linalg.normfunction that matches the behavior ofnumpy.linalg.norm.Additional changes:
frobenius_normandnuclear_normoutargument behavior fornuclear_normfrobenius_normallowed duplicates indimargument_norm_matrixCloses #24802