Re-enable test case to ensure EmbeddingBag gradient computation works #2289
Conversation
|
Thanks for making the change. we also disabled couple embeddingBag related test and I think your change might fix those tests. Could you give them a try? Searching for |
gmagogsfm
force-pushed
the
test_embedding
branch
2 times, most recently
from
e1cd942 to
651459a
Compare
gmagogsfm
changed the title
gmagogsfm
force-pushed
the
test_embedding
branch
2 times, most recently
from
59d1dab to
76af62c
Compare
| 'TestNNDeviceTypeXLA': { | ||
| 'test_embedding_backward', # sparse | ||
| 'test_embedding_dense_grad', # slow | ||
| 'test_EmbeddingBag_per_sample_weights_and_new_offsets', # FIXME! UndefinedTensorImpl::_singleton |
There was a problem hiding this comment.
Btw there're 3 more embeddingbag tests below, could you confirm they're still failing (are they different issues?) Thanks!
There was a problem hiding this comment.
Yes, they are still failing. Two of them are failing because they test sparse tensor behavior. The other fails due to a different runtime error.
There was a problem hiding this comment.
Would you mind updating them with the latest error message? (The singleton issue should be fixed by your PR) Thanks
There was a problem hiding this comment.
Done. And had a nice surprise of finding an extra test that's passing but missed before
Thanks, ended up enabling one of those tests instead of adding a new one. |
There was a problem hiding this comment.
Thanks @gmagogsfm ! Only had one minor comment about keeping our skipped list up to date.
| 'TestNNDeviceTypeXLA': { | ||
| 'test_embedding_backward', # sparse | ||
| 'test_embedding_dense_grad', # slow | ||
| 'test_EmbeddingBag_per_sample_weights_and_new_offsets', # FIXME! UndefinedTensorImpl::_singleton |
There was a problem hiding this comment.
Would you mind updating them with the latest error message? (The singleton issue should be fixed by your PR) Thanks
|
Pin removed. |
4 participants
This PR tests pytorch/pytorch#40557 in pytorch/pytorch.
Closes #2215