Closed
Conversation
Contributor
🔗 Helpful links
❌ 37 New FailuresAs of commit 44221ef (more details on the Dr. CI page): Expand to see more
🕵️ 35 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages
|
| Job | Step |
|---|---|
| Unknown | |
| Unknown |
This comment was automatically generated by Dr. CI (expand for details).
Please report bugs/suggestions to the (internal) Dr. CI Users group.
9f7bc95 to
44221ef
Compare
ezyang
reviewed
Sep 6, 2022
torch/_meta_registrations.py
Outdated
|
|
||
| # TODO: smarter way to copy? copy.deepcopy? | ||
| # can I just use list(shape)? | ||
| res = list(shape) |
ezyang
reviewed
Sep 6, 2022
torch/_meta_registrations.py
Outdated
| def reshape(self, proposed_shape): | ||
| if self.is_sparse: | ||
| # TODO: not sure what else to do here? | ||
| raise RuntimeError("reshape is not implemented for sparse tensors") |
Contributor
|
This seems fine but you will have to monkeypatch it in for it to do anything |
731fb9e to
3f5840d
Compare
ezyang
reviewed
Sep 9, 2022
| a = torch.rand(4, 4) | ||
| b = a.reshape((16, 1)) | ||
| #b = torch.reshape(a, (16, 1)) | ||
| print(b) |
Contributor
There was a problem hiding this comment.
this test is bad. There should be an OpInfo reshape test you can use
ezyang
reviewed
Sep 9, 2022
| _register_jit_decomposition_for_jvp(torch.ops.aten.log_sigmoid_forward.default) | ||
| _register_jit_decomposition_for_jvp(torch.ops.aten.native_layer_norm_backward.default) | ||
| _register_jit_decomposition_for_jvp(torch.ops.aten.native_batch_norm_backward.default) | ||
| #_register_jit_decomposition_for_jvp(torch.ops.aten.native_batch_norm_backward.default) |
ezyang
reviewed
Sep 9, 2022
| } | ||
| return from_complex.get(dtype, dtype) | ||
|
|
||
|
|
Contributor
There was a problem hiding this comment.
don't wobble lines unnecessarily; especially in a long lived branch these lead to spurious merge conflicts. Big problem!
ezyang
reviewed
Sep 9, 2022
| return self.view(self.clone(memory_format=torch.contiguous_format), shape) | ||
|
|
||
| torch.reshape = reshape | ||
| torch.Tensor.reshape = reshape |
Contributor
There was a problem hiding this comment.
This should ultimately live in torch._refs
Contributor
There was a problem hiding this comment.
Oh shoot, there's a reshape implementation in torch._refs.reshape
Contributor
|
/easycla As part of the transition to the PyTorch Foundation, this project now requires contributions be covered under the new CLA. See #85559 for additional details. This comment will trigger a new check of this PR. If you are already covered, you will simply see a new "EasyCLA" check that passes. If you are not covered, a bot will leave a new comment with a link to sign. |
|
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
ezyang
added a commit
that referenced
this pull request
Feb 22, 2023
This PR gets `reflect @ R @ reflect` working, where R has unbacked batch size. This pattern occurred in CrystalDPR. The billing of changes: * torch.broadcast_shapes avoids guarding on unbacked SymInts when testing for broadcastable dims. I extracted this to #95217 for separate review; it's repeated in this PR as it is necessary for the E2E test * I disable matrix multiply folding when there is an unbacked SymInt on any input. Folding is strictly a performance optimization and can be omitted. Also, I believe export would prefer to get matmul (rather than bmm/etc), so we should eventually actually get #91081 going * I add a direct Python transcription of the reshape composite adapted from #84584 . I cannot use the PrimTorch composite as it has problems when I register it pre-autograd. It has the same implementation as regular reshape, but at the beginning there is one more test for trivial reshapes, which is sufficient for the matmul example. * I hand-write a meta function for expand, rather than using the PrimTorch decomposition. I couldn't really figure out how to make the PrimTorch decomposition guard free, but with the hand-written meta it is clear where the divergence lies: we cannot easily choose the correct stride for the unbacked dim, as we need to know whether or not the size is one (in which case we give the predicted stride) versus non-one (in which case we MUST give zero.) In composability sync, we agreed that changes to striding behavior are fair game with unbacked SymInts, so I just unconditionally give these zero stride. Signed-off-by: Edward Z. Yang <ezyangmeta.com> [ghstack-poisoned]
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #ISSUE_NUMBER