[functorch] Fix torch.cat batching rule #86932
Conversation
The bug was discovered in #86842. torch.cat has an edge case where it ignores all tensors of shape [0]. So if any of the BatchedTensors have logical shape [0] but physical shape [B, 0], then we coerce them to shape [0] by slicing them. Why don't we just ignore those Tensors? We need to propagate requires_grad-ness somehow (e.g. if the BatchedTensor wraps a Tensor of shape [B, 0] that requires grad, then the output must require grad). Test Plan: - new tests [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/86932
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 Failures, 1 PendingAs of commit 167c1a6: The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
The bug was discovered in #86842. torch.cat has an edge case where it ignores all tensors of shape [0]. So if any of the BatchedTensors have logical shape [0] but physical shape [B, 0], then we coerce them to shape [0] by slicing them. Why don't we just ignore those Tensors? We need to propagate requires_grad-ness somehow (e.g. if the BatchedTensor wraps a Tensor of shape [B, 0] that requires grad, then the output must require grad). Test Plan: - new tests ghstack-source-id: 3cd4196977aff46c079b63b811b5056290f772f8 Pull Request resolved: #86932
Yes. The problem is the vmap codegen doesn't handle operators that accept TensorList yet |
|
@pytorchbot merge -f "failures look unrelated (they occur in the non-functorch shards) (also there are no logs)" |
1 similar comment
|
@pytorchbot merge -f "failures look unrelated (they occur in the non-functorch shards) (also there are no logs)" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
Hey @zou3519. |
The bug was discovered in pytorch#86842. torch.cat has an edge case where it ignores all tensors of shape [0]. So if any of the BatchedTensors have logical shape [0] but physical shape [B, 0], then we coerce them to shape [0] by slicing them. Why don't we just ignore those Tensors? We need to propagate requires_grad-ness somehow (e.g. if the BatchedTensor wraps a Tensor of shape [B, 0] that requires grad, then the output must require grad). Test Plan: - new tests Pull Request resolved: pytorch#86932 Approved by: https://github.com/Chillee
Stack from ghstack:
The bug was discovered in #86842.
torch.cat has an edge case where it ignores all tensors of shape [0]. So
if any of the BatchedTensors have logical shape [0] but physical shape
[B, 0], then we coerce them to shape [0] by slicing them.
Why don't we just ignore those Tensors? We need to propagate
requires_grad-ness somehow (e.g. if the BatchedTensor wraps a Tensor of
shape [B, 0] that requires grad, then the output must require grad).
Test Plan: