Fix affine quantized tensor to device calls #726
Conversation
Summary: Fixes: pytorch#698 Also added `TorchAOBaseTensor` addressing part of pytorch#710 Test Plan: python test/dtypes/test_affine_quantized.py Reviewers: Subscribers: Tasks: Tags:
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/726
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit f2e4a39 with merge base ac8ce4c ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
torchao/utils.py
Outdated
| memory_format = ( | ||
| memory_format if memory_format is not None else torch.preserve_format | ||
| ) |
There was a problem hiding this comment.
I have wondered about this for some time. Does it make sense to have memory_format applied to inner tensor? From what I understand, it's mainly for channels_last in convolution (https://pytorch.org/docs/stable/tensor_attributes.html#torch-memory-format). Maybe we can skip memory_format argument also?
There was a problem hiding this comment.
yeah this probably does not apply for most of the tensors, it won't impact things much if we remove this as well I think.
I guess if in the future some special cases need these args we can just copy paste this function and add the ommitted args
There was a problem hiding this comment.
if the current tests pass, I think should be good for now!
Summary:
Fixes: #698
Also added
TorchAOBaseTensoraddressing part of #710Test Plan:
python test/dtypes/test_affine_quantized.py
Reviewers:
Subscribers:
Tasks:
Tags: