bf16 support for fake_quantize_learnable_per_channel_affine #165098
Closed
Conversation
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/165098
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 9d89081 with merge base 34ac9b6 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
liangel-02
requested review from
digantdesai,
jerryzh168,
jianyuh,
kimishpatel and
salilsdesai
as code owners
|
@liangel-02 has imported this pull request. If you are a Meta employee, you can view this in D84286904. |
jerryzh168
reviewed
Oct 9, 2025
jerryzh168
reviewed
Oct 9, 2025
jerryzh168
reviewed
Oct 9, 2025
jerryzh168
reviewed
Oct 9, 2025
liangel-02
force-pushed
the
bf16_support_per_channel
branch
from
98bdb26 to
366d198
Compare
jerryzh168
approved these changes
Oct 9, 2025
andrewor14
approved these changes
Oct 9, 2025
liangel-02
force-pushed
the
bf16_support_per_channel
branch
from
366d198 to
9d89081
Compare
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. We increase the tolerance to 1e-2 for bf16 inputs because of a difference in casting calculations between python's `x.to(torch.bfloat16)` and cpp's `x.to(at::kBFloat16)` (after comparing intermediate tensors, we found that the numerics diverge after the final casting). We don't explicitly cast in the CPP op but rather let autograd/optimizer handle it. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 13, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. We increase the tolerance to 1e-2 for bf16 inputs because of a difference in casting calculations between python's `x.to(torch.bfloat16)` and cpp's `x.to(at::kBFloat16)` (after comparing intermediate tensors, we found that the numerics diverge after the final casting). We don't explicitly cast in the CPP op but rather let autograd/optimizer handle it. [ghstack-poisoned]
pytorchmergebot
pushed a commit
that referenced
this pull request
Oct 14, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. We increase the tolerance to 1e-2 for bf16 inputs because of a difference in casting calculations between python's `x.to(torch.bfloat16)` and cpp's `x.to(at::kBFloat16)` (after comparing intermediate tensors, we found that the numerics diverge after the final casting). We don't explicitly cast in the CPP op but rather let autograd/optimizer handle it. Pull Request resolved: #165325 Approved by: https://github.com/andrewor14
liangel-02
added a commit
that referenced
this pull request
Oct 14, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 14, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
liangel-02
added a commit
that referenced
this pull request
Oct 14, 2025
Follow up to #165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. [ghstack-poisoned]
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
…165098) Adding bf16 support for `torch._fake_quantize_learnable_per_channel_affine()` op by relaxing the type check on scale TODO: need to add bf16 support to `per_tensor_affine_` as `torch._fake_quantize_learnable_per_tensor_affine_backward` gets called in the backward pass **Test** Modified unit test in `test_workflow_ops.py` Pull Request resolved: pytorch#165098 Approved by: https://github.com/jerryzh168, https://github.com/andrewor14
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Oct 21, 2025
Follow up to pytorch#165098 - adding bf16 support for the backward pass. To avoid BC breaking changes/losing precision, we upcast the parameters to fp32 after the op gets called, and downcast the gradients to bf16 before returning. For testing, we upcast to fp32 before calling the reference function. We increase the tolerance to 1e-2 for bf16 inputs because of a difference in casting calculations between python's `x.to(torch.bfloat16)` and cpp's `x.to(at::kBFloat16)` (after comparing intermediate tensors, we found that the numerics diverge after the final casting). We don't explicitly cast in the CPP op but rather let autograd/optimizer handle it. Pull Request resolved: pytorch#165325 Approved by: https://github.com/andrewor14
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.
Adding bf16 support for
torch._fake_quantize_learnable_per_channel_affine()op by relaxing the type check on scaleTODO: need to add bf16 support to
per_tensor_affine_astorch._fake_quantize_learnable_per_tensor_affine_backwardgets called in the backward passTest
Modified unit test in
test_workflow_ops.py