Use FusedMovingAvgObsFakeQuantize instead of FakeQuantize for faster QAT #14740
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/pytorch/executorch/14740
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit d95e62f with merge base c997fe4 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
navsud
added a commit
to navsud/executorch
that referenced
this pull request
Oct 2, 2025
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Differential Revision: D83583655
navsud
force-pushed
the
export-D83583655
branch
from
698f6bf to
aa9860c
Compare
navsud
force-pushed
the
export-D83583655
branch
from
aa9860c to
5e2a9ef
Compare
navsud
added a commit
to navsud/executorch
that referenced
this pull request
Oct 2, 2025
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Differential Revision: D83583655
navsud
force-pushed
the
export-D83583655
branch
from
5e2a9ef to
c1f908c
Compare
navsud
added a commit
to navsud/executorch
that referenced
this pull request
Oct 2, 2025
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Differential Revision: D83583655
billmguo
approved these changes
Oct 3, 2025
cccclai
reviewed
Oct 3, 2025
| ) -> QuantizationConfig: | ||
| extra_args: Dict[str, Any] = {"eps": 2**-20} | ||
| act_fake_quant_ctr = FakeQuantize.with_args( | ||
| act_fake_quant_ctr = FusedMovingAvgObsFakeQuantize.with_args( |
There was a problem hiding this comment.
What's the difference between FakeQuantize and FusedMovingAvgObsFakeQuantize
There was a problem hiding this comment.
FusedMovingAvgObsFakeQuantize - as the name suggests, has a combined op for FakeQuantize and MovingAvgObserver which makes it faster than two separate ops: FakeQuantize and MovingAvgObserver.
navsud
added a commit
to navsud/executorch
that referenced
this pull request
Oct 3, 2025
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Reviewed By: billmguo Differential Revision: D83583655
navsud
force-pushed
the
export-D83583655
branch
from
c1f908c to
4cbcb42
Compare
navsud
added a commit
to navsud/executorch
that referenced
this pull request
Oct 3, 2025
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Reviewed By: billmguo Differential Revision: D83583655
navsud
force-pushed
the
export-D83583655
branch
from
4cbcb42 to
f7dc1f0
Compare
…QAT (pytorch#14740) Summary: FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly. Tested on llama model on HTP and got ~4x QAT speedup. Reviewed By: billmguo Differential Revision: D83583655
navsud
force-pushed
the
export-D83583655
branch
from
f7dc1f0 to
d95e62f
Compare
billmguo
approved these changes
Oct 3, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary:
FusedMovingAvgObsFakeQuantize speeds up by fusing FakeQuantize and MovingAverageMinMaxObserver into one CUDA op. Using it should give good speedups. This change updates the QAT qconfigs to accordingly.
Tested on llama model on HTP and got ~4x QAT speedup.
Differential Revision: D83583655