[qat] Add FakeQuantize Module #21767
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
Differential Revision: D15728503 Differential Version: 84919348
Differential Revision: D15728503 Differential Version: 84919576
Differential Revision: D15728503 Differential Version: 85103016
pytorchbot
added
module: cuda
Differential Revision: D15375695 Differential Version: 85144759
Differential Revision: D15483071 Differential Version: 85154440
Differential Revision: D15375695 Differential Version: 85155178
Differential Revision: D15554224 Differential Version: 85162142
Differential Revision: D15483071 Differential Version: 85187565
Differential Revision: D15483071 Differential Version: 85213062
Differential Revision: D15554224 Differential Version: 85213957
Differential Revision: D15375695 Differential Version: 85219380
Differential Revision: D15554224 Differential Version: 85217753
Differential Revision: D15554183 Differential Version: 85267282
Differential Revision: D15554183 Differential Version: 85267665
Differential Revision: D15483071 Differential Version: 85322492
Differential Revision: D15554224 Differential Version: 85325488
Differential Revision: D15554224 Differential Version: 85330125
Differential Revision: D15483071 Differential Version: 85331722
raghuramank100
suggested changes
Jun 21, 2019
Differential Revision: D15375695 Differential Version: 85439410
Differential Revision: D15483071 Differential Version: 85439430
Differential Revision: D15554224 Differential Version: 85439441
Differential Revision: D15554183 Differential Version: 85439445
Differential Revision: D16185556 Differential Version: 86296584
Differential Revision: D16185556 Differential Version: 86309733
Differential Revision: D15728503 Differential Version: 86310060
dzhulgakov
requested changes
Jul 11, 2019
| Y.copy_(X); // We might want to just return the input here. | ||
| return Y; | ||
| } | ||
| TORCH_CHECK(self.scalar_type() == ScalarType::Float); |
Collaborator
There was a problem hiding this comment.
in a sense of generality - anything preventing you supporting other float types?
Contributor
Author
There was a problem hiding this comment.
we only implemented quantize_linear and dequantize for float right now, but we could also do the double to float conversion when we convert to quantized model. Not sure how valuable it is to support more types though, do we train in double in practice?
| def enalbe(self): | ||
| self.enable_fq = True | ||
|
|
||
| def calculate_qparams(self): |
Collaborator
There was a problem hiding this comment.
nit: why do you need this method? is it part of public api?
Contributor
Author
There was a problem hiding this comment.
yes, we are using the same signature as Observer
Contributor
Author
There was a problem hiding this comment.
we need this because of the conversion, in from_float, we'll call calculate_params on observer(fake_quant). see later prs for more context.
Differential Revision: D15728503 Differential Version: 86365862
Differential Revision: D15728503 Differential Version: 86368540
dzhulgakov
requested changes
Jul 12, 2019
Collaborator
dzhulgakov
left a comment
dzhulgakov
left a comment
There was a problem hiding this comment.
build failures look legit
Differential Revision: D15728503 Differential Version: 86404688
Differential Revision: D15728503 Differential Version: 86408181
Differential Revision: D15728503 Differential Version: 86414971
Differential Revision: D15728503 Differential Version: 86421747
Differential Revision: D15728503 Differential Version: 86438127
Differential Revision: D15728503 Differential Version: 86495867
Differential Revision: D15728503 Differential Version: 86499899
dzhulgakov
approved these changes
Jul 15, 2019
Collaborator
dzhulgakov
left a comment
dzhulgakov
left a comment
There was a problem hiding this comment.
Looks good to me module one tiny issue
torch/quantization/fake_quantize.py
Outdated
| self.enabled = True | ||
|
|
||
| def disable(self): | ||
| self.enable(False) |
Collaborator
There was a problem hiding this comment.
does it work? I think you're missing the arg for enable
Differential Revision: D15728503 Differential Version: 86507651
Contributor
|
This pull request has been merged in f7de9be. |
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Jul 15, 2019
Summary: Pull Request resolved: pytorch/pytorch#21767 Adding FakeQuantize Module for quantization aware training Reviewed By: dzhulgakov Differential Revision: D15728503 fbshipit-source-id: 2a9a6a362812ede3deac42b93dddca35987bd8e6
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.
Stack:
:black_circle: #21767 [qat] Add FakeQuantize Module 💚
Adding FakeQuantize Module
for quantization aware training
Differential Revision: D15728503