Allow no quantization during QATConfig convert#2694
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2694
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ✅ No FailuresAs of commit 53af27a with merge base 418593c ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
b1f35a1 to
9fbcffe
Compare
**Summary:** This commit adds back the functionality to swap
`FakeQuantized*` modules back to the corresponding `torch.nn.*`
without performing post-training quantization:
```
QATConfig(base_config=None, step="convert")
```
This has the exact same functionality as this deprecated config:
```
FromIntXQuantizationAwareTrainingConfig()
```
This functionality is added back since it may be useful to users
who wish to save QAT trained checkpoints from models containing
only `torch.nn.*` modules (not `FakeQuanitzed*` modules), e.g.
when training and inference need to happen on different machines:
```
quantize_(model, QATConfig(base_ptq_config, step="prepare"))
train(model)
quantize_(model, QATConfig(step="convert"))
torch.save(model.state_dict(), "my_checkpoint.pt")
\# On a different machine
model.load_state_dict(torch.load("my_checkpoint.pt"))
quantize_(model, base_ptq_config)
```
**Test Plan:**
```
python test/quantization/test_qat.py -k qat_config_init
python test/quantization/test_qat.py -k qat_api_convert_no_quantization
```
9fbcffe to
53af27a
Compare
Contributor
|
I see, this is the step="load" in AWQConfig, would it be easier to just add that |
Contributor
Author
|
I don't think we need a different step? This is just separating the convert step into two phases: (1) swap back to nn.Linear, and (2) quantize the model, and these can happen on different machines |
jerryzh168
approved these changes
Aug 12, 2025
lisjin
reviewed
Aug 13, 2025
Comment on lines
+161
to
+162
| and self.base_config is None | ||
| and self.weight_config is None |
Contributor
There was a problem hiding this comment.
@andrewor14 Would it be possible to allow base_config=None and weight_config=None? This would support PARQ's activation-only quantization use case in #2743.
Contributor
Author
There was a problem hiding this comment.
yeah let's do that separately in your PR
liangel-02
pushed a commit
that referenced
this pull request
Aug 25, 2025
**Summary:** This commit adds back the functionality to swap
`FakeQuantized*` modules back to the corresponding `torch.nn.*`
without performing post-training quantization:
```
QATConfig(base_config=None, step="convert")
```
This has the exact same functionality as this deprecated config:
```
FromIntXQuantizationAwareTrainingConfig()
```
This functionality is added back since it may be useful to users
who wish to save QAT trained checkpoints from models containing
only `torch.nn.*` modules (not `FakeQuanitzed*` modules), e.g.
when training and inference need to happen on different machines:
```
quantize_(model, QATConfig(base_ptq_config, step="prepare"))
train(model)
quantize_(model, QATConfig(step="convert"))
torch.save(model.state_dict(), "my_checkpoint.pt")
\# On a different machine
model.load_state_dict(torch.load("my_checkpoint.pt"))
quantize_(model, base_ptq_config)
```
**Test Plan:**
```
python test/quantization/test_qat.py -k qat_config_init
python test/quantization/test_qat.py -k qat_api_convert_no_quantization
```
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Summary: This commit adds back the functionality to swap
FakeQuantized*modules back to the correspondingtorch.nn.*without performing post-training quantization:This has the exact same functionality as this deprecated config:
This functionality is added back since it may be useful to users who wish to save QAT trained checkpoints from models containing only
torch.nn.*modules (notFakeQuanitzed*modules), e.g. when training and inference need to happen on different machines:Test Plan: