Adding multiweight support to Quantized InceptionV3 #4850
Conversation
datumbox
added
enhancement
module: models.quantization
💊 CI failures summary and remediationsAs of commit bc51d53 (more details on the Dr. CI page):
1 failure not recognized by patterns:
1 job timed out:
This comment was automatically generated by Dr. CI (expand for details).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
datumbox
force-pushed
the
prototype/quantinception_multiweight
branch
from
beefef4 to
aa1adfb
Compare
| progress: bool = True, | ||
| quantize: bool = False, | ||
| **kwargs: Any, | ||
| ) -> "QuantizableInception3": |
There was a problem hiding this comment.
Moving the builder on the bottom of the page to use proper typing.
| progress: bool = True, | ||
| quantize: bool = False, | ||
| **kwargs: Any, | ||
| ) -> QuantizableInception3: |
There was a problem hiding this comment.
No review, just copy-paste.
| @@ -28,11 +28,11 @@ class Inception3Weights(Weights): | |||
| ) | |||
|
|
|||
|
|
|||
| def inception_v3(weights: Optional[Inception3Weights] = None, progress: bool = True, **kwargs: Any) -> Inception3: | |||
| def inception_v3(weights: Optional[InceptionV3Weights] = None, progress: bool = True, **kwargs: Any) -> Inception3: | |||
There was a problem hiding this comment.
The naming is inconsistent. The model class is called Inception3, while the mode builder inception_v3. It's unclear how we should name the weights. Following the convention from ResNets this should have been Inception_V3 which is ugly. I propose not to spend more time on this now and tackle it at #4652. I've already added a reference for this issue.
There was a problem hiding this comment.
Agree! Lets tackle it in the cleanup
| quantize_model(model, backend) | ||
|
|
||
| if weights is not None: | ||
| if quantize and not original_aux_logits: |
There was a problem hiding this comment.
Special handling needed here. See original implementation.
| @@ -28,11 +28,11 @@ class Inception3Weights(Weights): | |||
| ) | |||
|
|
|||
|
|
|||
| def inception_v3(weights: Optional[Inception3Weights] = None, progress: bool = True, **kwargs: Any) -> Inception3: | |||
| def inception_v3(weights: Optional[InceptionV3Weights] = None, progress: bool = True, **kwargs: Any) -> Inception3: | |||
There was a problem hiding this comment.
Agree! Lets tackle it in the cleanup
Summary: * Moving builder to the bottom to use proper typing. * Renaming weights. * Adding quantizated inception builder. * Correct meta info. * Fix linter. * Removing init_weights to avoid exposing it on the class. Reviewed By: kazhang Differential Revision: D32216665 fbshipit-source-id: 11dcacf95db067ab08a93408600f2b01e1b7367b
* Moving builder to the bottom to use proper typing. * Renaming weights. * Adding quantizated inception builder. * Correct meta info. * Fix linter. * Removing init_weights to avoid exposing it on the class.
Fixes partially #4674
Verified with:
cc @bjuncek