Adding multiweight support to Quantized GoogLeNet #4848
Conversation
datumbox
added
enhancement
module: models.quantization
💊 CI failures summary and remediationsAs of commit 1a150d9 (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. |
| @@ -84,6 +84,7 @@ def test_old_vs_new_factory(model_fn, module_name, dev): | |||
| }, | |||
| "quantization": { | |||
| "input_shape": (1, 3, 224, 224), | |||
| "quantize": True, | |||
There was a problem hiding this comment.
Needed to actually run the tests on the quantized weights.
| progress: bool = True, | ||
| quantize: bool = False, | ||
| **kwargs: Any, | ||
| ) -> "QuantizableGoogLeNet": |
There was a problem hiding this comment.
Just moving this builder below to the class definition to use proper typing.
| progress: bool = True, | ||
| quantize: bool = False, | ||
| **kwargs: Any, | ||
| ) -> QuantizableGoogLeNet: |
There was a problem hiding this comment.
copy-pasted, no need for review.
| @@ -14,14 +14,13 @@ | |||
|
|
|||
|
|
|||
| class GoogLeNetWeights(Weights): | |||
| ImageNet1K_Community = WeightEntry( | |||
| ImageNet1K_TFV1 = WeightEntry( | |||
There was a problem hiding this comment.
This was previously incorrectly tagged as community contribution while in reality the weights were ported from TF. Proof: #678 (comment)
| else: | ||
| weights = GoogLeNetWeights.verify(weights) | ||
|
|
||
| original_aux_logits = kwargs.get("aux_logits", False) |
There was a problem hiding this comment.
We simplify similarly to the unquantized builder.
| kwargs["num_classes"] = len(weights.meta["categories"]) | ||
| if "backend" in weights.meta: | ||
| kwargs["backend"] = weights.meta["backend"] | ||
| backend = kwargs.pop("backend", "fbgemm") |
There was a problem hiding this comment.
The rest is similar to the quantized implementation for resnet.
| "backend": "fbgemm", | ||
| "quantization": "ptq", |
There was a problem hiding this comment.
Adding some additional meta-data such as interpolation, type of quantization and a reference to the unquantized enum of the the weights to all previous models.
Summary: * Reordering the builders to use proper typing. * Adding additional meta-data on existing quantized models. * Fixing meta on unquantized model. * Adding quantized googlenet builder. * undo inception move. * Adding recipe information. Reviewed By: kazhang Differential Revision: D32216660 fbshipit-source-id: 2fbda7b7d39e97f156264b86fcf42076598d564b
* Reordering the builders to use proper typing. * Adding additional meta-data on existing quantized models. * Fixing meta on unquantized model. * Adding quantized googlenet builder. * undo inception move. * Adding recipe information.
Fixes partially #4674
Verified with:
cc @bjuncek