Passing the parameters of pretrained googlenet to an optimizer

Bill_Collins · May 3, 2022, 2:03am

I downloaded the pretrained quantized googlenet model to finetune on some data.

model = models.quantization.googlenet(pretrained=True, quantize=True)

However, list(model.parameters()) yields [].

How can I access the parameters to pass to an optimizer or set requires_grad?

There is no clear structure in the model containing the parameters, like ModuleList or something, which is mentioned in other posts about this type of topic. Here is a fragment of the model:

QuantizableGoogLeNet(
(conv1): QuantizableBasicConv2d(
(conv): QuantizedConvReLU2d(3, 64, kernel_size=(7, 7), stride=(2, 2), scale=0.08655554801225662, zero_point=0, padding=(3, 3))
(bn): Identity()
(relu): Identity()
)

Thanks

eqy · May 4, 2022, 6:32pm

From the docs, it seems that quantized models aren’t intended to be user with an optimizer (I’m assuming for training):

github.com

pytorch/vision/blob/62740807c18e68bb0acd85895dca527f9a655bd5/torchvision/models/quantization/googlenet.py#L147

      
        
            def googlenet(
                *,
                weights: Optional[Union[GoogLeNet_QuantizedWeights, GoogLeNet_Weights]] = None,
                progress: bool = True,
                quantize: bool = False,
                **kwargs: Any,
            ) -> QuantizableGoogLeNet:
                r"""GoogLeNet (Inception v1) model architecture from
                `"Going Deeper with Convolutions" <http://arxiv.org/abs/1409.4842>`_.
            
            
    Note that quantize = True returns a quantized model with 8 bit
                weights. Quantized models only support inference and run on CPUs.
                GPU inference is not yet supported
            
            
    Args:
                    weights (GoogLeNet_QuantizedWeights or GoogLeNet_Weights, optional): The pretrained
                        weights for the model
                    progress (bool): If True, displays a progress bar of the download to stderr
                    quantize (bool): If True, return a quantized version of the model
                """
                weights = (GoogLeNet_QuantizedWeights if quantize else GoogLeNet_Weights).verify(weights)