Loading quantized models with the right state dictionary

[mueedurrehman]

I am using distiller's resnet 56 model for CIFAR. After quantization and retraining using QuantAwareTrainRangeLinearQuantizer, what is the procedure for instantiating an instance of that model and loading the correct layers from the state dictionary. The saved state dictionary contains the floating point values for each layer. The model from distiller does not have the additional layers that the quantized model does. Is the procedure to instantiate the same QuantAwareTrainRangeLinearQuantizer quantizer but with train_with_fp_copy=False, call prepare_model on my model and then restore all layers from the checkpoint's state dictionary excluding the floating point version of the weights?