sairam954 / quantizationawaretrainingpcm Goto Github PK

First thing that I see that in your .ipynb file is not correct is the below line(which in there the trainedmodel is not declared.

driftedmodelstatedict = updateWeightsWithTimePCM(trainedmodel,vdict,kdict,t) # returns the state dictionary of the model w

Second thing that I want to ask you that why the weights in intmodel is not quantized and still in floating point precision ? I mean the results of the below lines execution!

print(intmodel.state_dict()['fc3.weight'])
print("====================================================")
print(driftedmodel.state_dict()['fc3.weight'])

I also changed quant type from FP to INT, but for example model.conv1.weight is in floating point precision
Thank for you'r great files.