How to reproduce result

quantization
1

I’m trying to run a easy deep learning model on embedding system without any framework

I have trained below model with my own dataset and quantize to int8

class Net(torch.nn.Module):
    def __init__(self, n_feature, n_hidden, n_output, quant=False):
        super(Net, self).__init__()
        self.fc1 = torch.nn.Linear(n_feature, n_hidden, bias=False)
        self.fc2 = torch.nn.Linear(n_hidden, n_output, bias=False)
        self.relu = torch.nn.ReLU()
        self.quant = quant
        if self.quant:
            self.quant = torch.quantization.QuantStub()
            self.dequant = torch.quantization.DeQuantStub()

    def forward(self, input):
        if self.quant:
            x = self.quant(input)
        else:
            x = input
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        if self.quant:
            x = self.dequant(x)

        return x

I can get each layer’s int8 weight as fc1.weight().int_repr()
But how to use these parameter to reproduce result like net.forward()?

2

Do you mean how to run the quantized pytorch model on your embedding system?