Finding dominant colors of an image using k-means clustering DominantColor

I used coremltools to convert a pytorch neural network (style transfer) model, then use add_activation(non_linearity='LINEAR') in coremltools.models.neural_network.builder.NeuralNetworkBuilder to make the output a matrix(NCHW) of [-1, 1], finally modified the output type to coremltools. ImageType. What should I do to convert the current mlmodel to mlprogram instead of re-converting the original pytorch project?