the file generated by .compile and .deploy for deep learning processor do not match each other

Hi, I'm using the deep learning processor in a Vivado project. I followed the documentation to generate the initialization files to use the processor without MATLAB connection as suggested by the documentation itself. I correctly generated the bias and weights files and the .dln containing the axi operations to correctly initialize the processor. I decode such file and translate its content in actual axi write and read.While analyzing the dcoded instructions I noticed that the weigths written on the memory are not the ones present on the weigths file I got generated. I'd like to know how this process is made and how to understand the order and the way in which the weigths are written on the memory