weight loading of quantized tiny-yolov2-seperable-conv2d + use separable conv model by default + testcases