Then you should use the fiber mode. I guess it might be due to polarization mis-match with Gaussian. In addition, Gaussian beam is an approximation of the fiber mode. If the fiber is too large to be included in the simulation, you can first simulate the fiber mode and record data (most likely in air or a uniform material) and then import it as source for edge coupling. However, it eliminates the possible reflection from the fiber end surface, which should be small.