YoloV4 having larger loss value than Yolov4_tiny

Question

I recently trained a yolov4 and yolov4-tiny model over the same dataset (batch size and subdivision are different). The loss value of Yolov4 (3.2) is a lot higher than the loss value of Yolov4_tiny (0.8) although yolov4 was having a mAp score of 75 and yolov4-tiny 64. Can someone please tell me the logical reason behind it?

apart from the network structure, the only difference is batch size and subdivision