Why are the model Flops smaller than ultralytics documentation?

[mschneider1711]

When I use the script get_flops.py in the analysis Tools the amount of Flops i get for YOLOv8 are roughly half the size than the official ultralytics documentation for every size:

E.g.
Yolov8n: 8.7 Flops (Ultralytics) compared to 4.4 (MMYOLO)

It it the same for larger Yolov8 versions as well.