Gyroscope readings unit

Hello all. I was trying to make sense of the gyroscope raw readings in hover. Since roll and pitch are zero (or extremely close to it), one can assume that the rate of change of yaw is approximately the same as the gyroscope reading G_z obtained under IMU (all three IMU readings are similar in this case) in the logs. Upon integrating G_z over a very short period of time (<2 seconds), I obtain a yaw attitude that is way less than the actual yaw. I’m assuming G_z has units of deg/secs as the log guide says. What would explain this discrepancy? How does ardupilot determine yaw rate to control it and try to drive it down to zero in flight modes like stabilize?