Can I export YOLOv5 to TensorFlow Lite using 8-bit quantization?

[mbe2014]

I was able to export YOLOv5 Nano to tflite using 8bit quantization, and a representative set of image.
The export went well, however, not surprisingly, it does not work well because the output tensor, which is a also 8Bit contains the BB
which in in range 0..IMAGE SIZE (640 by default), and confidence / probability numbers that are between 0..1.
and we cannot encode both with any degree of accuracy using scale+zero-point parametrization.

YOLO export doc (for YOLLO 11) only shows export to tflite:float32
https://docs.ultralytics.com/integrations/tflite/

The first question is: Did anyone successfully exported YOLOv5 to tflite 8bit?? and how?
The second question, Risking an XY question is: How can I change the model to produce normalized coordinates (between 0..1)?
I am not sure that it will work, but it will overcome the problem shown above.
(obviously I cannot normalize after the fact because the data already been lost)
Thank you!

[mbe2014]

Solved.
Export using Ultralytics export.py takes care of it.
Coordinates are normalized.