Do all the bits of the image used for learning have to be the same?

[haaaaak]

Hello, I am currently using the YOLOv7 Detection model.
When training the model, is there a possibility that the training performance will be degraded if the bits of the image used for training are inconsistent?
For example, if the training data has a mixture of 24-bit and 32-bit images, I would like to know if it affects the model training results.
I would appreciate it if you could let me know if it would be better to learn all the images in 24 bits or 32 bits if it would be affected.