Actually, I'm not sure if this is really bad news because I didn't have high expectations for SD3 at this point.
In summary, according to a source from [Reddit]
Stability AI (SAI) has removed a component from the model that allows for fine-tuning without crashes. This means that SAI has not released the official training code, and they have also removed the QK Norm layer, making fine-tuning the model practically impossible.