How does scale=True work in new v2 transforms?

vision
1

Mostly title, but, say in torchvision.transforms v2.ToDtype(torch.float32, scale=True) how exactly does scale=True scale the values? Min-max scaling? or something else.