Let's make UNet2DConditionModel take in not only class labels, but also continuous labels

[Foundsheep]

The current UNet2DConditoinModel can take in class labels as described here.

However, in my case and also some other people are using it with continuous labels and in that case there needs to be nn.Linear-like layer in addition to nn.Embedding in about this place

As I mentioned, there seems to be some needs for it, and even if the paper does not address the issue in the way I described, this paper (Continuous Conditional Diffusion Models) is one of the trials to use continuous labels in diffusion models.

To be clearer, I'm trying to generate images based on conditions that are regression-like values, such as heights and widths