is there a loss implementation for continuos targets(regression task)?

[tomer-erez]

i am encoding my input data into shape [batch,K], and my target is in shape [batch,K],
but each element in the target is continous between 0,1.
i want to make the encoder minimize the distance between vectors with closer embeddings, and push other away.
trying to do it without grouping items to bins/quantization.
do you have an idea how to modify the supcontrast to handle it?
basically asking for a suloution for a regression problem, not classification