Fine tuning a sentence transformer model for [single_sentence, label] format?

I tried finding a fine tuning example of that data format but couldnt find any, I’m using Tensorflow as my framework and looking to do a similarity-based classification.

My idea is:
Lets say I train on a (sentences, labels) format, I then want to sample random embedding of each label and during inference, classify the inference input by the most similar embedding.