It would be great to close the loop here by showing how to properly load the model and use it for inference later.
It seems simple, but using the above, and then loading using model = SentenceTransformer(save_dir) I get different results with the model before and after saving and loading.
