How to store embedding and then use it in the SamPredictor Class?

I would like to calculate embedding in a machine, then use it in another to do predictions. I have been trying to save it using

predictor = SamPredictor(sam)
predictor.set_image(img)
embedding = predictor.get_image_embedding().detach().cpu().numpy()
with open(pickle_file_name, "wb") as f:
pickle.dump(embedding, f)

then I upload the pickle file in another script but I don't know how to insert embedding again in the predictor avoiding to calculate embedding again with set_image(img).

Start with the SamPredictor methods named in the report—set_image and get_image_embedding—and the pickle save/load workflow. Trace how the predictor handles image embeddings, then document the cross-machine usage path; done means a saved embedding can be used for prediction without recomputing it.

Written by the indexing model from the issue text.