Yes, embeddings can be shared across systems. One of the key advantages of embeddings is their ability to be used as a generalized representation of data that can be transferred between different applications and systems. Once embeddings are generated, they can be serialized and stored in formats like JSON, binary, or HDF5, and then shared with other systems through APIs, cloud storage, or file transfers.
For example, in a recommendation system, embeddings for users and items can be computed in one system and shared with another system responsible for generating recommendations. Similarly, embeddings generated for text or images in one service can be shared with a search engine for similarity comparison. By sharing embeddings, organizations can integrate different services and models while maintaining efficiency, as the embeddings already encode useful features and relationships.
To ensure compatibility across systems, the systems must agree on the embedding's dimensionality, encoding format, and any preprocessing steps (e.g., tokenization or normalization) used before generating the embeddings. Additionally, it’s essential to consider the security of the shared embeddings to protect sensitive data when sharing across different environments.