How to save and load tensors in the best way when you have very large data

I am facing a question that I would need more support from the community.
For example, I want to make a recommendation system. I turn every part of the product into an embed.

product_url_embedding: Embedding = self.get_product_url_embedding(product_url=product.product_url)
        categories_embedding: Embedding = self.get_categories_embedding(categories=product.categories)
        price_embedding: Embedding = self.get_price_embedding(price=product.price)

All good so far. But for example, I need products when I make recommendations in this form so that I can calculate the similarities with the user’s actions.
My question is:
How could I keep this data so that I don’t have a big load?
For example, I now save each embed in the database as a BYTEA, but this is expensive in terms of performance…
Any ideas?