Can I get un-normalized vectors from the TF USE model?

I'm using this Universal Sentence Encoder (USE) model to get embeddings of a set of texts, each text corresponding to a newspaper article. In order to build a Recommender System, I generate user embeddings by averaging the embeddings of items a user has read, and then I look for other texts that are cosine-similar to this user (basically, the method returns a set of items that are similar to this user embedding).

Now, the problem is that the mentioned model returns approximately normalized vectors, as you can see here. I'm not sure if the averaging over text-vectors to get a user-vector makes sense if the vectors are already normalized.

1. Can the above USE model also return the original embeddings (not normalized)? Or does it fit the normalized vectors?
2. What would be the alternative way to combine the normalized text-vectors to get a user-vector, given that the text-vectors are normalized?

Please feel free to suggest other models where the above described averaging would make sense.