What pooling technique is used to get an embedding vector for a question or text chunk?

[RonanKMcGovern]

I've been reading through the privateGPT code and langchain docs:

1. It's unclear to me whether the question is represented as a tensor (vector of token embeddings) or a vector (which I assume would be achieved through some form of pooling). Same for document chunks (is the embedding a tensor or vector).
2. If indeed embeddings are vectors, then what kind of pooling is being used of the individual token embeddings?

Appreciate if someone could point me to the right part of the docs. Thanks