Expression language: parallel answering of retrieved documents

[andyrandy87]

Hi guys,
I'm using the new expression language and I managed to get n documents from the retriever, based on a question. Now I want to answer the question based on each individual document in parallel, to get n individual answers, which I will stuff together later. So my chain should look something like this:

chat_history | standalone_question | retrieve_documents | answer_each_doc_in_parallel | create_final_answer

Basically I want to implement map_reduce with the expression language. How do I do that? I built a retrieve that is for example returning 7 documents and I want to pass them to something that is implementing map_reduce.

Does someone know how to do it?
Many thanks in advance!

[dom-vaz]

I am also very interested in this. Is it already possible to use a retriever and distribute the retrieved documents in a map step and then combine them in a reduce step to a final answer via expression language?

[thiswillbeyourgithub]

Also looking for this, 8 months later

[dom-vaz]

It was added to Langchain via the map method. You can map the results of the retrieval to a chain.

Something like:

chain = retriever | map _chain.map() | reduce_chain

[thiswillbeyourgithub]

Thank you so much!