Does anyone know how to increase the "source documents return" - k?

[dkampien]

In the updated repo, the makechain changed, there is no K variable anymore.

[Haste171]

This is the old code which uses a different function, not sure if you can with the new function logic it uses...

[palef-re]

interested in this - if you can't set a higher K value, is there a way to tweak the logic so more source documents are generally returned?

[flickersgit]

Here,

Put the number you desired in the 'as.retriever'.

const chain = ConversationalRetrievalQAChain.fromLLM(
  model,
  vectorstore.asRetriever(12),
  {
    qaTemplate: QA_PROMPT,
    questionGeneratorTemplate: CONDENSE_PROMPT,
    returnSourceDocuments: true,  
    //The number of source documents returned is 4 by default
  },
);
return chain;

[amitaknkpt]

Yes

Here,

Put the number you desired in the 'as.retriever'.

const chain = ConversationalRetrievalQAChain.fromLLM(
  model,
  vectorstore.asRetriever(12),
  {
    qaTemplate: QA_PROMPT,
    questionGeneratorTemplate: CONDENSE_PROMPT,
    returnSourceDocuments: true,  
    //The number of source documents returned is 4 by default
  },
);
return chain;

It worked for me

[ahhany]

if the number is high like vectorstore.asRetriever(40) the interface shows "Request failed with status code 400" , is there a way to change the interface to be similar to the chatgpt writing the response as it gets. and how can I get detailed response from my files if the response should be an analysis not just a simple question?

[jon-fuller-ukhsa]

Is there an equivalent but for Python?

[onalv]

seems to be like this as per this link Stackoverflow

In Python you can pass search_kwargs param to as_retriever() like: as_retriever(search_kwargs={"k": 3}). There could be a similar interface in JS binding probably.