Retrieval QA and prompt templates

[ebayes]

Hello!

To improve the performance and accuracy of my document QA application, I want to add a prompt template but I'm unsure on how to incorporate LLMChain + Retrieval QA.

I found this helpful thread for the RetrievalQAWithSourcesChain library in python, but does anyone know if it's possible to add a custom prompt template for the RetrievalQAChain in JS too?

Thanks!

Ed

[onhate]

here it is

const template =
  'You are an customer support agent working for {organization}. ' +
  'Use the following pieces of context to answer the question at the end. ' +
  "If you don't know the answer, just say that you don't know, don't try to make up an answer.\n\n" +
  '{context}\n\n' +
  'Question: {question}\n' +
  'Helpful Answer:';

export const qaPrompt = new PromptTemplate({
  template,
  inputVariables: ['context', 'question', 'organization']
});

const retrievalChain = RetrievalQAChain.fromLLM(model, retriever, {
      returnSourceDocuments: true,
      prompt: qaPrompt
});

[DmitryKatson]

And how to pass 'organization' here?

[onhate]

const result = await retrievalChain.call({
  context: "",
  question: "",
  organization: ""
});

[ebayes]

Hey! Any idea how to change prompt template for ConversationalRetrievalQAChain in this context?

Code:
import {OpenAI} from "langchain/llms/openai";
import {pinecone} from "@/utils/pinecone-client";
import {PineconeStore} from "langchain/vectorstores/pinecone";
import {OpenAIEmbeddings} from "langchain/embeddings/openai";
import {ConversationalRetrievalQAChain} from "langchain/chains";

async function initChain() {
const model = new OpenAI({ modelName: "gpt-3.5-turbo-16k" }); //

const pineconeIndex = pinecone.Index(process.env.PINECONE_INDEX ?? '');

/* create vectorstore*/
const vectorStore = await PineconeStore.fromExistingIndex(
    new OpenAIEmbeddings({}),
    {
        pineconeIndex: pineconeIndex,
        textKey: 'text',
    },
);

return ConversationalRetrievalQAChain.fromLLM(
    model,
    vectorStore.asRetriever(),
    {returnSourceDocuments: true}
);

}

export const chain = await initChain()

Link: https://github.com/ebayes/gov-gpt/blob/main/utils/chain.ts

[mikeblackk]

did you figure this out?