qa_response_synth.py

from typing import List
from llama_index.response_synthesizers import BaseSynthesizer
from llama_index.indices.service_context import ServiceContext
from llama_index.prompts.prompts import RefinePrompt, QuestionAnswerPrompt
from llama_index.prompts.prompt_type import PromptType
from app.schema import Document as DocumentSchema
from app.chat.utils import build_title_for_document
from llama_index.response_synthesizers.factory import get_response_synthesizer


def get_custom_response_synth(
    service_context: ServiceContext, documents: List[DocumentSchema]
) -> BaseSynthesizer:
    doc_titles = "\n".join("- " + build_title_for_document(doc) for doc in documents)
    refine_template_str = f"""
A user has selected a set of SEC filing documents and has asked a question about them. \
The SEC documents have the following titles:
{doc_titles}
The original query is as follows: {{query_str}}
We have provided an existing answer: {{existing_answer}}
We have the opportunity to refine the existing answer \
(only if needed) with some more context below.
------------
{{context_msg}}
------------
Given the new context, refine the original answer to better \
answer the query. \
If the context isn't useful, return the original answer.
Refined Answer:
""".strip()
    refine_prompt = RefinePrompt(
        template=refine_template_str,
        prompt_type=PromptType.REFINE,
    )

    qa_template_str = f"""
A user has selected a set of SEC filing documents and has asked a question about them. \
The SEC documents have the following titles:
{doc_titles}
Context information is below.
---------------------
{{context_str}}
---------------------
Given the context information and not prior knowledge, \
answer the query.
Query: {{query_str}}
Answer:
""".strip()
    qa_prompt = QuestionAnswerPrompt(
        template=qa_template_str,
        prompt_type=PromptType.QUESTION_ANSWER,
    )

    return get_response_synthesizer(
        service_context,
        refine_template=refine_prompt,
        text_qa_template=qa_prompt,
        structured_answer_filtering=True,
    )