app.py

import numpy as np
import pandas as pd
import openai
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores.faiss import FAISS
from langchain.docstore.document import Document
from langchain.prompts import PromptTemplate
from langchain.chains.question_answering import load_qa_chain
from langchain.llms import OpenAI
from langchain.text_splitter import RecursiveCharacterTextSplitter
import gradio as gr
import os
import PyPDF2
import zipfile

openai.api_key = "sk-"+os.environ['OPENAI_API_KEY']
embeddings = OpenAIEmbeddings()

df = pd.read_csv('reglamento-cdmx.csv')
text = df['text'].tolist()

text_splitter = RecursiveCharacterTextSplitter(
    # Set a really small chunk size, just to show.
    chunk_size=500,
    chunk_overlap=0,
    length_function=len,
)

texts = text_splitter.split_text(text)

docsearch = FAISS.from_texts(texts, embeddings)


def proper_query(query):
    prompt = f"The following text is a users' question: {query}\n\nHow should that question be modified so that it uses correct language?\nReturn the question in the same language.\nCorrected Question:"
    response = openai.Completion.create(
        engine="text-davinci-003", prompt=prompt, max_tokens=1000, temperature=0.2)
    return response.choices[0].text


def answer_question(query):
    query = proper_query(query)
    docs = docsearch.similarity_search(query)
    refine_prompt_template = (
        "The original question is as follows: {question}\n"
        "We have provided an existing answer: {existing_answer}\n"
        "You have the opportunity to refine the existing answer,"
        "only if needed, with the context below.\n"
        "------------\n"
        "{context_str}\n"
        "------------\n"
        "If that context is not helpful to answer the question, then omit it.\n"
        "Your answer should be correct, and concise.\n"
        "Shorten the answer if possible.\n"
        "Reply in the same language as the question.\n"
        "Answer:"
    )
    refine_prompt = PromptTemplate(
        input_variables=["question", "existing_answer", "context_str"],
        template=refine_prompt_template,
    )

    initial_qa_template = (
        "Context information is below. \n"
        "---------------------\n"
        "{context_str}"
        "\n---------------------\n"
        "Given the context information and not prior knowledge, "
        "answer the question: {question}\n"
    )
    initial_qa_prompt = PromptTemplate(
        input_variables=["context_str", "question"], template=initial_qa_template
    )
    chain = load_qa_chain(OpenAI(temperature=0), chain_type="refine", return_refine_steps=False,
                          question_prompt=initial_qa_prompt, refine_prompt=refine_prompt)
    answer = chain({"input_documents": docs, "question": query},
                   return_only_outputs=True)['output_text']
    return answer


demo = gr.Interface(
    fn=answer_question,
    inputs=[
        gr.Textbox(label="Hola soy tu asesor personal de tránsito, ¿cuál es tu pregunta? / Hi, I am your CDMX transit law personal assistant, ask me anything about Mexico City's transit law in any language.", lines=3,),
    ],
    outputs=[gr.Textbox(label="Respuesta / Answer: ")],
    title="Asesor de Reglamento de Tránsito CDMX",
    examples=[
        ["cuál es la multa por no llevar casco?"],
        ["qué pasa si no tengo licencia de conducir?"],
        ["What would happen if I drove under the influence of alcohol?"]
    ],
)

if __name__ == "__main__":
    demo.launch()