app.py

from llama_index.core import SimpleDirectoryReader, GPTListIndex, GPTVectorStoreIndex, PromptHelper
from llama_index.core import StorageContext, load_index_from_storage, get_response_synthesizer
from llama_index.core.retrievers import VectorIndexRetriever
from llama_index.core.query_engine import RetrieverQueryEngine
from llama_index.core.postprocessor import SimilarityPostprocessor
#from llama_index import LLMPredictor, PromptHelper
#from langchain.chat_models import ChatOpenAI
from langchain_community.embeddings import OpenAIEmbeddings
import gradio as gr
import os
import openai
from gradio.themes.utils import colors, fonts, sizes

#os.environ['OPENAI_API_KEY'] = api_key

messages = [
    {"role": "system", "content": "follow the 4 instructions below for your outputs:"},
    {"role": "system", "content": "1. make sure all expressions are compatible with Polish"},
    {"role": "system", "content": "2. use Polish only for outputs"},
    {"role": "system", "content": "3. if you cannot answer, reply that you do not have enough information"},
    {"role": "system", "content": "4. do not make up any answer if you do know the answer"},
]

def construct_index(directory_path):
    max_input_size = 4096
    num_outputs = 512
    max_chunk_overlap = 0.05
    chunk_size_limit = 1000
    temperature = 0.1

    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
#    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=temperature, model_name="gpt-3.5-turbo-instruct", max_tokens=num_outputs))
    documents = SimpleDirectoryReader(directory_path).load_data()
    #index = GPTVectorStoreIndex(documents, llm_predictor=llm_predictor, prompt_helper=prompt_helper)
    index = GPTVectorStoreIndex.from_documents(documents, urls=[
                                                                'https://trio.house/',
                                                                'https://trio.house/kontakt/',
                                                                'https://trio.house/o-nas/',
                                                                'https://trio.house/w-sprzedazy/',
                                                                'https://trio.house/dzialki/',
                                                                'https://trio.house/zainwestuj-z-nami/',
                                                                'https://trio.house/potrzebujesz-konsultacji-rynku-nieruchomosci/',
                                                                'https://trio.house/potrzebujesz-remontu/',
                                                                'https://trio.house/potrzebujesz-projektu-wnetrza/',
                                                                'https://trio.house/potrzebujesz-mebli-na-wymiar/',
                                                                'https://trio.house/potrzebujesz-kredytu-na-zakup-nieruchomosci/',
                                                                'https://trio.house/makroekonomia/',
                                                                'https://trio.house/rynek-nieruchomosci/',
                                                                'https://trio.house/2023/05/24/deweloperzy-buduja-coraz-mniej/',
                                                                'https://trio.house/2023/04/27/prognozy-na-2023-2025-co-nas-czeka/',
                                                                'https://trio.house/2023/04/18/wycinka-drzew-na-wlasnej-dzialce-w-2023/',
                                                                'https://trio.house/2023/04/03/lipiec-rozpoczynamy-juz-w-kwietniu/',
                                                                'https://trio.house/2023/04/03/zmiany-w-podatku-od-czynnosci-cywilnoprawnych/',
                                                                'https://trio.house/2023/03/23/czy-aby-napewno-najdrozsze-mieszkania-sa-w-stolicy/',
                                                                'https://trio.house/2023/06/15/rekomendacja-s-korzystniejsza-dla-bezpiecznego-kredytu-2/',
                                                                'https://trio.house/2023/07/20/warszawski-rynek-nieruchomosci-mieszkaniowych-na-6-biegu/',
                                                                'https://livesmarter.pl/najlepsze-lokaty-maj-2023/',
                                                                'https://www.money.pl/gospodarka/inflacja-maj-2023-r-finalny-odczyt-gus-6909186710817344a.html',
                                                                'https://ksiegowosc.infor.pl/wiadomosci/5754337,oprocentowanie-lokat-bankowych-i-kont-oszczednosciowych-2023-koniec-maja-poczatek-czerwca-tabela.html#:~:text=7%2C05%25%20%2D%20takie%20jest,proc.',        
#        ], llm_predictor=llm_predictor, prompt_helper=prompt_helper)
        ], prompt_helper=prompt_helper)
    index.storage_context.persist('index.json')

    return index

def chatbotCustom(input):
    storage_context = StorageContext.from_defaults(persist_dir="index.json")
    index = load_index_from_storage(storage_context)
#    query_engine = index.as_query_engine()
#    response = query_engine.query(input, similarity_top_k=5, response_mode="tree_summarize")
##    response = index.query(input, similarity_top_k=5, response_mode="tree_summarize")
# configure retriever
    retriever = VectorIndexRetriever(
        index=index,
        similarity_top_k=10,
        response_mode="tree_summarize",
    )

# configure response synthesizer
    response_synthesizer = get_response_synthesizer()

# assemble query engine
    query_engine = RetrieverQueryEngine(
        retriever=retriever,
        response_synthesizer=response_synthesizer,
        node_postprocessors=[SimilarityPostprocessor(similarity_cutoff=0.7)],
    )

    response = query_engine.query(input)        
    return response.response

#def chatbotGPT(input):
#    if input:
#        messages.append({"role": "user", "content": input})
#        chat = openai.ChatCompletion.create(
#            model="gpt-3.5-turbo-instruct", messages=messages
#        )
#        reply = chat.choices[0].message.content
#        messages.append({"role": "assistant", "content": reply})
#        return reply

def clear():
    return None, None

theme = gr.themes.Default(font=[gr.themes.GoogleFont("Roboto"), "sans-serif", "sans-serif"], primary_hue="neutral", secondary_hue="neutral", neutral_hue="neutral").set(
    button_primary_background_fill="#3FCCA5",
    button_primary_background_fill_dark="#3FCCA5",
    button_primary_text_color="#003F62",
    body_background_fill="FFFFFF",
    body_background_fill_dark="FFFFFF"
)

with gr.Blocks(theme=theme) as trioGPT:
	inputs = gr.Textbox(lines=4, elem_id="inputs", label="Zadaj mi pytanie")#, elem_classes="textbox")
	outputs = gr.Textbox(label="Odpowiedź", elem_id="outputs")#, elem_classes="textbox")
	with gr.Row():
		submit_btn = gr.Button("Wyślij", variant="primary")
		clear_btn = gr.Button("Wyczyść")

	submit_btn.click(chatbotCustom, inputs=inputs, outputs=outputs)
	clear_btn.click(fn=clear, inputs=None, outputs=[inputs, outputs])

index = construct_index("data")
trioGPT.launch()#(share=True)