app.py

import os
import openai
from transformers import pipeline, Conversation
import gradio as gr
import json
from dotenv import load_dotenv

# Load environment variables from the .env file de forma local
load_dotenv()
import base64

with open("Iso_Logotipo_Ceibal.png", "rb") as image_file:
    encoded_image = base64.b64encode(image_file.read()).decode()

openai.api_key = os.environ['OPENAI_API_KEY']

def clear_chat(message, chat_history):
     return "", []

def add_new_message(message, rubrica, chat_history):
     new_chat = []
     
     new_chat.append({"role": "system", "content": 'Sos {} y tendrás que responder preguntas que te harán niños de escuela, las respuestas tienen que ser cómo si tu fueras {} y responder con la información de su vida. Las respuestas tienen que estar orientadas a niños entre 9 y 10 años. No respondas preguntas hasta que el usuario te pregunte sobre algún tema.'.format(rubrica,estilo)})
   
     for turn in chat_history:
          user, bot = turn
          new_chat.append({"role": "user", "content": user})
          new_chat.append({"role": "assistant","content":bot})
     new_chat.append({"role": "user","content":message})
     return new_chat

def respond(message, rubrica, chat_history):
    prompt = add_new_message(message, rubrica, chat_history)
    # stream = client.generate_stream(prompt,
    #                                   max_new_tokens=1024,
    #                                   stop_sequences=["\nUser:", "<|endoftext|>"],
    #                                   temperature=temperature)
    #                                   #stop_sequences to not generate the user answer
    # acc_text = ""
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages= prompt,
        temperature=0.5,
        max_tokens=1000,
        stream=True,
        )#.choices[0].message.content
    #chat_history.append((message, response))

    token_counter = 0 
    partial_words = "" 

    counter=0
    for chunk in response:
        chunk_message = chunk['choices'][0]['delta']
        if(len(chat_history))<1:
            # print("entró acaá")
            partial_words += chunk_message.content
            chat_history.append([message,chunk_message.content])
        else:
            # print("antes", chat_history)
            if(len(chunk_message)!=0):
                if(len(chunk_message)==2):
                    partial_words += chunk_message.content
                    chat_history.append([message,chunk_message.content])
                else:
                    partial_words += chunk_message.content
                    chat_history[-1] =([message,partial_words])
        yield "",chat_history


with gr.Blocks() as demo:
    gr.Markdown("""
    <center>
    <h1>
    Uso de AI para un chatbot.
    </h1>
    <img src='data:image/jpg;base64,{}' width=200px>
    <h3>
    Este espacio permite probar la generación mediante IA de devoluciones en base a una rúbrica.
    </h3>
    </center>
    """.format(encoded_image))
    with gr.Row():
        rubrica = gr.Textbox(lines=5, label="Escribe la rúbrica que quieres usar para generar la devolución.")
    with gr.Row():
        with gr.Row():
            with gr.Column(scale=4):
                msg = gr.Textbox(lines=5, label="Texto de entrada para ser evaluado y generar devolución.")
            with gr.Column(scale=1):
                btn = gr.Button("Enviar")
                clear = gr.ClearButton(components=[msg, chatbot], value="Borrar chat")        
    with gr.Row():
        chatbot = gr.Chatbot(lines=10) #just to fit the notebook

    btn.click(respond, inputs=[msg, rubrica, chatbot], outputs=[msg, chatbot])
    msg.submit(respond, inputs=[msg, rubrica,chatbot], outputs=[msg, chatbot]) #Press enter to submit
    clear.click(clear_chat,inputs=[msg, chatbot], outputs=[msg, chatbot])
demo.queue()
demo.launch()