app.py

import gradio as gr
import requests
import json
import os

# Definir variáveis de ambiente ou substituir com sua chave de API real
API_KEY = os.getenv('API_KEY')
INVOKE_URL = "https://api.nvcf.nvidia.com/v2/nvcf/pexec/functions/df2bee43-fb69-42b9-9ee5-f4eabbeaf3a8"

headers = {
    "Authorization": f"Bearer {API_KEY}",
    "accept": "text/event-stream",
    "content-type": "application/json",
}

BASE_SYSTEM_MESSAGE = "I carefully provide accurate, factual, thoughtful, nuanced answers and am brilliant at reasoning."

def clear_chat(chat_history_state, chat_message):
    print("Clearing chat...")
    chat_history_state = []
    chat_message = ''
    return chat_history_state, chat_message

def user(message, history):
    print(f"User message: {message}")
    history = history or []
    history.append({"role": "user", "content": message})
    return history

def call_nvidia_api(history, max_tokens, temperature, top_p, seed=42):
    # Preparar o payload com o histórico de chat formatado
    messages = [{"role": "user" if i % 2 == 0 else "assistant", "content": msg} for i, msg in enumerate(history)]
    payload = {
        "messages": messages,
        "temperature": temperature,
        "top_p": top_p,
        "max_tokens": max_tokens,
        "seed": seed,
        "stream": True
    }
    
    response = requests.post(INVOKE_URL, headers=headers, json=payload, stream=True)
    full_response = ""
    for line in response.iter_lines():
        if line:
            decoded_line = line.decode("utf-8").strip()
            if decoded_line.startswith("data:"):
                try:
                    json_data = json.loads(decoded_line[5:])
                    # Processar a resposta da API aqui
                    # Supondo que a resposta da API seja diretamente o texto a ser adicionado ao chat
                    full_response += json_data.get("content", "")
                except json.JSONDecodeError:
                    print(f"Invalid JSON: {decoded_line[5:]}")
    return full_response

def chat(history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
    print("Starting chat...")
    # Chamar a API da NVIDIA aqui com o histórico formatado
    assistant_response = call_nvidia_api(history, max_tokens, temperature, top_p)
    # Atualizar o histórico com a resposta do assistente
    if history:
        history[-1][1] += assistant_response
    else:
        history.append(["", assistant_response])
    return history, history, ""


# Gradio interface setup
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column():
            gr.Markdown("## Your Chatbot Interface")
    chatbot = gr.Chatbot()
    message = gr.Textbox(label="What do you want to chat about?", placeholder="Ask me anything.", lines=3)
    submit = gr.Button(value="Send message")
    clear = gr.Button(value="New topic")
    system_msg = gr.Textbox(BASE_SYSTEM_MESSAGE, label="System Message", placeholder="System prompt.", lines=5)
    max_tokens = gr.Slider(20, 512, label="Max Tokens", step=20, value=500)
    temperature = gr.Slider(0.0, 1.0, label="Temperature", step=0.1, value=0.7)
    top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.95)
    chat_history_state = gr.State([])

    def update_chatbot(message, chat_history):
        print("Updating chatbot...")
        # Certifique-se de que a mensagem do usuário não seja adicionada duas vezes
        if not chat_history or (chat_history and chat_history[-1][0] != "user"):
            chat_history = user(message, chat_history)
        chat_history, _, _ = chat(chat_history, system_msg.value, max_tokens.value, temperature.value, top_p.value, 40, 1.1)
        # Formate as mensagens para exibição, removendo roles do conteúdo
        formatted_chat_history = [(msg["role"], msg["content"]) for msg in chat_history]
        return formatted_chat_history, chat_history, ""

    submit.click(
        fn=update_chatbot,
        inputs=[message, chat_history_state],
        outputs=[chatbot, chat_history_state, message]
    )

    clear.click(
        fn=clear_chat,
        inputs=[chat_history_state, message],
        outputs=[chat_history_state, message]
    )

demo.launch()