app.py

import os
import gradio as gr
from huggingface_hub import InferenceClient

# Pega o token do Hugging Face dos Secrets
HF_TOKEN = os.environ.get("HF_TOKEN")

# Inicializa os clientes dos modelos
client_main = InferenceClient(token=HF_TOKEN, model="meta-llama/Llama-3.1-8B-Instruct")
client_aux1 = InferenceClient(token=HF_TOKEN, model="google/flan-t5-large")
client_aux2 = InferenceClient(token=HF_TOKEN, model="facebook/bart-large-cnn")

# Função principal de resposta
def respond(message, history, system_message, max_tokens, temperature, top_p):
    try:
        # Prompt principal incluindo system_message
        full_prompt = f"{system_message}\nUsuário: {message}"

        # --- Passo 1: Llama 3.1 (conversational) ---
        result_main = client_main.conversational(
            prompt=full_prompt,        # string diretamente
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p
        )
        response_main = result_main.generated_responses[-1]  # pega a última resposta

        # --- Passo 2: FLAN-T5 ---
        result_aux1 = client_aux1.text_generation(
            prompt=f"Reformule este texto de forma clara:\n{response_main}",
            max_new_tokens=max_tokens
        )
        response_aux1 = result_aux1.generated_text

        # --- Passo 3: BART ---
        result_aux2 = client_aux2.text_generation(
            prompt=f"Resuma este texto em 3 frases:\n{response_aux1}",
            max_new_tokens=150
        )
        response_aux2 = result_aux2.generated_text

    except Exception as e:
        response_aux2 = f"Erro ao gerar resposta: {e}"

    # Atualiza histórico no formato correto para o Gradio Chatbot
    history.append({"role": "user", "content": message})
    history.append({"role": "assistant", "content": response_aux2})

    return history, history

# Interface Gradio
with gr.Blocks() as demo:
    gr.Markdown("## 🤖 Chatbot em Cascata (Llama 3.1 + FLAN-T5 + BART)")

    system_message = gr.Textbox(
        value="Você é um chatbot amigável e prestativo.", 
        label="System Message"
    )

    chatbot = gr.Chatbot()
    msg = gr.Textbox(label="Digite sua mensagem")
    max_tokens = gr.Slider(50, 2048, 512, step=50, label="Max Tokens")
    temperature = gr.Slider(0.1, 1.0, 0.7, step=0.05, label="Temperature")
    top_p = gr.Slider(0.1, 1.0, 0.95, step=0.05, label="Top-p (nucleus sampling)")
    
    history = gr.State([])

    def handle_submit(message, history, system_message, max_tokens, temperature, top_p):
        return respond(message, history, system_message, max_tokens, temperature, top_p)

    msg.submit(
        handle_submit, 
        inputs=[msg, history, system_message, max_tokens, temperature, top_p], 
        outputs=[chatbot, history]
    )

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