app.py

import gradio as gr
from transformers import pipeline

# Carrega o modelo para rodar na CPU do seu Space
# Usamos o modelo de 0.5B pois ele cabe nos 16GB de RAM gratuitos
model_id = "Qwen/Qwen2.5-0.5B-Instruct"
pipe = pipeline("text-generation", model=model_id, device_map="cpu")

def chat_function(message, history):
    # Formata a mensagem para o modelo
    messages = [{"role": "user", "content": message}]
    
    # Gera a resposta
    results = pipe(messages, max_new_tokens=512, truncation=True)
    return results[0]['generated_text'][-1]['content']

# Cria a interface de Chat
demo = gr.ChatInterface(fn=chat_function, title="Meu Chat Qwen Local")

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