app.py

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

model_id = "TheBloke/phi-2-GPTQ"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=False  # Força desabilitar quantização
)

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    trust_remote_code=True,
    quantization_config=bnb_config
)
# Pipeline de texto
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Função do chat + salvar memória
def chat(user_input, history):
    prompt = user_input
    result = pipe(prompt, max_new_tokens=256, temperature=0.7)[0]["generated_text"]

    # Salvar memória em arquivo
    with open("memoria.txt", "a", encoding="utf-8") as f:
        f.write(f"User: {user_input}\nAI: {result}\n")

    return result

# Interface Gradio
with gr.Blocks() as demo:
    chat_history = gr.State([])
    chatbot = gr.Chatbot()
    msg = gr.Textbox(label="Digite sua pergunta:")

    def respond(user_input, chat_history):
        answer = chat(user_input, chat_history)
        chat_history.append((user_input, answer))
        return chat_history, chat_history

    msg.submit(respond, [msg, chat_history], [chatbot, chat_history])

demo.launch()