from ctransformers import AutoModelForCausalLM
import gradio as gr

# Carrega CodeLlama 7B Instruct quantizado (GGUF) direto do Hugging Face
llm = AutoModelForCausalLM.from_pretrained(
    "TheBloke/CodeLlama-7B-Instruct-GGUF",
    model_file="codellama-7b-instruct.Q4_K_M.gguf",  # versão leve
    model_type="llama"
)

def chat(message, history):
    prompt = f"### Instrução:\n{message}\n### Resposta:\n"
    response = llm(prompt, max_new_tokens=512)
    return response

# Layout igual ao anterior (estilo chat)
iface = gr.ChatInterface(
    fn=chat,
    title="MEu ChatBot CodeLlama",
    description="Assistente de Programação rodando no Hugging Face Spaces (CodeLlama-7B Instruct).",
    theme="soft"
)

iface.launch()