app.py

import gradio as gr
from ctransformers import AutoModelForCausalLM

# Cargar modelo GGUF
model = AutoModelForCausalLM.from_pretrained(
    "tinyllama-1.1b-chat-v1.0.Q4_0.gguf",
    model_type="llama",
    gpu_layers=0,
    threads=4
)

def chat_fn(message, history):
    prompt = (
        "<|system|>Eres un asistente útil especializado en trámites y administración en España.<|end|>\n"
        f"<|user|>{message}<|end|>\n"
        "<|assistant|>"
    )
    
    output = model(
        prompt,
        max_new_tokens=200,
        temperature=0.7
    )

    if isinstance(output, bytes):
        output = output.decode("utf-8", errors="ignore")

    return output

demo = gr.ChatInterface(
    fn=chat_fn,
    title="Lumen Admin Assistant",
    description="Chat en español usando Spanish-TinyLLaMA."
)

demo.launch()