app.py

import gradio as gr
import spaces
import transformers_gradio

demo = gr.load(name="MakiAi/Llama-3.2-3B-Instruct-bnb-4bit-OKU_wiki_llama3.1_8b_inst_Reflexive_chunk200_overlap700-10epochs", src=transformers_gradio.registry)
demo.fn = spaces.GPU()(demo.fn)

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