app.py

import gradio as gr

gr.load("models/Qwen/Qwen2.5-72B").launch()
model = gr.load("model_path")  # Load the model only once at startup
def predict(input_data):
    return model(input_data)