app.py

import spaces
import transformers
import gradio as gr

def greet(name):
    return "Hello " + name + "!!"


@spaces.GPU
def infer():
    from transformers import pipeline

    messages = [
        {"role": "user", "content": "Who are you?"},
    ]
    pipe = pipeline("text-generation", model="microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True)
    return pipe(messages)

demo = gr.Interface(fn=infer, inputs="text", outputs="text")
demo.launch()