app.py

import gradio as gr
from transformers import pipeline

pipe  = pipeline(
    "text-generation",
    model="microsoft/Phi-3-mini-4k-instruct"
)

def chat(message,history):
    response = pipe(message,max_new_tokens=150)[0]["generated_text"]
    return response

demo = gr.ChatInterface(fn=chat,title="Simple CPU chatbot")
demo.launch()