app.py

import gradio as gr
from transformers import pipeline

# Load your large model
model = pipeline("text-generation", model="cognitivecomputations/dolphin-2.9-llama3-8b")

def generate_text(prompt):
    return model(prompt)[0]['generated_text']

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