app.py

import gradio as gr
from llama_cpp import Llama

llm = Llama(model_path="ggml-alpaca-7b-q4.bin")

def generate_text(input_text):
    output = llm(input_text, max_tokens=32, stop=["Q:", "\n"], echo=True)
    return output

input_text = gr.inputs.Textbox(label="Enter your input text")
output_text = gr.outputs.Textbox(label="Output text")

gr.Interface(fn=generate_text, inputs=input_text, outputs=output_text, title="Llama Language Model", description="Enter your input text to generate output text.").launch()