app.py

import gradio as gr
from llama_cpp import Llama

llm = Llama(model_path="ggml-model-q4_0.bin", n_ctx=256, n_batch=128)

def generate_text(input_text):
    print(input_text)
    output = llm(f"### Instruction:\n{input_text}\n\n### Response:\n", max_tokens=128, echo=False)
    print(output)
    return output['choices'][0]['text']

input_text = gr.inputs.Textbox(label="Enter your input text")
output_text = gr.outputs.Textbox(lines= 10, label="Output text")

gr.Interface(fn=generate_text, inputs=input_text, outputs=output_text, title="Alpaca GGML").launch()