import gradio as gr import os import subprocess subprocess.run(["python3", "-m", "llama_cpp.server", "--hf_model_repo_id bartowski/gemma-2-9b-it-GGUF", "--model gemma-2-9b-it-Q6_K.gguf", "--host localhost", "--port 8000", "--chat_format gemma"]) with gr.Blocks() as demo: gr.Markdown("Running...") demo.launch()