import subprocess
import sys

# Uninstall the existing llama-cpp-python package
# uninstall_command = ["pip", "uninstall", "-y", "llama-cpp-python"]
# subprocess.run(uninstall_command, shell=True)

# Install llama-cpp-python with CUDA support
install_command = ["pip", "install", "--upgrade", "--force-reinstall", "llama-cpp-python", "--no-cache-dir"]
install_command_cmake_args = ["CMAKE_ARGS=-DLLAMA_CUBLAS=on", "FORCE_CMAKE=1"]
install_command.extend(install_command_cmake_args)
subprocess.run(install_command, shell=True)

# Start the Hugging Face Space
uvicorn_command = ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
subprocess.run(uvicorn_command, shell=True)
subprocess.run(install_command, shell=True)