Does not actually limit the gpu within container

Python version: 3.9
Docker-py version: 7.0.0

When creating the container, I limit the GPU ID that can be seen within the container. However, In the container, it can still see all the GPUs. Key codes are as follows:

host_config=client.create_host_config(
    device_requests=[
        {
            'Driver': 'nvidia',
            'DeviceIDs': ['0'],
            'Capabilities': [["nvidia", "compute", "graphics","utility"]]
        }
    ]
)

container = client.create_container(image=image, command=None, detach=True, stdin_open=True, name=container_name, host_config=host_config)
print(container)

The result is below:

Somebody else has seen the same problem ?

Start with the Python 3.9 and docker-py 7.0.0 setup shown in the issue, then inspect the create_host_config device_requests and create_container calls. Reproduce the container with DeviceIDs set to ['0'] and compare the GPUs visible inside it with the expected restriction; done means the cause and a verified fix or limitation are documented.