def get_user_agent()

in src/optimum/nvidia/utils/hub.py [0:0]

• 40 lines of code
• 10 McCabe index (conditional complexity)

def get_user_agent() -> str:
    """
    Get the library user-agent when calling the hub
    :return:
    """
    ua = USER_AGENT_BASE.copy()

    # Nvidia driver / devices
    try:
        nvmlInit()
        ua.append(f"nvidia/{nvmlSystemGetDriverVersion()}")

        num_gpus = nvmlDeviceGetCount()
        if num_gpus > 0:
            sm = []
            for device_idx in range(num_gpus):
                compute_capabilities = get_device_compute_capabilities(device_idx)
                if compute_capabilities:
                    major, minor = compute_capabilities
                    sm.append(f"{major}{minor}")

            ua.append(f"gpus/{num_gpus}")
            ua.append(f"sm/{'|'.join(sm)}")
    except (RuntimeError, ImportError):
        ua.append("nvidia/unknown")

    # Torch / CUDA related version, (from torch)
    try:
        from torch import __version__ as pt_version
        from torch.version import cuda, cudnn

        ua.append(f"cuda/{cuda}")
        ua.append(f"cudnn/{cudnn}")
        ua.append(f"torch/{pt_version}")
    except ImportError:
        pass

    # transformers version
    try:
        from transformers import __version__ as tfrs_version

        ua.append(f"transformers/{tfrs_version}")
    except ImportError:
        pass

    # TRTLLM version
    try:
        from tensorrt_llm._utils import trt_version

        ua.append(f"tensorrt/{trt_version()}")
    except ImportError:
        pass

    # Add a flag for CI
    if parse_flag_from_env("OPTIMUM_NVIDIA_IS_CI", False):
        ua.append("is_ci/true")
    else:
        ua.append("is_ci/false")

    return "; ".join(ua)