def load_from_name(name: str, device: Union[str, torch.device] = "cuda" if torch.cuda.is_available() else "cpu",
                   download_root: str = None, vision_model_name: str = None, text_model_name: str = None, input_resolution: int = None):
    if name in _MODELS:
        model_path = _download(_MODELS[name], download_root or os.path.expanduser("~/.cache/clip"))
        model_name, model_input_resolution = _MODEL_INFO[name]['struct'], _MODEL_INFO[name]['input_resolution']
    elif os.path.isfile(name):
        assert vision_model_name and text_model_name and input_resolution, "Please specify specific 'vision_model_name', 'text_model_name', and 'input_resolution'"
        model_path = name
        model_name, model_input_resolution = f'{vision_model_name}@{text_model_name}', input_resolution
    else:
        raise RuntimeError(f"Model {name} not found; available models = {available_models()}")

    with open(model_path, 'rb') as opened_file:
        # loading saved checkpoint
        checkpoint = torch.load(opened_file, map_location="cpu")

    model = create_model(model_name, checkpoint)
    if str(device) == "cpu":
        model.float()
    else:
        model.to(device)
    return model, image_transform(model_input_resolution)