def add_cli_args()

in arctic_inference/vllm/args.py [0:0]

• 24 lines of code
• 1 McCabe index (conditional complexity)

    def add_cli_args(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
        parser = EngineArgsPatch._orig_add_cli_args(parser)
        arctic_group = parser.add_argument_group(
            title="Arctic Inference",
            description="Arctic Inference configuration.",
        )
        arctic_group.add_argument(
            "--ulysses-sequence-parallel-size",
            type=int,
            default=ArcticEngineArgs.ulysses_sequence_parallel_size,
            help="Number of Ulysses sequence parallel replicas",
        )
        arctic_group.add_argument(
            "--enable-shift-parallel",
            action='store_true',
            help='If True, enable shift parallelism.')
        arctic_group.add_argument(
            "--shift-parallel-threshold",
            type=int,
            default=ArcticEngineArgs.shift_parallel_threshold,
            help=("Ulysses sequence parallel if batch size > threshold, "
                  "otherwise tensor parallel across the whole world size"),
        )
        return parser