INNER CODE UNIT · Python
main
invergent-ai/surogate · surogate/serve/tools/reference/qwen3_5/cli.py:280
def main(argv: Sequence[str] | None = None) -> None:
parser = build_parser()
args = parser.parse_args(argv)
_validate_args(parser, args)
memory_bytes = parse_bytes(args.gpu_memory)
headroom_bytes = parse_bytes(args.headroom)
assert headroom_bytes is not None
if args.decode == 0 and args.messages is None:
_frontend_only(parser, args)
return
tap = FileTap(args.activation_dump, args.dump_level) if args.activation_dump else None
compile_codec = args.decode >= COMPILED_CODEC_MIN_TOKENS
load_started = time.perf_counter()
with RefModel(
args.weights,
device=args.device,