"""Select a benchmark-owned service without importing native runtimes or models.""" from __future__ import annotations import argparse from importlib import import_module # These are executable service entrypoints, not the legacy adapter/campaign # registry. Keep task selection, native semantics and launch options in benchmarks. SERVICE_MODULES = { "behavior": "benchmarks.simBenchmarks.Behavior.server", "calvin": "benchmarks.simBenchmarks.calvin.server", "domino": "benchmarks.simBenchmarks.DOMINO.server", "libero": "benchmarks.simBenchmarks.LIBERO.server", "libero_plus": "benchmarks.simBenchmarks.LIBERO-plus.server", "metaworld": "benchmarks.simBenchmarks.MetaWorld.server", "robocasa365": "benchmarks.simBenchmarks.Robocasa_365.server", "robocasa_tabletop": "benchmarks.simBenchmarks.Robocasa_tabletop.server", "robodojo": "benchmarks.simBenchmarks.RoboDojo.server", "robotwin": "benchmarks.simBenchmarks.Robotwin.server", "simple_humanoid": "benchmarks.simBenchmarks.SIMPLE.server", "simpler_env": "benchmarks.simBenchmarks.SimplerEnv.server", "vla_arena": "benchmarks.simBenchmarks.VLA-Arena.server", } ALIASES = {"robocasa_365": "robocasa365"} class _SingleBenchmark(argparse.Action): def __call__(self, parser, namespace, value, option_string=None): if getattr(namespace, self.dest, None) is not None: raise argparse.ArgumentError(self, "specify --benchmark exactly once") setattr(namespace, self.dest, value) def main(argv=None) -> int: parser = argparse.ArgumentParser( prog="rsi-bench", allow_abbrev=False, description="List or launch a benchmark service for your agent's trusted driver.") commands = parser.add_subparsers(dest="command", required=True) commands.add_parser("list", allow_abbrev=False, help="list implemented development services") serve = commands.add_parser( "serve", add_help=False, allow_abbrev=False, help="launch one benchmark-owned service", description="Select a service and pass its launch arguments unchanged.", epilog="Use serve --benchmark ID --help for the selected service's launch arguments. " "Native environments and assets must be prepared separately.") serve.add_argument("--benchmark", choices=tuple(SERVICE_MODULES) + tuple(ALIASES), action=_SingleBenchmark, help="exact service ID (required except for generic help)") args, remaining = parser.parse_known_args(argv) if args.command == "list": if remaining: parser.error("unrecognized arguments: " + " ".join(remaining)) print("Development services; listing does not verify native readiness or full-suite support.") print("Documentation paths are repository-relative; installed wheels may omit these documents.") for identity, module in SERVICE_MODULES.items(): aliases = [alias for alias, target in ALIASES.items() if target == identity] suffix = f" (alias: {', '.join(aliases)})" if aliases else "" guide = module.rsplit(".", 1)[0].replace(".", "/") + "/README.md" print(f"{identity}{suffix}\t{guide}") return 0 # parse_known_args preserves benchmark-owned arguments. Do not let an # abbreviated selector accidentally reach a different service's parser. for value in remaining: option = value.split("=", 1)[0] if option.startswith("--") and option != "--" and "--benchmark".startswith(option): serve.error("use the exact --benchmark selector; abbreviations are not supported") if args.benchmark is None: if remaining and all(value in {"-h", "--help"} for value in remaining): serve.print_help() return 0 serve.error("the following argument is required: --benchmark") identity = ALIASES.get(args.benchmark, args.benchmark) # Selected help follows the same path: the existing launcher prints its # actual options before token validation, construction or output creation. return import_module(SERVICE_MODULES[identity]).main(remaining) if __name__ == "__main__": raise SystemExit(main())