File size: 9,380 Bytes
6aa6c12 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 | """Command-line entry point for catalog inspection and evaluation runs."""
from __future__ import annotations
import argparse
import json
import os
import sys
from pathlib import Path
from typing import Any, Sequence
from benchkit.catalog import load_catalog, validate_repository_catalog
from benchkit.registry import create_benchmark, create_policy, registered
from AgentPolicies.policy_support.runner import EvaluationRunner, RunRequest
def _json_object(value: str) -> dict[str, Any]:
try:
parsed = json.loads(value)
except json.JSONDecodeError as exc:
raise argparse.ArgumentTypeError(str(exc)) from exc
if not isinstance(parsed, dict):
raise argparse.ArgumentTypeError("expected a JSON object")
return parsed
def _repository_root() -> Path:
return Path(__file__).resolve().parents[1]
def build_parser() -> argparse.ArgumentParser:
"""Create the CLI parser without importing simulator dependencies."""
parser = argparse.ArgumentParser(prog="starvla-eval", description="StarVLA-RSI embodied-agent evaluation")
subparsers = parser.add_subparsers(dest="command", required=True)
catalog_parser = subparsers.add_parser("catalog", help="list benchmark integration targets and status")
catalog_parser.add_argument("--json", action="store_true", help="emit machine-readable JSON")
catalog_parser.add_argument("--check", action="store_true", help="validate directories and protocol files")
list_parser = subparsers.add_parser("list", help="list runnable adapters installed in this environment")
list_parser.add_argument("--json", action="store_true")
preflight_parser = subparsers.add_parser(
"preflight-policy", help="check model-owned local runtime readiness without inference or a benchmark")
preflight_parser.add_argument("--model-profile", required=True)
preflight_parser.add_argument("--policy-args", type=_json_object, default={})
run_parser = subparsers.add_parser("run", help="run one benchmark-policy evaluation cell")
run_parser.add_argument("--benchmark", required=True)
policy_selection = run_parser.add_mutually_exclusive_group(required=True)
policy_selection.add_argument("--policy")
policy_selection.add_argument("--model-profile", help="select a frozen model-folder policy profile")
run_parser.add_argument("--benchmark-args", type=_json_object, default={})
run_parser.add_argument("--sim-server", help="evaluator-only http://127.0.0.1:port SIM endpoint")
run_parser.add_argument("--sim-server-token-env", default="RSI_SIM_SERVER_TOKEN")
run_parser.add_argument("--sim-server-timeout", type=float, default=1200.0)
run_parser.add_argument("--policy-args", type=_json_object, default={})
run_parser.add_argument("--output-dir", type=Path,
help="defaults to the selected model's results/<benchmark>/raw directory")
run_parser.add_argument("--seed", type=int, default=0)
run_parser.add_argument("--episodes", type=int)
run_parser.add_argument("--max-steps", type=int, default=500)
run_parser.add_argument("--max-policy-calls", type=int)
run_parser.add_argument("--replan-interval", type=int)
run_parser.add_argument("--budget-mode", choices=("horizon", "decisions"), default="horizon")
run_parser.add_argument(
"--policy-min-action-decisions-before-give-up", type=int,
help="opt-in persistence preference of the selected policy, not a benchmark budget",
)
run_parser.add_argument("--run-id")
run_parser.add_argument("--fail-fast", action="store_true")
smoke_parser = subparsers.add_parser("smoke", help="run the dependency-free end-to-end self-check")
smoke_parser.add_argument("--output-dir", type=Path, default=Path("results/rsi-smoke"))
smoke_parser.add_argument("--episodes", type=int, default=3)
smoke_parser.add_argument("--seed", type=int, default=0)
return parser
def _print_catalog(as_json: bool) -> None:
entries = load_catalog()
if as_json:
print(json.dumps([entry.__dict__ | {"tags": list(entry.tags)} for entry in entries], indent=2))
return
print(f"{'ID':<20} {'STATUS':<18} {'PRIORITY':<10} NAME")
for entry in entries:
print(f"{entry.id:<20} {entry.migration_status:<18} {entry.priority:<10} {entry.display_name}")
def _run(args: argparse.Namespace) -> Path:
profile = getattr(args, "model_profile", None)
if profile is None and str(getattr(args, "policy", "")).strip().lower() == "mailbox":
# Compatibility spelling, not a provenance bypass: the executable
# anonymous policy still gets its own exact model-source bundle.
profile = "anonymous_code_agent"
output_dir = args.output_dir
if output_dir is None:
output_dir = (_repository_root() / "AgentPolicies" / profile / "results" / args.benchmark / "raw"
if profile else Path("results/rsi"))
request = RunRequest(
output_dir=output_dir,
seed=args.seed,
episode_limit=args.episodes,
max_steps=args.max_steps,
max_policy_calls=args.max_policy_calls,
replan_interval=args.replan_interval,
run_id=args.run_id,
fail_fast=args.fail_fast,
repository=_repository_root(),
budget_mode=args.budget_mode,
)
policy_args = dict(args.policy_args)
minimum = getattr(args, "policy_min_action_decisions_before_give_up", None)
if minimum is not None:
key = "min_action_decisions_before_give_up"
if key in policy_args:
raise ValueError("policy persistence rule must be set once: use the flag or --policy-args")
policy_args[key] = minimum
if profile:
from benchkit.model_profiles import create_policy as create_profile_policy, load_profile, preflight_profile
if load_profile(profile).get("runtime_preflight"):
policy = preflight_profile(profile, **policy_args)
else:
policy = create_profile_policy(profile, **policy_args)
else:
policy = create_policy(args.policy, policy_args)
if getattr(args, "sim_server", None):
from benchkit.sim_server.client import SimServerBenchmarkAdapter
if args.benchmark_args:
raise ValueError("server owns benchmark parameters; --benchmark-args must be omitted")
benchmark = SimServerBenchmarkAdapter(args.sim_server,
token=os.environ.get(args.sim_server_token_env, ""), timeout=args.sim_server_timeout)
if benchmark.descriptor.name != args.benchmark:
raise ValueError("--benchmark must equal the connected server's descriptor.name")
else:
benchmark = create_benchmark(args.benchmark, args.benchmark_args)
return EvaluationRunner().run(benchmark, policy, request)
def _preflight(args: argparse.Namespace) -> int:
"""A model-only command: no run directory, benchmark import or request."""
from benchkit.errors import PolicyPreflightError
from benchkit.model_profiles import preflight_profile
code = None
try:
preflight_profile(args.model_profile, **args.policy_args)
except PolicyPreflightError as exc:
code = exc.code if exc.code in PolicyPreflightError.CODES else "initialization_failed"
except Exception:
code = "initialization_failed"
print(json.dumps({"status": "failed" if code else "passed", "scope": "local_runtime_only",
"error_code": code}))
return 1 if code else 0
def main(argv: Sequence[str] | None = None) -> int:
"""Execute one CLI command and return a process status code."""
parser = build_parser()
args = parser.parse_args(argv)
try:
if args.command == "preflight-policy":
return _preflight(args)
if args.command == "catalog":
_print_catalog(args.json)
if args.check:
errors = validate_repository_catalog(_repository_root())
if errors:
for error in errors:
print(f"ERROR: {error}", file=sys.stderr)
return 1
print("catalog layout: OK", file=sys.stderr)
return 0
if args.command == "list":
entries = registered()
if args.json:
print(json.dumps(entries, indent=2, sort_keys=True))
else:
for kind, names in entries.items():
print(f"{kind}: {', '.join(names) or '(none)'}")
return 0
if args.command == "smoke":
args.benchmark = "smoke"
args.policy = "scripted"
args.benchmark_args = {}
args.policy_args = {}
args.max_steps = 10
args.max_policy_calls = 10
args.replan_interval = 1
args.budget_mode = "horizon"
args.run_id = None
args.fail_fast = True
run_dir = _run(args)
elif args.command == "run":
run_dir = _run(args)
else:
parser.error(f"unsupported command: {args.command}")
return 2
except Exception as exc:
print(f"error: {type(exc).__name__}: {exc}", file=sys.stderr)
return 1
print(run_dir)
summary = json.loads((run_dir / "summary.json").read_text(encoding="utf-8"))
return 1 if summary["errors"] else 0
|