Datasets:
File size: 3,459 Bytes
37d21bb | 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 | """The generation side of the command line: `simready generate`, `simready campaign`, `simready review`.
simready generate --track scenes|objects|demos --engine mujoco|isaac --id ID --cli claude|codex|opencode --model M [--budget 900 1800]
simready campaign run|status CONFIG [--generation-gpus GPU-a,GPU-b] [--evaluation-gpus GPU-c] [--render-gpu GPU-c]
simready review [--port 8794]
"""
from pathlib import Path
TRACKS = ("scenes", "objects", "demos")
def generate(args):
from .session import Session, shown
session = Session(args.track, args.engine, args.id, args.cli, args.model, args.effort, args.gpu, args.claude_config_dir,
campaign=args.campaign, runs_root=args.runs_root)
for run in session.run(args.budget):
print(shown(run), flush=True)
return 0
def campaign(args):
from . import campaign as runner
if args.action == "status":
print(runner.status(args.config))
else:
runner.main_run(args.config, args.generation_gpus, args.evaluation_gpus, args.render_gpu)
return 0
def review(args):
import runpy
from ..common.io import REPO_ROOT
server = runpy.run_path(str(REPO_ROOT / "web/review/serve.py"))
server["serve"](args.port)
return 0
def add_commands(commands):
"""Register the three subcommands; each sets `handler`."""
parser = commands.add_parser("generate", help="one isolated agent session on one benchmark item, over one or more budgets")
parser.add_argument("--track", choices=TRACKS, required=True)
parser.add_argument("--engine", choices=["mujoco", "isaac"], required=True)
parser.add_argument("--id", required=True, help="benchmark item id (data/benchmark/<track>/<id>)")
parser.add_argument("--cli", choices=["claude", "codex", "opencode"], required=True)
parser.add_argument("--model", required=True, help="e.g. claude-opus-5-5, gpt-6.1-sol, opencode-go/kimi-k3.1-pro")
parser.add_argument("--effort", help="reasoning effort (default medium; OpenCode: the model's default, or a --variant name)")
parser.add_argument("--budget", type=int, nargs="+", default=[900], help="cumulative wall-clock seconds; the first is what the agent is told")
parser.add_argument("--claude-config-dir", type=Path, help="Claude Code login folder (default ~/.claude), e.g. ~/.claude-huan")
parser.add_argument("--gpu", help="Isaac Sim: GPU UUID")
parser.add_argument("--campaign", default="", help="label recorded in run.json (e.g. smoke-test)")
parser.add_argument("--runs-root", type=Path, help="where run folders go (default data/runs)")
parser.set_defaults(handler=generate)
parser = commands.add_parser("campaign", help="run a campaign config (sessions and their evaluation), or show its status")
parser.add_argument("action", choices=["run", "status"])
parser.add_argument("config", type=Path)
parser.add_argument("--generation-gpus", default="", help="Isaac sessions: comma-separated GPU UUIDs")
parser.add_argument("--evaluation-gpus", default="", help="Isaac evaluation: comma-separated GPU UUIDs")
parser.add_argument("--render-gpu", help="GPU for MuJoCo evaluation's rendering")
parser.set_defaults(handler=campaign)
parser = commands.add_parser("review", help="the local review site (benchmark inputs as staged, evaluator-only data, metrics)")
parser.add_argument("--port", type=int, default=8794)
parser.set_defaults(handler=review)
|