Datasets:
File size: 2,633 Bytes
af5b4af | 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 | """Build submission.json for the OpenEnv Arena.
Reads the schema from a running server, so the submitted schema is exactly what
GET /schema returns. Tasks come from --tasks (comma list) or a file with one id per line.
Usage:
python tools/make_submission.py --url http://localhost:18001 --image ghcr.io/syndicate604/science-desk@sha256:... \
--tasks-file runs/selected.txt --id science-desk-v5 --out submission.json
"""
import argparse
import json
import sys
import httpx
sys.path.insert(0, ".")
from terminal_desk.families import parse_task_id # noqa: E402
# Copied verbatim from the arena operator's own smoke submission (benchflow/openenv-smoke-fix-git),
# so training runs under the budgets the private tasks are believed to use.
LIMITS = {
"split": "train",
"cleanup_grace_s": 30,
"completion_tokens": 4096,
"context_tokens": 8192,
"cpu_floor_vcpus": 1.0,
"memory_gib": 2.0,
"reset_wall_s": 300,
"rollout_wall_s": 902,
"tool_calls_inflight": 1,
"tool_calls_per_minute": 60,
"tool_calls_total": 32,
"tool_rate_burst": 4,
"tool_wall_s": 110,
"verifier_wall_s": 600,
"workspace_gib": 10,
}
def main() -> None:
p = argparse.ArgumentParser()
p.add_argument("--url", required=True)
p.add_argument("--image", required=True)
p.add_argument("--id", required=True)
p.add_argument("--tasks", default="")
p.add_argument("--tasks-file", default="")
p.add_argument("--dataset", default="Syndicate604/science-desk")
p.add_argument("--name", default="Science Desk")
p.add_argument("--out", required=True)
a = p.parse_args()
ids = [t for t in a.tasks.split(",") if t] if a.tasks else [
ln.strip() for ln in open(a.tasks_file) if ln.strip() and not ln.startswith("#")
]
for t in ids:
parse_task_id(t)
if not 1 <= len(ids) <= 50:
raise SystemExit(f"need 1-50 tasks, got {len(ids)}")
schema = httpx.get(f"{a.url}/schema", timeout=10).json()
sub = {
"submission_id": a.id,
"name": a.name,
"image": a.image,
"schema": {"action": schema["action"], "observation": schema["observation"]},
"tasks": [{"task_id": t, **LIMITS} for t in ids],
"example_actions": [{"action_type": "exec", "command": "ls -la"}, {"action_type": "evaluate"}],
"finish_action": {"action_type": "evaluate"},
"dataset": a.dataset,
"source": f"https://huggingface.co/datasets/{a.dataset}",
}
json.dump(sub, open(a.out, "w"), indent=2)
print(f"wrote {a.out}: {len(ids)} tasks, image {a.image}")
if __name__ == "__main__":
main()
|