Datasets:
Download source/tools/replay.py from Syndicate604/science-desk: direct link, hf CLI and curl.
- Browser
- Download file 2.36 kB
-
https://huggingface.co/datasets/Syndicate604/science-desk/resolve/main/source/tools/replay.py
- Command line
-
hf download hf://datasets/Syndicate604/science-desk/source/tools/replay.py
-
curl -L -o replay.py https://huggingface.co/datasets/Syndicate604/science-desk/resolve/main/source/tools/replay.py
2.36 kB
| """Replay episodes against a running server over WebSocket, the way the arena plays them. | |
| For every task: reset, run `ls`, then play the given terminal action and check | |
| done=true with a reward in [0, 1]. Exits non-zero on the first failure. | |
| Usage: python tools/replay.py http://localhost:8000 [--tasks all] [--actions '[{"action_type":"exec","command":"ls -la"},{"action_type":"evaluate"}]'] | |
| """ | |
| import argparse | |
| import asyncio | |
| import json | |
| import sys | |
| import websockets | |
| sys.path.insert(0, ".") | |
| from terminal_desk.families import all_task_ids # noqa: E402 | |
| async def episode(url: str, task_id: str, actions: list[dict]) -> dict: | |
| async with websockets.connect(url.replace("http", "ws", 1) + "/ws", max_size=2**24) as ws: | |
| await ws.send(json.dumps({"type": "reset", "data": {"task_id": task_id}})) | |
| reset = json.loads(await ws.recv()) | |
| if reset.get("type") != "observation": | |
| raise SystemExit(f"{task_id}: reset failed: {reset}") | |
| obs = reset["data"] | |
| assert not obs["done"], obs | |
| result = None | |
| for action in actions: | |
| await ws.send(json.dumps({"type": "step", "data": action})) | |
| msg = json.loads(await ws.recv()) | |
| if msg.get("type") != "observation": | |
| raise SystemExit(f"{task_id}: step {action} failed: {msg}") | |
| result = msg["data"] | |
| if result["done"]: | |
| break | |
| if not (result and result["done"] and 0.0 <= float(result["reward"]) <= 1.0): | |
| raise SystemExit(f"{task_id}: no terminal reward in [0,1]: {result}") | |
| return {"task_id": task_id, "instruction_chars": len(obs["observation"]["instruction"]), "reward": result["reward"], | |
| "last_output": result["observation"]["output"][:80]} | |
| async def main() -> None: | |
| p = argparse.ArgumentParser() | |
| p.add_argument("url") | |
| p.add_argument("--tasks", default="all") | |
| p.add_argument("--actions", default='[{"action_type": "exec", "command": "ls -la"}, {"action_type": "evaluate"}]') | |
| args = p.parse_args() | |
| tasks = all_task_ids() if args.tasks == "all" else args.tasks.split(",") | |
| actions = json.loads(args.actions) | |
| for t in tasks: | |
| r = await episode(args.url, t, actions) | |
| print(json.dumps(r)) | |
| print(f"OK {len(tasks)} episodes") | |
| if __name__ == "__main__": | |
| asyncio.run(main()) | |