File size: 2,662 Bytes
32030e4 | 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 | #!/usr/bin/env python3
"""Write the train split as chat prompts for RL fine-tuning (TRL, verl, Tinker, ...).
One JSON line per single-turn train task: ``prompt`` (chat messages, exactly what
``run_model.py`` sends), ``task_id`` and ``family``. Pair it with
``ejbench_eval.reward.TaskRewarder``, which scores completions by ``task_id``:
python eval/prepare_rl_prompts.py --dataset . --out rl/train_prompts.jsonl
from datasets import load_dataset
from trl import GRPOConfig, GRPOTrainer
from ejbench_eval.data import load_tasks
from ejbench_eval.reward import TaskRewarder
trainer = GRPOTrainer(model="Qwen/Qwen3.5-9B",
reward_funcs=[TaskRewarder(load_tasks(".", "train"))],
train_dataset=load_dataset("json", data_files="rl/train_prompts.jsonl")["train"],
args=GRPOConfig(output_dir="runs/grpo"))
Multi-step families need an environment loop (``StoredDataEnv``) and are left out here;
they are still evaluated, so their before/after change measures transfer.
Only ``train`` is accepted: never train on ``test`` or ``test_unseen_cell``.
"""
from __future__ import annotations
import argparse
import json
import sys
from collections import Counter
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent))
from ejbench_eval.data import FAMILIES, load_tasks # noqa: E402
from ejbench_eval.prompting import single_turn_messages # noqa: E402
def main(argv: list[str] | None = None) -> int:
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("--dataset", default=str(Path(__file__).resolve().parents[1]))
ap.add_argument("--families", default=",".join(FAMILIES))
ap.add_argument("--out", required=True)
args = ap.parse_args(argv)
tasks = [t for t in load_tasks(args.dataset, "train", [f for f in args.families.split(",") if f])
if t["interaction"] == "single_turn"]
if any(t["split"] != "train" for t in tasks):
raise ValueError("RL prompts must come from the train split only")
out = Path(args.out)
out.parent.mkdir(parents=True, exist_ok=True)
with out.open("w") as handle:
for task in tasks:
handle.write(json.dumps({"prompt": single_turn_messages(task), "task_id": task["task_id"],
"family": task["family"]}) + "\n")
print(json.dumps({"out": str(out), "prompts": len(tasks),
"by_family": dict(Counter(t["family"] for t in tasks))}, indent=2))
return 0
if __name__ == "__main__":
raise SystemExit(main())
|