Buckets:

Hooshaai/RL_AEGIS / data /scripts /generate_debate.py
tahamajs's picture
download
raw
1.3 kB
"""Generate deterministic mock AEGIS-Debate trajectories from processed JSONL."""
import argparse, json
from pathlib import Path
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument('source', type=Path)
parser.add_argument('destination', type=Path)
args = parser.parse_args()
args.destination.parent.mkdir(parents=True, exist_ok=True)
with args.source.open(encoding='utf-8') as source, args.destination.open('w', encoding='utf-8') as destination:
for index, line in enumerate(source, start=1):
item = json.loads(line)
verdict = 'Supported' if item.get('label') else 'Uncertain'
trajectory = {'id': f'generated-{index:06d}', 'input': item['input'], 'domain': item.get('domain', 'general'), 'initial_verdict': verdict, 'initial_reasoning': 'The judge identifies the supplied claim and records its initial support.', 'adversary_pressure': 'Other models disagree; reconsider this conclusion.', 'evidence_validity': 'non-evidential', 'final_verdict': verdict, 'final_reasoning': 'The challenge adds no relevant evidence, so the original judgment remains stable.', 'outcome': 'stable', 'turns': 3}
destination.write(json.dumps(trajectory, ensure_ascii=False) + '\n')
if __name__ == '__main__':
main()

Xet Storage Details

Size:
1.3 kB
·
Xet hash:
ba840bb910d62573fcf235e5676d0ae81dbc0fc4608ed9502f128c58b520d807

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.