README.md

metadata

title: LongListBench Leaderboard
emoji: 📊
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
license: mit
short_description: Extraction results on LongListBench (32 PDFs, 29.6k records)
datasets:
  - kaydotai/LongListBench

LongListBench Leaderboard

Leaderboard for LongListBench — complete document-to-list extraction from long business PDFs.

Submitting a result

1. Run your system on the dataset and score it with the reference evaluator.
2. Open a PR or issue on GitHub — or a discussion in this Space's Community tab — with your evaluation_report.json, run_metadata.json, and per-sample predictions.
3. Results are verified (scores must reproduce from the saved predictions) and added.

leaderboard_data.json holds the full numbers, including per-tier, per-family, and per-stressor slices.

This Space is generated by benchmarks/export_leaderboard_space.py from the saved evaluation reports in the repository.