OpenRaiser/PaperFit

---
pretty_name: PaperFit Benchmark
license: other
size_categories:
- 100<n<1K
configs:
- config_name: cases
data_files:
- split: test
path: metadata/cases.jsonl
- config_name: perturbations
data_files:
- split: test
path: metadata/perturbations.jsonl
---

# PaperFit Benchmark

This dataset packages the PaperFit benchmark as a Hugging Face dataset repo.

## Included

- `metadata/cases.jsonl`: one row per benchmark case.
- `metadata/perturbations.jsonl`: one row per applied perturbation.
- Raw case archives are stored under `raw/<venue>/<paper_id>.tar.gz`.

- `metadata/benchmark_generation_report.json`: sanitized generation summary.

## Current Snapshot

- Cases: 200
- Perturbation rows: 761
- Selected cases in source report: 200
- Generated cases in source report: 180
- Skipped existing cases in source report: 20
- Failed cases in source report: 0

## Loading

```python
from datasets import load_dataset

cases = load_dataset("xuxinglong/paperfit-benchmark", "cases", split="test")
perturbations = load_dataset("xuxinglong/paperfit-benchmark", "perturbations", split="test")
```

## Notes

- Published metadata has local absolute paths removed.
- This staging export excludes macOS `._*` files and LaTeX build artifacts such as `.aux`, `.fls`, `.fdb_latexmk`, `.log`, `.out`, `.bbl`, and `.blg`.
- Check redistribution rights for source papers, figures, and templates before making the repo public.