Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,30 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
viewer: false
|
| 4 |
+
tags:
|
| 5 |
+
- mathematical-reasoning
|
| 6 |
+
- llm-evaluation
|
| 7 |
+
- automated-grading
|
| 8 |
+
- olympiad-mathematics
|
| 9 |
+
pretty_name: RefGrader - Data and Logs
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# RefGrader: Data and Logs
|
| 13 |
+
|
| 14 |
+
Grading data and run logs from **RefGrader**, an automated system for grading
|
| 15 |
+
mathematical competition proofs against reference solutions.
|
| 16 |
+
|
| 17 |
+
RefGrader derives problem-specific grading rubrics from reference solutions and
|
| 18 |
+
applies them to score model-generated proofs, bringing automated grading
|
| 19 |
+
substantially closer to expert human judgment than a single-pass baseline.
|
| 20 |
+
|
| 21 |
+
## Contents
|
| 22 |
+
- `imo-shortlist.json` — model solutions to IMO Shortlist problems together with
|
| 23 |
+
reference-based grades and intermediate grading logs.
|
| 24 |
+
|
| 25 |
+
## Related publications
|
| 26 |
+
- **RefGrader: Automated Grading of Mathematical Competition Proofs** (MATH-AI Workshop, NeurIPS 2025) — arXiv:2510.09021
|
| 27 |
+
- **Brains vs. Bytes: Evaluating LLM Proficiency in Olympiad Mathematics** (COLM 2025) — arXiv:2504.01995
|
| 28 |
+
|
| 29 |
+
## Citation
|
| 30 |
+
If you use this data, please cite the RefGrader paper (arXiv:2510.09021).
|