Add dataset card and link to paper
#2
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- question-answering
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# CorVer-infini-gram-index
|
| 8 |
+
|
| 9 |
+
This repository contains the prebuilt Wikipedia infini-gram index (tokenized with Llama-2) used for **CorVer** (*Corpus Verify*), a lightweight process reward for reinforcement learning fine-tuning on factual question answering.
|
| 10 |
+
|
| 11 |
+
The dataset was introduced in the paper [Verifiable Rewards Beyond Math and Code: Lightweight Corpus-Grounded Process Supervision for Factual Question Answering](https://huggingface.co/papers/2605.29648).
|
| 12 |
+
|
| 13 |
+
- **GitHub Repository**: [https://github.com/shichengf/CorVer](https://github.com/shichengf/CorVer)
|
| 14 |
+
- **Paper**: [https://huggingface.co/papers/2605.29648](https://huggingface.co/papers/2605.29648)
|
| 15 |
+
|
| 16 |
+
## Sample Usage
|
| 17 |
+
|
| 18 |
+
You can download the index using the `huggingface-cli`:
|
| 19 |
+
|
| 20 |
+
```bash
|
| 21 |
+
huggingface-cli download Shichengf/CorVer-infini-gram-index \
|
| 22 |
+
--repo-type dataset --local-dir ./infigram_index
|
| 23 |
+
export INFIGRAM_LOCAL_INDEX_DIR=./infigram_index
|
| 24 |
+
```
|
| 25 |
+
|
| 26 |
+
## Citation
|
| 27 |
+
|
| 28 |
+
```bibtex
|
| 29 |
+
@misc{fan2026corver,
|
| 30 |
+
title = {Verifiable Rewards Beyond Math and Code: Lightweight Corpus-Grounded Process Supervision for Factual Question Answering},
|
| 31 |
+
author = {Shicheng Fan and Haochang Hao and Dehai Min and Weihao Liu and Philip S. Yu and Lu Cheng},
|
| 32 |
+
year = {2026},
|
| 33 |
+
eprint = {2605.29648},
|
| 34 |
+
archivePrefix = {arXiv},
|
| 35 |
+
primaryClass = {cs.CL},
|
| 36 |
+
url = {https://arxiv.org/abs/2605.29648},
|
| 37 |
+
}
|
| 38 |
+
```
|