Add dataset card for QuCo-RAG BM25 index

Hi, I'm Niels, part of the community science team at Hugging Face. I've added a dataset card for the QuCo-RAG Elasticsearch data archive. This PR includes:
- Relevant metadata including the `text-retrieval` task category.
- Links to the [associated paper](https://huggingface.co/papers/2512.19134) and the [official GitHub repository](https://github.com/ZhishanQ/QuCo-RAG).
- Usage instructions based on the GitHub documentation to help researchers set up the retrieval environment quickly.
- A BibTeX citation for the paper.

README.md

@@ -0,0 +1,54 @@
+---
+task_categories:
+- text-retrieval
+tags:
+- RAG
+- Elasticsearch
+- BM25
+- Wikipedia
+---
+
+# QuCo-RAG Elasticsearch Data Archive
+
+This repository contains the pre-built BM25 index (for Elasticsearch) of the Wikipedia dump used in the paper [QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation](https://huggingface.co/papers/2512.19134).
+
+- **GitHub Repository:** [https://github.com/ZhishanQ/QuCo-RAG](https://github.com/ZhishanQ/QuCo-RAG)
+- **Paper:** [QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation](https://huggingface.co/papers/2512.19134)
+
+## Introduction
+
+QuCo-RAG is a dynamic Retrieval-Augmented Generation (RAG) framework that adaptively determines when to retrieve during generation by quantifying uncertainty grounded in the pre-training corpus. This archive provides a pre-built BM25 index of approximately 21 million Wikipedia passages, allowing researchers to skip the time-consuming indexing process and quickly set up the retrieval environment.
+
+## Usage
+
+As documented in the official repository, you can use the provided startup script to automatically download this archive and configure Elasticsearch.
+
+To use the pre-built index, run the following command from the root of the QuCo-RAG repository:
+
+```bash
+bash Start_Elasticsearch_from_hf.sh
+```
+
+The script will:
+1. Check if Elasticsearch is already running.
+2. Download the pre-built index (~10GB) from this Hugging Face repository.
+3. Extract and start Elasticsearch with the index pre-loaded.
+4. Verify the index is working correctly.
+
+For users on HPC clusters with fast local SSD storage at `/tmp`, an optimized script is available:
+```bash
+bash Start_Elasticsearch_from_hf_HPC.sh
+```
+
+## Citation
+
+If you use this data or the QuCo-RAG framework in your research, please cite:
+
+```bibtex
+@article{min2025qucorag,
+  title={QuCo-RAG: Quantifying Uncertainty from the Pre-training Corpus for Dynamic Retrieval-Augmented Generation},
+  author={Min, Dehai and Zhang, Kailin and Wu, Tongtong and Cheng, Lu},
+  journal={arXiv preprint arXiv:2512.19134},
+  year={2025}
+}
+```