Upload folder using huggingface_hub

README.md

@@ -0,0 +1,101 @@
+---
+language:
+- eng
+license: mit
+task_categories:
+- text-retrieval
+task_ids:
+- document-retrieval
+tags:
+- decontaminated
+- beir
+- information-retrieval
+configs:
+- config_name: corpus
+  data_files:
+  - split: corpus
+    path: corpus.parquet
+- config_name: queries
+  data_files:
+  - split: queries
+    path: queries.parquet
+- config_name: qrels-test
+  data_files:
+  - split: test
+    path: qrels_test.parquet
+- config_name: qrels-train
+  data_files:
+  - split: train
+    path: qrels_train.parquet
+- config_name: qrels-validation
+  data_files:
+  - split: validation
+    path: qrels_validation.parquet
+---
+
+# hotpotqa (Decontaminated)
+
+A decontaminated version of the [hotpotqa](https://huggingface.co/datasets/BeIR/hotpotqa) dataset from the BEIR benchmark, with samples found in the [mgte-en](https://huggingface.co/datasets/Alibaba-NLP/mgte-en) pre-training dataset removed.
+
+## Decontamination methodology
+
+Contamination was detected using a two-pass approach against the full mgte-en dataset (484 GB, 1,235 parquet files):
+
+### Pass 1: Exact hash matching
+
+All texts (queries and corpus documents) were normalized (lowercased, unicode NFKD, whitespace collapsed) and hashed with xxHash-64. The same normalization + hashing was applied to every `query` and `document` field in mgte-en. Any sample whose hash appeared in mgte-en was flagged as contaminated.
+
+### Pass 2: 13-gram containment (GPT-3 style)
+
+Following the methodology introduced in the GPT-3 paper (Brown et al., 2020), word-level 13-grams were extracted from all remaining samples. For each sample, containment was computed as:
+
+```
+containment = |ngrams_in_sample ∩ ngrams_in_mgte| / |ngrams_in_sample|
+```
+
+Samples with containment >= 0.5 were flagged as near-duplicates.
+
+### Qrels filtering
+
+Relevance judgments (qrels) referencing any removed query or corpus document were also removed.
+
+## Decontamination results
+
+| Component | Original | Clean | Removed |
+|---|---|---|---|
+| Corpus | 5,233,329 | 2,314,813 | 2,918,516 |
+| Queries | 97,852 | 12,932 | 84,920 |
+
+### Qrels per split
+
+| Split | Original | Clean | Removed |
+|---|---|---|---|
+| test | 14,810 | 2,296 | 12,514 |
+| train | 170,000 | 61 | 169,939 |
+| validation | 10,894 | 1,788 | 9,106 |
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+corpus = load_dataset("lightonai/hotpotqa-decontaminated", "corpus", split="corpus")
+queries = load_dataset("lightonai/hotpotqa-decontaminated", "queries", split="queries")
+```
+
+## Citation
+
+Please cite the original BEIR benchmark:
+
+```bibtex
+@inproceedings{thakur2021beir,
+  title={BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models},
+  author={Thakur, Nandan and Reimers, Nils and Rücklé, Andreas and Srivastava, Abhishek and Gurevych, Irena},
+  booktitle={NeurIPS Datasets and Benchmarks},
+  year={2021}
+}
+```
+
+## License
+
+MIT (same as original BEIR)

corpus.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c8150fae4283f9bbc7d884368061330571b22808c8faa4441ef8b15d83aa95d
+size 329922088

qrels_test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d8fa3cef2e42d323a526216f9afb26fd9367780b5442527ec13a1a33d927fe4
+size 53516

qrels_train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9e028c831056c13da6822aab69b69b98046f2ec6ade61fb6c8f4125a617dff3
+size 4304

qrels_validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f97298e30442a2f845339819e09e881797e7c7c48c5924568e62d3614e5f3969
+size 42005

queries.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70b6e881789842730e72496c1c35583e04397ef948a3dfce894bef76ebf75e68
+size 1105402