Datasetpublic

Hyukkyu/train-dpr-wikipedia

DPRWikipedia — Training, unified schema A normalised copy of the dataset behind the mteb task DPRWikipedia, a retrieval training set built from Tevatron/wikipedia-nq-corpus. Same queries, documents and relevance judgements as the benchmark evaluates — reshaped into one strict schema shared by every dataset in this collection. Source Tevatron/wikipedia-nq-corpus @ 56c6e2438c13 (the revision pinned in mteb) Domain · languages Wikipedia QA (DPR) · eng Queries /… See the full description on the dataset page: https://huggingface.co/datasets/Hyukkyu/train-dpr-wikipedia.

sourceHugging Facecc-by-sa-3.0updated 9d agoView on Hugging Face

0likes155downloads

Dataset Card

DPRWikipedia — Training, unified schema

A normalised copy of the dataset behind the mteb task `DPRWikipedia`, a retrieval training set built from `Tevatron/wikipedia-nq-corpus`. Same queries, documents and relevance judgements as the benchmark evaluates — reshaped into one strict schema shared by every dataset in this collection.


Source	`Tevatron/wikipedia-nq-corpus` @ `56c6e2438c13` (the revision pinned in `mteb`)
Domain · languages	Wikipedia QA (DPR) · eng
Queries / documents / qrels (all splits)	210,110 / 21,015,324 / 1,711,566
Qrels per query	min 1 · mean 7.495 · max 93
Score values	1 ×439,345
Layout	`{subset}-queries` · `{subset}-corpus` · `{subset}-qrels` for subset ∈ {nq, trivia, squad}, split `test`
Splits	`corpus`: train · `nq-hard-negatives`: train · `nq-qrels`: train, dev · `nq-queries`: train, dev · `nq-teacher-scores`: train · `squad-hard-negatives`: train · `squad-qrels`: train, dev · `squad-queries`: train, dev · `squad-teacher-scores`: train · `trivia-hard-negatives`: train · `trivia-qrels`: train, dev · `trivia-queries`: train, dev · `trivia-teacher-scores`: train
Hard negatives	`nq-hard-negatives`: sources: `dense` · 5,701,040 rows; `squad-hard-negatives`: sources: `dense` · 6,891,826 rows; `trivia-hard-negatives`: sources: `dense` · 5,752,957 rows
Teacher scores	`nq-teacher-scores`: `jinaai/jina-reranker-v3.5` · 5,759,662 rows (positives included); `squad-teacher-scores`: `jinaai/jina-reranker-v3.5` · 6,961,864 rows (positives included); `trivia-teacher-scores`: `jinaai/jina-reranker-v3.5` · 5,813,253 rows (positives included)
Ids	`source ids (Tevatron query_id / DPR docid)`; identical texts collapse to one document (0 collapsed)
Pair recovery	1,793,915 of 1,793,915 source pairs reconstructed from `queries` × `qrels` × `corpus` with byte-equal text; 1,693 of them have a title that the source spells in a different Unicode normalization form than its corpus file (counted, kept as in the corpus)
License	`cc-by-sa-3.0`

subset	split	queries	corpus	qrels	qrels/query
nq	train	58,622	21,015,324	439,345	7.495
nq	dev	6,489	21,015,324	48,723	7.509
trivia	train	60,296	21,015,324	730,095	12.109
trivia	dev	6,751	21,015,324	81,778	12.113
squad	train	70,038	21,015,324	369,436	5.275
squad	dev	7,914	21,015,324	42,189	5.331

Schema

config	columns	rules
`queries`	`id: string`, `text: string`	ids unique and non-empty; every query has ≥ 1 qrel
`corpus`	`id: string`, `title: string`, `text: string`	`title` is always present (`""` when the source has none)
`qrels`	`query-id: string`, `corpus-id: string`, `score: int32`	referential integrity to both tables; no duplicate pairs; no floats
`hard-negatives`	`query-id: string`, `corpus-id: string`, `rank: int32`, `source: string`	one row per mined negative; `(query-id, corpus-id, source)` unique; never a labelled positive of the same query
`teacher-scores`	`query-id: string`, `corpus-id: string`, `teacher: string`, `score: float32`	one row per scored pair (positives included); a row means scored — never a placeholder

Files are Parquet, sorted by id, zstd-compressed, sharded at 500 MB. Every rule above is enforced by a validator before publishing; provenance.json records the source file hashes, what changed, and the output file hashes.

What changed from the source

—byte-preserved all text — no whitespace, newline, or control-character normalisation
—dropped non-schema columns from qrels: negative_passages (BM25 negatives, left out by D9)
—dropped non-schema columns from queries: answers
—kept 44,540 documents whose text is identical to another document's — a property of the source; note that models embed such documents identically and retrieval ties among them are broken by corpus order

Hard negatives and teacher scores

Filled by the owner's annotation pipeline (annotation=jina35) for the train split of the query set(s) below; queries without a labelled positive are left out.

—Candidates: dense retrieval with jinaai/jina-embeddings-v5-text-small over the full corpus to depth 1,000; 100 candidates per query drawn from the rank windows 1–30 (30), 31–100 (30), 101–300 (20), 301–1000 (20), the query's labelled positives excluded. rank is the dense rank; source is dense for a mined row and dataset for a negative the source labels itself (those are kept for every query of the split, sampled or not).
—Teacher: jinaai/jina-reranker-v3.5, listwise: a query's positive and all of its candidates are scored together in one context of up to 32,768 tokens. score is the raw cosine score, one row per (query, positive) and per (query, candidate); a labelled negative that was also mined is scored once. No filtering is applied to the tables.

configs	queries	hard negatives	teacher scores
`nq-hard-negatives` · `nq-teacher-scores`	58,622 (all)	5,701,040 (5,701,040 dense)	5,759,662
`squad-hard-negatives` · `squad-teacher-scores`	70,038 (all)	6,891,826 (6,891,826 dense)	6,961,864
`trivia-hard-negatives` · `trivia-teacher-scores`	60,296 (all)	5,752,957 (5,752,957 dense)	5,813,253

python

from datasets import load_dataset
negatives = load_dataset("Hyukkyu/train-dpr-wikipedia", "squad-hard-negatives", split="train")
scores    = load_dataset("Hyukkyu/train-dpr-wikipedia", "squad-teacher-scores", split="train")

Load it

python

from datasets import load_dataset
queries = load_dataset("Hyukkyu/train-dpr-wikipedia", "nq-queries", split="train")
corpus  = load_dataset("Hyukkyu/train-dpr-wikipedia", "nq-corpus", split="train")
qrels   = load_dataset("Hyukkyu/train-dpr-wikipedia", "qrels", split="train")

License and attribution

The data is redistributed under the source's terms — cc-by-sa-3.0. All credit belongs to the original authors; see the source repository and the references in mteb's task metadata (https://github.com/facebookresearch/DPR). This repository is an independent repackaging and is not affiliated with the RTEB or MTEB maintainers.