Hyukkyu/rteb-open

RTEB Open — unified schema

The 15 open retrieval datasets listed on the RTEB leaderboard, each republished as its own repo in one strict, identical schema, with a provenance.json that ties it to the exact source revision mteb evaluates. This repo is just the index; the data lives in the linked repos.

Two further datasets — SWEbenchCodeRetrieval and MIRACLRetrievalHardNegatives — belong to mteb's RTEB(beta) benchmark definition but are not on the RTEB leaderboard above. They were built and verified with the same pipeline (manifest.json records them) but are not published, so they are not in the table.

task	repo	languages	domain	license	queries	documents	qrels	= mteb
AILACasedocs	rteb-AILACasedocs	eng	legal	cc-by-4.0	50	186	195	✓
AILAStatutes	rteb-AILAStatutes	eng	legal	cc-by-4.0	50	82	217	✓
AppsRetrieval	rteb-AppsRetrieval	eng	code	mit	3,765	8,765	3,765	✓
ChatDoctorRetrieval	rteb-ChatDoctorRetrieval	eng	healthcare	unspecified	5,591	5,545	5,591	✓
CUREv1	rteb-CUREv1	eng, spa, fra	healthcare	cc-by-nc-4.0	6,000	269,271	80,716	✓
DS1000Retrieval	rteb-DS1000Retrieval	eng	code	cc-by-sa-4.0	1,998	1,998	1,998	✓
FinanceBenchRetrieval	rteb-FinanceBenchRetrieval	eng	finance	apache-2.0	150	145	150	✓
FinQARetrieval	rteb-FinQARetrieval	eng	finance	cc-by-4.0	1,138	380	1,138	✓
FreshStackRetrieval	rteb-FreshStackRetrieval	eng	code	mit	672	3,804	5,445	✓
HC3FinanceRetrieval	rteb-HC3FinanceRetrieval	eng	finance	cc-by-sa-4.0	415	415	415	✓
HumanEvalRetrieval	rteb-HumanEvalRetrieval	eng	code	mit	158	158	158	✓
LegalQuAD	rteb-LegalQuAD	deu	legal	cc-by-4.0	200	200	200	✓
LegalSummarization	rteb-LegalSummarization	eng	legal	apache-2.0	284	438	439	✓
MBPPRetrieval	rteb-MBPPRetrieval	eng	code	cc-by-4.0	974	974	974	✓
WikiSQLRetrieval	rteb-WikiSQLRetrieval	eng	code	bsd-3-clause	2,048	2,048	2,048	✓

"= mteb" ✓ means: running mteb on the task as shipped and on this build, with a deterministic encoder, produced identical metrics to 1e-9 — the data is the data the benchmark scores.

The schema (every repo)

config	columns	split
queries	id: string, text: string	test
corpus	id: string, title: string, text: string	test
qrels	query-id: string, corpus-id: string, score: int32	test

Multilingual repos prefix the configs (ar-queries, ar-corpus, ar-qrels, …). CUREv1 keeps its ten medical-specialty splits and shares one corpus/qrels across en-/es-/fr-queries. Rules: ids unique and non-empty; title always present ("" if none); every query has ≥ 1 qrel; no duplicate pairs; referential integrity — relaxed only for MIRACL, whose source judgements reference documents outside its hard-negative pool (kept, counted, documented).

Use with mteb

import mteb, types
from mteb.abstasks.retrieval import AbsTaskRetrieval

task = mteb.get_task("HumanEvalRetrieval")
task.metadata = task.metadata.model_copy(update={"dataset": {"path": "Hyukkyu/rteb-HumanEvalRetrieval", "revision": "main"}})
# nine RTEB tasks ship a custom load_data hardcoded to their source's split names; the generic loader reads this layout
if type(task).load_data is not AbsTaskRetrieval.load_data:
    task.load_data = types.MethodType(AbsTaskRetrieval.load_data, task)
mteb.evaluate(mteb.get_model("intfloat/multilingual-e5-small"), [task])

Provenance and licensing

Each repo's provenance.json records the source repo and revision, the sha256 of every source file, every change made (renames, casts, dropped columns, orphan/duplicate counts), and the sha256 of every output file. Licenses are per dataset and follow the upstream card where it disagrees with mteb's metadata; the index carries none of its own. Not affiliated with the RTEB or MTEB maintainers. Built 2026-09-08 from mteb main.