datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hotpotqa
HotpotQA
An MTEB dataset
Massive Text Embedding Benchmark
HotpotQA is a question answering dataset featuring natural, multi-hop questions, with strong supervision for supporting facts to enable more explainable question answering systems.
Task category
t2t
Domains
Web, Written
Reference
https://hotpotqa.github.io/
How to evaluate on this task
You can evaluate an embedding model on this dataset using the following code:
import mteb
task =… See the full description on the dataset page: https://huggingface.co/datasets/mteb/hotpotqa.hotpotqa-fr-abstracts
HotpotQA-fr (abstracts)
65 565 questions multi-sauts en français, au format de
HotpotQA (Yang et al., 2018). Les questions sont
construites directement sur Wikipédia français, sans traduction. L'annotation
humaine du jeu original est remplacée par une génération par modèle de langue
suivie d'une validation automatique par ablations.
Chaque contexte contient dix paragraphes : les deux abstracts nécessaires à la
réponse et huit distracteurs. Tous les paragraphes sont des abstracts… See the full description on the dataset page: https://huggingface.co/datasets/Mvanypersele/hotpotqa-fr-abstracts.hotpotqa-generated-queries
Dataset Card for BEIR Benchmark
Dataset Summary
BEIR is a heterogeneous benchmark that has been built from 18 diverse datasets representing 9 information retrieval tasks:
Fact-checking: FEVER, Climate-FEVER, SciFact
Question-Answering: NQ, HotpotQA, FiQA-2018
Bio-Medical IR: TREC-COVID, BioASQ, NFCorpus
News Retrieval: TREC-NEWS, Robust04
Argument Retrieval: Touche-2020, ArguAna
Duplicate Question Retrieval: Quora, CqaDupstack
Citation-Prediction: SCIDOCS
Tweet… See the full description on the dataset page: https://huggingface.co/datasets/BeIR/hotpotqa-generated-queries.wiki_hotpotqaHotpotQA_FA_test_top_250_only_w_correct-v2beir-nl-hotpotqa
Dataset Card for BEIR-NL Benchmark
Dataset Summary
BEIR-NL is a Dutch-translated version of the BEIR benchmark, a diverse and heterogeneous collection of datasets covering various domains from biomedical and financial texts to general web content. Our benchmark is integrated into the Massive Multilingual Text Embedding Benchmark (MMTEB).
BEIR-NL contains the following tasks:
Fact-checking: FEVER, Climate-FEVER, SciFact
Question-Answering: NQ, HotpotQA, FiQA-2018… See the full description on the dataset page: https://huggingface.co/datasets/clips/beir-nl-hotpotqa.KBevo-SFT-hotpotqa-6k
KBevo-SFT-hotpotqa-6k
Supervised fine-tuning (SFT) trajectories for the KBevo two-phase policy, generated on HotpotQA. Accompanies Co-Evolving Structured Knowledge and Reasoning in Language Models (COLM 2026).
This is the exact SFT dataset used to produce kilian-group/KBevo-Qwen3-1.7B-SFT and kilian-group/KBevo-Qwen3-4B-SFT, which in turn initialise the KBevo-Qwen3-1.7B-GRPO and KBevo-Qwen3-4B-GRPO runs.
What's in the file
A single JSON file, trajectories.json… See the full description on the dataset page: https://huggingface.co/datasets/kilian-group/KBevo-SFT-hotpotqa-6k.hotpotqa_with_qa_gpt35
HotpotQA Dataset with GPT-3.5 Generated Questions
Overview
This repository hosts an enhanced version of the HotpotQA dataset, where each supporting sentence in the dataset has been supplemented with questions generated using OpenAI's GPT-3.5 turbo API. The aim is to provide a richer context for each entry, potentially benefiting various NLP tasks, such as question answering and context understanding.
Dataset Format
Each entry in the dataset is formatted as… See the full description on the dataset page: https://huggingface.co/datasets/scholarly-shadows-syndicate/hotpotqa_with_qa_gpt35.hotpotqa-fa
Dataset Summary
HotpotQA-Fa is a Persian (Farsi) dataset designed for the Retrieval task, specifically focused on multi-hop question answering. It is a translated version of the original English HotpotQA dataset and a key part of the FaMTEB (Farsi Massive Text Embedding Benchmark), under the BEIR-Fa collection.
Language(s): Persian (Farsi)
Task(s): Retrieval (Multi-hop Question Answering)
Source: Translated from the English HotpotQA dataset
Part of FaMTEB: Yes — under BEIR-Fa… See the full description on the dataset page: https://huggingface.co/datasets/MCINext/hotpotqa-fa.chem-hotpotqa
chem-hotpotqa
Chemistry subset of HotpotQA.
Corpus: 464
Queries: 274 (train 234, dev 14, test 26)
Qrels: 548 (train 468, dev 28, test 52)
hotpotqa-hard-neg-reasoning-embedding
Do Reasoning Models Enhance Embedding Models?
Introduction
This is the dataset used to train the embedding models in the paper Do Reasoning Models Enhance Embedding Models?. We use Qwen3-Embedding-0.6B to mine 3 hard negatives per query, and employ the positive-aware hard negative mining technique introduced in NV-Retriever with 95% margin to the positive score.
Abstract
State-of-the-art embedding models are… See the full description on the dataset page: https://huggingface.co/datasets/lucaswychan/hotpotqa-hard-neg-reasoning-embedding.hop-1-hotpotqa-decomposition
hop-1-hotpotqa-decomposition
What this is. For each question in this dataset, GPT-4o produced a decomposition of the first reasoning step needed to answer a multi-hop HotpotQA question: what to look up first, not the final answer. I used these labels to fine-tune Gemma-3-270M into a Hop-1 specialist, Omokemi/hop-1-gemma-270m.
Source. Questions come from HotpotQA (Yang et al., 2018), released under CC BY-SA 4.0 since it's built from Wikipedia text. The Hop-1 decomposition for… See the full description on the dataset page: https://huggingface.co/datasets/Omokemi/hop-1-hotpotqa-decomposition.hotpotqa-dev-raft-subsetFollows RAFT to generate question, documents, answer triplets
from the first 2010 512-token chunks of the HotPotQA dev set (fullwiki) with 2 questions per chunk and 3 distractor docs
and formatted into chat. Instead of GPT-4, meta-llama/Meta-Llama-3-70B-Instruct (hosted on a vLLM server) is used as the question generator.
Generating the 51.4k dataset (train+test) took ~27.5 hours.
hotpotqa-knowledge-coupling
Knowledge Coupling Analysis on HotpotQA Dataset
Dataset Description
This dataset contains the results of a comprehensive knowledge coupling analysis performed on the HotpotQA dataset using LLaMA2-7B model. The analysis investigates how different pieces of knowledge interact within the model's parameter space through gradient-based coupling measurements.
Research Overview
Model: meta-llama/Llama-2-7b-hf (layers 28-31 focused analysis)
Dataset: HotpotQA (train +… See the full description on the dataset page: https://huggingface.co/datasets/Wuhuwill/hotpotqa-knowledge-coupling.hotpot-qa-kg-processedTruthRL-HotpotQAhotpotqahotpotqa-top-20-gen-queries
NFCorpus: 20 generated queries (BEIR Benchmark)
This HF dataset contains the top-20 synthetic queries generated for each passage in the above BEIR benchmark dataset.
DocT5query model used: BeIR/query-gen-msmarco-t5-base-v1
id (str): unique document id in NFCorpus in the BEIR benchmark (corpus.jsonl).
Questions generated: 20
Code used for generation: evaluate_anserini_docT5query_parallel.py
Below contains the old dataset card for the BEIR benchmark.
Dataset Card for BEIR… See the full description on the dataset page: https://huggingface.co/datasets/income/hotpotqa-top-20-gen-queries.hotpotqa-dev-raft-subset-completionFollows RAFT to generate question, documents, answer triplets
from the first 110 512-token chunks of the HotPotQA dev set (fullwiki) with 2 questions per chunk and 3 distractor docs
and formatted into completion.
hotpotqa_devhotpotqa-distractor-embed-bge-large-v3hotpotqa-instructiondpo_hotpotqa_workforcehotpot-qa-contamination-scoresTruthRL-HotpotQAgpl-hotpotqaHotPotQA-jsonhotpotqa-hard-negatives
Dataset Card
Dataset Details
This dataset contains a set of candidate documents for second-stage re-ranking on hotpotqa
(test split in BEIR). Those candidate documents are composed of hard negatives mined from
gtr-t5-xl as Stage 1 ranker
and ground-truth documents that are known to be relevant to the query. This is a release from our paper
Policy-Gradient Training of Language Models for Ranking, so
please cite it if using this dataset.
Direct Use
You… See the full description on the dataset page: https://huggingface.co/datasets/NeuralPGRank/hotpotqa-hard-negatives.benchmark-hotpotqaqwen2.5-3b-sft-grpo-hotpotqa-dataset