datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
CareQA
CareQA
Dataset Summary
CareQA is a healthcare QA dataset with two versions:
Closed-Ended Version: A multichoice question answering (MCQA) dataset containing 5,621 QA pairs across six categories. Available in English and Spanish.
Open-Ended Version: A free-response dataset derived from the closed version, containing 2,769 QA pairs (English only).
The dataset originates from… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/CareQA.medical-specialities
Medical Question Classification Dataset
Dataset Summary
This dataset is designed for medical language models evaluation. It merges several of the most important medical QA datasets into a common format and classifies them into 35 distinct medical categories. This structure enables users to identify any specific categories where the model's performance may be lacking and address these areas accordingly.
Dataset Structure
Data Fields
id: Unique… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/medical-specialities.multi_lmentry
Multi-LMentry
This dataset card provides documentation for Multi-LMentry, a multilingual benchmark designed for evaluating large language models (LLMs) on fundamental, elementary-level tasks across nine languages. It is the official dataset release accompanying the EMNLP 2025 paper "Multi-LMentry: Can Multilingual LLMs Solve Elementary Tasks Across Languages?".
Dataset Details
Dataset Description
Multi-LMentry is a multilingual extension of LMentry (Efrat et… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/multi_lmentry.MedQA-Mixtral-CoT
Dataset Card for medqa-cot
Synthetically enhanced responses to the medqa dataset using mixtral.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the MedQA dataset, we leverage Mixtral-8x7B to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a
hand-crafted list of few-shot examples. For a multichoice answer, we ask the model to rephrase and explain the question… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/MedQA-Mixtral-CoT.headqa-cot-llama31
headqa-cot
Synthetically enhanced responses to the HeadQA dataset. Used to train Aloe-Beta model.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the HeadQA dataset, we leverage Llama-3.1-70B-Instruct to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a hand-crafted list of… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/headqa-cot-llama31.MMLU-medical-cot-llama31
MMLU-medical-cot
Synthetically enhanced responses to the medical-related questions of the auxiliary train set of the MMLU dataset. Used to train Aloe-Beta model.
Dataset Details
Dataset Description
First, we use Llama-3.1-70B-Instruct to filter the medical-related questions of the auxiliary train set of the MMLU dataset. Next, we leverage Mixtral-8x7B to… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/MMLU-medical-cot-llama31.EsBBQ
Spanish Bias Benchmark for Question Answering (EsBBQ)
The Spanish Bias Benchmark for Question Answering (EsBBQ) is an adaptation of the original BBQ to the Spanish language and the social context of Spain.
Dataset Description
This dataset is used to evaluate social bias in LLMs in a multiple-choice Question Answering (QA) setting and along 10 social categories: Age, Disability Status, Gender, LGBTQIA, Nationality, Physical Appearance, Race/Ethnicity, Religion… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/EsBBQ.CaBBQ
Catalan Bias Benchmark for Question Answering (CaBBQ)
The Catalan Bias Benchmark for Question Answering (CaBBQ) is an adaptation of the original BBQ to the Catalan language and the social context of Spain.
Dataset Description
This dataset is used to evaluate social bias in LLMs in a multiple-choice Question Answering (QA) setting and along 10 social categories: Age, Disability Status, Gender, LGBTQIA, Nationality, Physical Appearance, Race/Ethnicity, Religion… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/CaBBQ.BSC_ParaMT_8
Dataset Card for BSC_ParaMT_8
Dataset Summary
Large-scale multilingual parallel corpus covering Catalan, Spanish, and English paired with Arabic, Hindi, Chinese, Japanese, and Korean. Built by aggregating and carefully filtering multiple public sources, it provides sentence-level alignments for training Machine Translation systems. The Spanish portion of the dataset includes synthetic data generated by translating original English sentences into Spanish. Similarly… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/BSC_ParaMT_8.m-personas
mPersonas: Multilingual Persona‑Driven Conversational Dataset
Dataset Summary
mPersonas is a multilingual open-source dataset with high-quality persona descriptions synthetically generated by DeepSeek-V3–0324. It follows a persona-driven data synthesis methodology, similar to PersonaHub.
Instances: 510,000
Total tokens: 173M
28M in personas
145M in conversations (105M in assistant turns)
Languages: 15
License: Apache 2.0
Methodology
This section… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/m-personas.HEART
HEART
Dataset Summary
The HEART dataset is composed of multiple splits that differ in the type of injected cues. It includes a baseline split with no injected cues and four cue-based splits. Each cue is instantiated in two variants: assistive, where the cue is consistent with the GT, and adversarial,
where the cue supports an incorrect option. Cue types, summarized below… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/HEART.COPA-es
Dataset Card for COPA-es
COPA-es is a textual entailment dataset in Spanish, professionally translated from the COPA dataset in English. The dataset consists of 600 premises, each given a question and two choices with a label encoding which of the choices is more plausible given the annotator.
Dataset Details
Dataset Description
COPA-es (Choice of Plausible Alternatives - Spanish) is designed to simulate causal reasoning of text from commonsense subjects.… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/COPA-es.MedMCQA-Mixtral-CoT
Dataset Card for medmcqa-cot
Synthetically enhanced responses to the medmcqa dataset using mixtral.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the MedMCQA dataset, we leverage Mixtral-8x7B to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a
hand-crafted list of few-shot examples. For a multichoice answer, we ask the model to rephrase and explain the… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/MedMCQA-Mixtral-CoT.Egida
Dataset Card for Egida
Dataset Summary
Egida is an expanded collection of unsafe requests gathered from a variety of external sources. This dataset is boosted and extended (1) through a manual fine-grained topic classification, and (2) by applying a variety of jailbreaking attacks to all their samples.
Dataset Curation
Sources and data collection
In total, the dataset is composed of 2,949… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/Egida.openbookqa-es
Dataset Card for openbookqa_es
openbookqa_es is a question answering dataset in Spanish, professionally translated from the main version of the OpenBookQA dataset in English.
Dataset Details
Dataset Description
openbookqa_es (Open Book Question Answering - Spanish) is designed to simulate open book exams and assess human-like understanding of a subject. The dataset comprises 500 instances in the validation split and another 500 instances in the test split.… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/openbookqa-es.Aloe-Beta-Medical-Collection
Aloe-Beta-Medical-Collection
Collection of curated datasets used to fine-tune Aloe-Beta.
Dataset Details
Dataset Description
We curated data from many publicly available medical instruction tuning data sources (QA format). Most data samples correspond to single-turn QA pairs, while a small proportion contain multi-turn. All data sources are publicly available for… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/Aloe-Beta-Medical-Collection.medqa-cot-llama31
medqa-cot-llama31
Synthetically enhanced responses to the MedQa dataset. Used to train Aloe-Beta model.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the MedQA dataset, we leverage Llama-3.1-70B-Instruct to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a hand-crafted… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/medqa-cot-llama31.ALIA_mixed_authentic_synthetic_MT
Dataset Card for ALIA_mixed_authentic_synthetic_MT
Dataset Summary
Large-scale multilingual parallel corpus covering English and Spanish paired with Arabic, Hindi, Chinese, Japanese, and Korean. Built by aggregating and carefully filtering multiple public sources, it provides sentence-level alignments for training Machine Translation systems. The Spanish–Hindi and Spanish–Chinese portions of the dataset include synthetic Spanish translations generated from English using… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/ALIA_mixed_authentic_synthetic_MT.MedS-Ins
HPAI-BSC MedS-Ins
Collection of curated data from the MedS-Ins dataset. Used to train Aloe-Beta model.
Dataset Details
Dataset Description
This is the curated version of the MedS-Ins dataset included in the training set of the Aloe-Beta models.
First, we selected 75 out of the 122 existing tasks, excluding the tasks that were already in the training set, and the… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/MedS-Ins.Aloe-Beta-General-Collection
Aloe-Beta-Medical-Collection
Collection of curated general datasets used to fine-tune Aloe-Beta.
Dataset Details
Dataset Description
We curated data from many publicly available general instruction tuning data sources (QA format). It consists of 400k instructions including:
Coding, math, data analysis, STEM, etc.
Function calling
Creative writing, advice seeking… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/Aloe-Beta-General-Collection.medmcqa-cot-llama31
medqa-cot-llama31
Synthetically enhanced responses to the MedMCQA dataset. Used to train Aloe-Beta model.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the MedMCQA dataset, we leverage Llama-3.1-70B-Instruct to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/medmcqa-cot-llama31.salamandra-guard-dataset
Salamandra Guard Dataset
Dataset Description
The Salamandra Guard dataset is a comprehensive multilingual safety classification corpus designed for training and evaluating content moderation systems in Catalan, Spanish. It consists of 21,335 carefully curated conversational examples annotated across a hierarchical safety taxonomy.
This dataset represents a significant advancement in culturally-grounded safety data, with particular emphasis on Catalan—a language… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/salamandra-guard-dataset.Aloe-Beta-DPO
Aloe-Beta-Medical-Collection
Collection of curated DPO datasets used to align Aloe-Beta.
Dataset Details
Dataset Description
The first stage of the Aloe-Beta alignment process. We curated data from many publicly available data sources, including three different types of data:
Medical preference data: TsinghuaC3I/UltraMedical-Preference
General preference data:… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/Aloe-Beta-DPO.IFEval_es
Dataset Card for IFEval_es
IFEval_es is a prompt dataset in Spanish, professionally translated from the main version of the IFEval dataset in English.
Dataset Details
Dataset Description
IFEval_es (Instruction-Following Eval benchmark - Spanish) is designed to evaluating chat or instruction fine-tuned language models. The dataset comprises 541 "verifiable instructions" such as "write in more than 400 words" and "mention the keyword of AI at least 3 times"… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/IFEval_es.EQ-bench_es
Dataset Card for EQ Bench Dataset (Spanish Version)
This dataset card documents the Spanish adaptation of the EQ-Bench benchmark. The original dataset was designed to evaluate emotional reasoning in language models through dialogue-based prompts.
Dataset Details
Dataset Description
EQ-Bench (Spanish Version) is a translated and linguistically adapted version of the original EQ-Bench dataset.
Its design responds to the need to adapt the emotional detection… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/EQ-bench_es.PubmedQA-Mixtral-CoT
Dataset Card for pubmedqa-cot
Synthetically enhanced responses to the pubmedqa dataset using mixtral.
Dataset Details
Dataset Description
To increase the quality of answers from the training splits of the PubMedQA dataset, we leverage Mixtral-8x7B to generate Chain of Thought(CoT) answers. We create a custom prompt for the dataset, along with a
hand-crafted list of few-shot examples. For a multichoice answer, we ask the model to rephrase and explain the… See the full description on the dataset page: https://huggingface.co/datasets/HPAI-BSC/PubmedQA-Mixtral-CoT.EQ-bench_ca
Dataset Card for EQ Bench Dataset (Catalan Version)
This dataset card documents the Catalan adaptation of the EQ-Bench benchmark. The original dataset was designed to evaluate emotional reasoning in language models through dialogue-based prompts.
Dataset Details
Dataset Description
EQ-Bench (Catalan Version) is a translated and linguistically adapted version of the original EQ-Bench dataset.
Its design responds to the need to adapt the emotional detection… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/EQ-bench_ca.arc_es
Dataset Card for ARC (Spanish Version)
Dataset summary
This dataset provides the Spanish translation and adaptation of the ARC (AI2 Reasoning Challenge) validation set. The original dataset was designed to evaluate scientific reasoning in LLMs by presenting a collection of authentic, grade-school science multiple choice questions split into two sets of varying difficulty: ARC Easy and ARC Challenging.
This Spanish adaptation enables evaluation of the ARC task in… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/arc_es.Catalan-Aranese_Parallel_Corpus
Dataset Card for Catalan-Aranese Parallel Corpus
Dataset Summary
A bilingual parallel corpus for the low-resource language pair Catalan-Aranese. Built by aggregating and filtering multiple public sources, along with data obtained through direct data sharing with external partners, it provides sentence-level alignments for training Machine Translation systems. The dataset includes both authentically parallel data as well as synthetic Catalan translations generated from… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/Catalan-Aranese_Parallel_Corpus.LexBOE
Dataset Card for LexBOE
Dataset summary
LexBOE is a Spanish legal text classification dataset built from articles extracted from the Boletín Oficial del Estado (BOE), the official source of legislation and administrative acts in Spain. The articles included in the dataset were published between 2022 and 2024.
LexBOE reflects contemporary legal-administrative language and is intended for the training and evaluation of language models on legal text classification tasks.… See the full description on the dataset page: https://huggingface.co/datasets/BSC-LT/LexBOE.