datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
in1k_clip_qwen25vl_3b_224res_64tokens_new_ptin1k_clip_qwen25vl_3b_448res_256tokens_new_merged_ptMT-intents-dataset-pt-PTSentiments-FinBERT-PT-BR
Dataset
A manually annotated dataset was created to enable supervised training for the FinBERT-PT-BR model, which focuses on sentiment analysis of Brazilian Portuguese financial texts.
More than 1.4 million financial news texts in Portuguese were collected and used for the initial language modeling phase. From this corpus, a sample of 1,000 texts was manually annotated with sentiment labels.
Annotation Process
Three annotators participated in the process.
All texts were… See the full description on the dataset page: https://huggingface.co/datasets/lucas-leme/Sentiments-FinBERT-PT-BR.harmful-prompts-pt
Harmful Prompts PT-BR
harmful-prompts-pt is a Brazilian Portuguese adaptation of the
WildJailbreak dataset,
constructed to support research on the robustness of language models against
harmful and adversarial prompts in Portuguese.
This dataset was used to train and evaluate SecBERT, a Portuguese harmful
prompt classifier presented at the International Joint Conference on Neural
Networks (IJCNN). The full paper and source code are available at
[Paper] [Code].
Caution: This dataset… See the full description on the dataset page: https://huggingface.co/datasets/Edu-p/harmful-prompts-pt.SQuAD-pt_BR-V1.1_
Dataset Card para o SQuAD 1.1 em Português Brasil
O conjunto de dados "Stanford Question Answering Dataset" (SQuAD),
para tarefa de perguntas e respostas extrativas, foi desenvolvido em 2016. Ele utiliza perguntas geradas a partir de
536 artigos da Wikipedia* com mais de 100.000 linhas de dados. É construído na forma de uma pergunta e um contexto dos artigos da
Wikipedia contendo a resposta à pergunta. [1]Originalmente este dataset foi construído no idioma inglês, contudo, o grupo… See the full description on the dataset page: https://huggingface.co/datasets/vsvasconcelos/SQuAD-pt_BR-V1.1_.toxic_classificationcombination of SetFit/toxic_conversations_50k, Arsive/toxicity_classification_jigsaw
qa-pt
Dataset Card for QA-Portuguese
Dataset Summary
Portuguese preprocessed split from MQA dataset.
Supported Tasks and Leaderboards
[More Information Needed]
Languages
The dataset is Portuguese.
Dataset Structure
Data Instances
[More Information Needed]
Data Fields
[More Information Needed]
Data Splits
[More Information Needed]
Dataset Creation
Curation Rationale
[More Information Needed]… See the full description on the dataset page: https://huggingface.co/datasets/ju-resplande/qa-pt.SQuAD-pt_BR-V1.1
Dataset Card for Dataset Name
O conjunto de dados "Stanford Question Answering Dataset" (SQuAD), para tarefa de perguntas e respostas extrativas, foi desenvolvido em 2016. Ele utiliza perguntas geradas
a partir de 536 artigos da Wikipedia com mais de 100.000 linhas de dados. É construído na forma de uma pergunta e um contexto dos artigos da Wikipedia contendo a resposta
à pergunta.
Originalmente este dataset foi construído no idioma inglês, contudo, o grupo Deep Learning Brasil… See the full description on the dataset page: https://huggingface.co/datasets/vsvasconcelos/SQuAD-pt_BR-V1.1.pt-recognition
PT Interaction Dataset
Dataset Description
The PT (Peptide-TCR) interaction dataset is designed for training and evaluating T-Cell Receptor (TCR) binding prediction models with full TCR sequence information. This dataset contains paired peptide sequences and complete TCR alpha/beta chain sequences (including all 6 CDR regions: A1-A3, B1-B3), along with binary binding labels.
Key Features
Full TCR Information: Contains all 6 CDR regions (A1, A2, A3, B1, B2, B3)… See the full description on the dataset page: https://huggingface.co/datasets/YYJMAY/pt-recognition.banking77-pt-br
Resumo do conjunto de dados
Tradução revisada para o português brasileiro do conjunto de dados BANKING77. O conjunto é composto por consultas online realizadas a sistemas conversacionais de bancos, rotuladas de acordo com suas intenções correspondentes. De acordo com a documentação original, as 13.083 consultas que compõem o conjunto são referentes a atendimentos ao cliente, e foram rotuladas em 77 intenções diferentes. O foco principal é suportar análises em domínio específico e… See the full description on the dataset page: https://huggingface.co/datasets/c2d-usp/banking77-pt-br.JBB-Behaviors-pt
JBB-Behaviors-pt
Dataset Description
JBB-Behaviors-pt is a dataset of behaviors in Portuguese, including both jailbreak prompts and safe behaviors. The dataset is intended for behavioral testing of language models to evaluate their robustness against jailbreak attempts in Portuguese.
What is a jailbreak?
Jailbreak prompts are inputs designed to bypass a language model's safety guardrails, potentially causing it to generate harmful, unethical, or otherwise… See the full description on the dataset page: https://huggingface.co/datasets/tech4humans/JBB-Behaviors-pt.humaneval_mt_pt
HumanEval-PT
Portuguese version of HumanEval, a code generation benchmark with programming problems and test cases.
Translated using NLLB-200.
Original Dataset: https://huggingface.co/datasets/openai/openai_humaneval
Note: This dataset is machine translated and may contain translation errors or artifacts.
This dataset is provided as part of the AMALIA project and is included in AMALIA-Bench, a comprehensive benchmark suite for evaluating large language models on… See the full description on the dataset page: https://huggingface.co/datasets/amalia-llm/humaneval_mt_pt.pt-parliament-interventionsTechnicalDebt_GitHubIssues_PT
Technical Debt in GitHub Issues (Portuguese)
Visão Geral
Este dataset reúne issues públicas extraídas do GitHub que mencionam o termo "dívida técnica" ou suas variações em português. A base foi construída com o objetivo de apoiar pesquisas em Engenharia de Software, Processamento de Linguagem Natural (PLN) e Inteligência Artificial, com foco na compreensão e classificação de como o conceito de dívida técnica é comunicado por desenvolvedores.
A estrutura e categorização… See the full description on the dataset page: https://huggingface.co/datasets/IuryCavalcante/TechnicalDebt_GitHubIssues_PT.XStoryCloze_ptARC-Challenge_PT-PT
Dataset Card for ARC-Challenge_PT-PT
Dataset Summary
This repository contains a European Portuguese (pt-PT) translation of ARC-Challenge (AI2 Reasoning Challenge – Challenge subset), a benchmark of non-trivial, grade-school science questions that require background knowledge and reasoning. Each example presents a question and four answer choices.Load with:
import datasets
data = datasets.load_dataset("ruibrogandrade/ARC-Challenge_PT-PT")
Supported Tasks and… See the full description on the dataset page: https://huggingface.co/datasets/ruibrogandrade/ARC-Challenge_PT-PT.sst_pt-pt
Simple Safety Tests-PT
Portuguese machine translation of Simple Safety Tests, a benchmark for evaluating model safety and harmful content detection.
Translated using a Finetuned GemmaX2-9B for pt-PT.
Original Dataset: https://huggingface.co/datasets/Bertievidgen/SimpleSafetyTests
Note: This dataset is machine translated and may contain translation errors or artifacts.
This dataset is provided as part of the AMALIA project and is included in AMALIA-Bench, a… See the full description on the dataset page: https://huggingface.co/datasets/amalia-llm/sst_pt-pt.santfordimdb-ptHellaSwag_PT-PT
Dataset Card for HellaSwag_PT-PT
Dataset Summary
This repository provides a European Portuguese (pt-PT) translation of HellaSwag, an adversarial benchmark for grounded commonsense inference. Given a short scenario, the model must choose the most plausible continuation from four options.Load with:
import datasets
data = datasets.load_dataset("ruibrogandrade/HellaSwag_PT-PT")
Supported Tasks and Leaderboards
Multiple-Choice Question Answering (commonsense… See the full description on the dataset page: https://huggingface.co/datasets/ruibrogandrade/HellaSwag_PT-PT.Squad_PT
Dataset Card para o SQuAD 1.1 em Português Brasil
O conjunto de dados "Stanford Question Answering Dataset" (SQuAD),
para tarefa de perguntas e respostas extrativas, foi desenvolvido em 2016. Ele utiliza perguntas geradas a partir de
536 artigos da Wikipedia* com mais de 100.000 linhas de dados. É construído na forma de uma pergunta e um contexto dos artigos da
Wikipedia contendo a resposta à pergunta. [1]Originalmente este dataset foi construído no idioma inglês, contudo, o grupo… See the full description on the dataset page: https://huggingface.co/datasets/vitorandrade/Squad_PT.gemma-3-1b-pt-blind-spots
Gemma-3-1b-pt Blind Spots Dataset
Dataset Description
This dataset documents blind spots (systematic errors) found in
google/gemma-3-1b-pt,
a 1-billion-parameter pretrained base model (not instruction-tuned)
released by Google in March 2025 as part of the Gemma 3 family.
Each row contains:
Column
Description
id
Unique probe index
category
Type of reasoning tested
prompt
The input fed to the model (text-completion style)
expected_output
The… See the full description on the dataset page: https://huggingface.co/datasets/Junaid687/gemma-3-1b-pt-blind-spots.pt-all-words
Dataset Card for Dicionário Português
It is a list of portuguese words with its inflections
How to use it:
from datasets import load_dataset
remote_dataset = load_dataset("VanessaSchenkel/pt-all-words")
remote_dataset
toxic_classification_balancedcombination of SetFit/toxic_conversations_50k, Arsive/toxicity_classification_jigsaw
taken sample from toxic classification - to balance the dataset
PAWS_ptcstnews-pttwitter-sentiment-pt-BR-md-2-lSocial_I_QA_PT-PT
Dataset Card for Social_I_QA_PT-PT
Dataset Summary
This repository contains a European Portuguese (pt-PT) translation of Social IQa, a benchmark for social commonsense reasoning. Each example includes a short context about a social situation, a question, and three answer candidates. The task is to choose the most plausible answer about intents, reactions, or social outcomes.Load with:
import datasets
data = datasets.load_dataset("ruibrogandrade/Social_I_QA_PT-PT")… See the full description on the dataset page: https://huggingface.co/datasets/ruibrogandrade/Social_I_QA_PT-PT.imdb-pt-brgemma-3-1b-pt-blind-spots
Blind Spots of google/gemma-3-1b-pt
Model Tested
Model: google/gemma-3-1b-ptParameters: 1BType: Pre-trained base language model (not instruction-tuned)Tested by: Toka-Tarek | Biotechnology graduate & Pharmacogenetics Lab Specialist
How I Loaded the Model
Tested on Google Colab (free T4 GPU, 16GB VRAM).
Note: torch.float16 caused numerical instability (NaN/inf errors)
on the T4 GPU, so torch.float32 was used instead for stable generation.
from huggingface_hub… See the full description on the dataset page: https://huggingface.co/datasets/Toka-Tarek/gemma-3-1b-pt-blind-spots.