datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
FinQA-hallucination-detection
FinQA Hallucination Detection
Dataset Summary
This dataset was created from a subset of the original FinQA dataset. For each user query (financial questions), we prompted an LLM to generate a response to this query based on provided context (financial statements and tables from the original FinQA).
Each generated LLM response is labeled based on whether it is correct or not. This dataset is thus useful for benchmarking reference-free LLM Eval and Hallucination… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/FinQA-hallucination-detection.insurance-claims-extractionThis dataset can be used for benchmarking LLM Structured Outputs via the code here:
https://github.com/cleanlab/structured-output-benchmark/
stanford-politeness
Stanford Politeness Dataset
This dataset contains politeness classification data based on the Stanford Politeness Corpus for active learning and fine-tuning tasks.
Dataset Description
The dataset is organized into two main directories:
Active Learning
X_labeled_full.csv - Labeled examples
X_unlabeled.csv - Unlabeled examples for active learning
extra_annotations.npy - Additional annotation data
test.csv - Test set
Fine-tuning
train.csv - Training… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/stanford-politeness.fire-financial-ner-extractionThis dataset can be used for benchmarking LLM Structured Outputs via the code here:
https://github.com/cleanlab/structured-output-benchmark/
amazon-reviews
Amazon Reviews Dataset
This dataset contains Amazon product reviews with binary sentiment labels (positive, negative) for text classification tasks.
Dataset Description
The dataset includes:
train.csv - Training set, 5000 samples
test.csv - Test set, 1000 samples
Usage
import pandas as pd
from huggingface_hub import hf_hub_download
# Download the training set
file_path = hf_hub_download(
repo_id="Cleanlab/amazon-reviews",
filename="train.csv"… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/amazon-reviews.pii-extractionThis dataset can be used for benchmarking LLM Structured Outputs via the code here:
https://github.com/cleanlab/structured-output-benchmark/
object-detection-tutorial
Object Detection Tutorial Dataset
Dataset Description
This dataset contains object detection annotations and predictions used in the cleanlab tutorial: Object Detection.
The dataset demonstrates how to use cleanlab to identify and correct label issues in object detection datasets, where labels consist of bounding boxes around objects in images.
Dataset Summary
Total Examples: 118 images with bounding box annotations
Task: Object detection with bounding boxes… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/object-detection-tutorial.banking-intent-50
Banking Intent 50 Dataset
This dataset contains banking intent classification data with multiple CSV files for different use cases.
Dataset Description
The dataset includes:
examples-pool.csv - Pool of examples for labeling
examples-pool-ground-truth.csv - Ground truth labels
test.csv - Test set
Usage
import pandas as pd
from huggingface_hub import hf_hub_download
# Download a specific file
file_path = hf_hub_download(… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/banking-intent-50.student-grades-demo
Student Grades Demo Dataset
This dataset contains student grades data with both true labels and noisy (corrupted) labels.
Dataset Description
The dataset includes:
Student exam scores (exam_1, exam_2, exam_3)
Notes field
True letter grades (letter_grade)
Noisy/corrupted letter grades (noisy_letter_grade)
This is useful for demonstrating and validating label error detection methods.
Usage
import pandas as pd
# Load the dataset
df =… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/student-grades-demo.segmentation-tutorial
Segmentation Tutorial Dataset
Dataset Description
This dataset contains image segmentation masks used in the cleanlab tutorial: Image Segmentation.
The dataset demonstrates how to use cleanlab to identify and correct label issues in semantic segmentation datasets, where each pixel in an image is assigned a class label.
Dataset Summary
Total Examples: 30 images with segmentation masks
Task: Semantic image segmentation
Image Size: 1088 x 1920 pixels
Files:… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/segmentation-tutorial.bad_data_gsm8k_svamp.csvSome bad data discovered in the popular GSM8K and SVAMP LLM benchmarking datasets.
These examples have incorrect answers in the corresponding math problem benchmark dataset, and should not be used to evaluate AI models.
We detected this bad data automatically using Cleanlab's Trustworthy Language Model. TLM's estimated trustworthiness score for each example is also provided.
Example error found in the GSM8K dataset:
Question: After scoring 14 points, Erin now has three times… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/bad_data_gsm8k_svamp.csv.banking-intent-classification
Banking Intent Classification Dataset
This dataset contains text samples for banking intent classification tasks.
Dataset Description
The dataset consists of customer queries/messages related to banking services, each labeled with an intent category.
Usage
import pandas as pd
# Load the dataset
df = pd.read_csv("hf://datasets/Cleanlab/banking-intent-classification/banking-intent-classification.csv")
print(df.head())
License
MIT License
token-classification-tutorial
Token Classification Tutorial Dataset
Dataset Description
This dataset contains predicted probabilities for token classification used in the cleanlab tutorial: Token Classification.
The dataset demonstrates how to use cleanlab to identify and correct label issues in token classification datasets, such as Named Entity Recognition (NER) tasks where each token in a sequence is assigned a class label.
Dataset Summary
Task: Token classification / Named Entity… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/token-classification-tutorial.databricks-dolly-15k-cleaned
Summary
databricks-dolly-15k-cleaned is a cleaned up version of the popular databricks-dolly-15k dataset after automatically removing low quality datapoints detected using Cleanlab.
student-grades
Student Grades Dataset
Dataset Description
This dataset contains student grade data used in the cleanlab tutorial: Improving ML Performance via Data Curation with Train vs Test Splits.
The task is to predict each student's final letter grade (A, B, C, D, F) based on their exam scores and notes.
Dataset Summary
Total Examples: ~750 (train + test)
Task: Multi-class classification
Features:
exam_1: Score on first exam (0-100)
exam_2: Score on second exam (0-100)… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/student-grades.cleanlab-label_errorsdata-table-analysisThis dataset can be used for benchmarking LLM Structured Outputs via the code here:
https://github.com/cleanlab/structured-output-benchmark/
cifar-10-subset
CIFAR-10 Subset
This dataset contains a subset of the CIFAR-10 image classification dataset.
Dataset Description
A curated subset of the CIFAR-10 dataset useful for demonstrating image classification and data quality techniques without requiring the full dataset.
Usage
# Download the dataset
wget https://huggingface.co/datasets/Cleanlab/cifar-10-subset/resolve/main/CIFAR-10-subset.zip
unzip CIFAR-10-subset.zip
from huggingface_hub import hf_hub_download
#… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/cifar-10-subset.databricks-dolly-15k-cleanset
Summary
databricks-dolly-15k-cleanset can be used to produced CLEANed up versions of the popular databricks-dolly-15k dataSET, which was used to fine-tune the Dolly 2.0. The original databricks-dolly-15k contains 15,000 human-annotated instruction-response pairs covering various categories. However, there are many low-quality responses, incomplete/vague prompts, and other problematic text lurking in the dataset (as with for all real-world instruction tuning datasets). We ran… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/databricks-dolly-15k-cleanset.legal-ediscovery
Legal eDiscovery Dataset
This dataset contains text data for legal eDiscovery classification tasks.
Dataset Description
The dataset includes:
legal-ediscovery.csv - Main dataset
legal-ediscovery-full.csv - Full version of the dataset
Usage
import pandas as pd
from huggingface_hub import hf_hub_download
# Download the dataset
file_path = hf_hub_download(
repo_id="Cleanlab/legal-ediscovery",
filename="legal-ediscovery.csv",
repo_type="dataset"
)
df… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/legal-ediscovery.stanford-politeness-prompt-selection
Stanford Politeness Prompt Selection Dataset
This dataset contains data for prompt selection tasks based on the Stanford Politeness Corpus.
Dataset Description
The dataset is designed for evaluating and selecting prompts for politeness classification tasks.
Usage
import pandas as pd
# Load the dataset
df = pd.read_csv("hf://datasets/Cleanlab/stanford-politeness-prompt-selection/stanford-politeness-prompt-selection.csv")
print(df.head())
License
MIT… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/stanford-politeness-prompt-selection.student-grades-regression
Student Grades Regression Dataset
This dataset contains student grades data for regression tasks. It is designed for demonstrating regression techniques and data quality methods.
Dataset Description
The dataset includes:
train.csv - Training data with student features and grade labels
test.csv - Test data for evaluation
This is a regression version of the student grades dataset, where the target variable is continuous.
Usage
import pandas as pd
# Load the… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/student-grades-regression.multilabel-datasets
Multilabel Datasets
This dataset contains multilabel classification data for student grades.
Dataset Description
The dataset includes:
student-grades-multilabel.csv - Student grades with multilabel annotations
student-grades-multilabel-v2.csv - Version 2 of the multilabel dataset
student-grades-one-hot-encoded.csv - One-hot encoded version
Usage
import pandas as pd
from huggingface_hub import hf_hub_download
# Download a specific file
file_path =… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/multilabel-datasets.grades-tabular-demo-with-outliers
Grades Tabular Demo with Outliers Dataset
This dataset contains student grades data for tabular classification tasks, including some outliers to demonstrate data quality techniques.
Dataset Description
A tabular dataset containing student information and their grades, useful for demonstrating data quality techniques on tabular data.
Usage
import pandas as pd
# Load the dataset
df =… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/grades-tabular-demo-with-outliers.grades-tabular-demo
Grades Tabular Demo Dataset
This dataset contains student grades data for tabular classification tasks.
Dataset Description
A tabular dataset containing student information and their grades, useful for demonstrating data quality techniques on tabular data.
Usage
import pandas as pd
# Load the dataset
df = pd.read_csv("hf://datasets/Cleanlab/grades-tabular-demo/grades-tabular-demo.csv")
print(df.head())
License
MIT License
cleanvision-sample-images
CleanVision Sample Images
This dataset contains sample images for demonstrating CleanVision, a Python package for automatically detecting issues in image datasets.
Dataset Description
This collection includes various images with different quality issues such as:
Blurry images
Dark/underexposed images
Overexposed images
Near-duplicate images
Images with various other quality issues
Usage
Download the dataset:
wget… See the full description on the dataset page: https://huggingface.co/datasets/Cleanlab/cleanvision-sample-images.agnews-cleanlab-cleanedmnist
MNIST Dataset
This dataset contains MNIST handwritten digit images.
Dataset Description
The classic MNIST dataset containing images of handwritten digits (0-9) for image classification tasks.
Usage
# Download the dataset
wget https://huggingface.co/datasets/Cleanlab/mnist/resolve/main/mnist.tar.gz
tar -xzf mnist.tar.gz
License
MIT License
clean_labeled_datasetfootwear-demo
Footwear Demo Dataset
This dataset contains footwear images for image classification tasks.
Dataset Description
A collection of footwear images for demonstrating image classification and data quality techniques.
Usage
# Download the dataset
wget https://huggingface.co/datasets/Cleanlab/footwear-demo/resolve/main/footwear-demo.zip
unzip footwear-demo.zip
License
MIT License