datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
da-code-evaluation-resultsinaturalist
Dataset Description
The iNaturalist dataset is a large-scale species classification dataset for fine-grained recognition. This split is derived from the OpenOOD benchmark OOD evaluation splits.
Homepage: https://github.com/visipedia/inat_comp
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation
@inproceedings{vanhorn2018inaturalist,
title={The iNaturalist species classification and detection dataset},
author={Van Horn, Grant and others}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/inaturalist.CIFAR-CThe license is to the original authors (see below)!
This repository contains the CIFAR-10-C dataset from Benchmarking Neural Network Robustness to Common Corruptions and Perturbations. We are currently hosting it on Hugging Face due to an increased latency from Zenodo.
We are not the original authors. If you find this useful in your research, please consider citing:
@article{hendrycks2019robustness,
title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/CIFAR-C.Places365
Dataset Description
Places365 is a large-scale scene recognition dataset with 1.8M images across 365 scene categories. This split is derived from the OpenOOD benchmark OOD evaluation splits.
Homepage: http://places2.csail.mit.edu/
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation
@article{zhou2017places,
title={Places: A 10 million Image Database for Scene Recognition},
author={Zhou, Bolei and others},
journal={IEEE TPAMI},
year={2017}
}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/Places365.query-level-uncertaintyUncertaintyGym
UncertaintyGym
A Standardized Benchmark for LLM Epistemic Calibration & Uncertainty Expression
Abstract
UncertaintyGym evaluates whether language models recognize the boundaries of their knowledge. Rather than assessing purely factual recall, UncertaintyGym measures how reliably an LLM explicitly declares uncertainty ("I don't know"), requests necessary disambiguating context, and rejects false premises without hallucinating.
Benchmark Taxonomy… See the full description on the dataset page: https://huggingface.co/datasets/Muse-Ltd/UncertaintyGym.Openimage-O
Dataset Description
OpenImage-O is an OOD benchmark subset derived from Open Images dataset. This split is derived from the OpenOOD benchmark OOD evaluation splits.
Homepage: https://storage.googleapis.com/openimages/web/index.html
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation
@article{kuznetsova2020open,
title={The open images dataset v4: Unified image classification, object detection, and visual relationship detection at scale}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/Openimage-O.Imagenet1k
Dataset Description
The ImageNet-1K dataset contains over 1.2 million training images across 1,000 object categories.
Homepage: https://www.image-net.org/
Note : What's hosted in this repo is only the validation split. If you wish to downlaod the train split please use the official website.
Citation
@inproceedings{deng2009imagenet,
title={ImageNet: A large-scale hierarchical image database},
author={Deng, Jia and others},
booktitle={CVPR},
year={2009}
}
Segformer_b0_no_uncertaintyPythonic-ZHTWCheckpointsThe Checkpoints dataset as trained and used in A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors published at ICLR 2024. All models all trained and uploaded in a float16 format to reduce the memory footprint.
Usage
Untar the models
Just untar the desired models available in models, for instance with:
tar -xvf models/cifar10-resnet18/cifar10-resnet18-0-1023.tgz
Most of them are regrouped in tar files containing 1024 models each. This will create a new… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/Checkpoints.Texture
Dataset Description
The Describable Textures Dataset (DTD) includes 5,640 images of textures annotated with human-centric attributes. This split is derived from the OpenOOD benchmark OOD evaluation splits.
Homepage: https://www.robots.ox.ac.uk/~vgg/data/dtd/
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation
@inproceedings{cimpoi2014describing,
title={Describing textures in the wild},
author={Cimpoi, Mircea and others},
booktitle={CVPR}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/Texture.honest-uncertainty-sft-100k
Honest Uncertainty SFT (100K)
100,000 ShareGPT conversations demonstrating calibrated epistemic humility across 21 scenarios. Each example shows a model correctly expressing what it knows, what it doesn't know, and why -- without being uselessly vague or confidently wrong.
Targets the hallucination and overconfidence failure modes that are the #1 complaint in enterprise AI deployments.
Motivation
LLMs have a systematic bias toward confident-sounding responses… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/honest-uncertainty-sft-100k.Uncertainty-Aware-Narrative-Dataset
Dataset Description (Uncertainty-Aware Narrative Synthesis)
This Uncertainty-Aware Narrative Synthesis Dataset is a public resource under active development by the Bridge-AI Lab under the direction of Dr. Shubhra Kanti (Santu) Karmaker. The dataset is being developed in collaboration with the Orange County Office of Emergency Management and the Seminole County Office of Emergency Management in Florida, incorporating the operational expertise of emergency-management professionals… See the full description on the dataset page: https://huggingface.co/datasets/BridgeAI-Lab/Uncertainty-Aware-Narrative-Dataset.SSB_hard
Dataset Description
SSB-Hard is an OOD dataset focusing on semantically similar but distinct categories to ImageNet. This split is derived from the OpenOOD benchmark OOD evaluation splits.
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation
@inproceedings{vaze2022openset,
title={Open-set Recognition: A Good Closed-set Classifier is All You Need?},
author={Vaze, Siddharth and others},
booktitle={ICLR},
year={2022}
}… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/SSB_hard.LLM_COTtiny-imagenet-200
Dataset Description
Tiny ImageNet is a reduced version of the original ImageNet dataset, containing 200 classes (a subset of the 1,000 ImageNet categories)
Homepage: https://www.image-net.org/
Citation
@inproceedings{deng2009imagenet,
title={ImageNet: A large-scale hierarchical image database},
author={Deng, Jia and others},
booktitle={CVPR},
year={2009}
}
data_gpt54_only_answer_lossNinco
Dataset Description
NINCO (Natural In-Distribution Corruptions) is a dataset designed for evaluating robustness and OOD detection under natural corruptions.It includes real-world corrupted versions of ImageNet images to test model generalization beyond synthetic distortions. This split is derived from the OpenOOD benchmark OOD evaluation splits.
Homepage: https://github.com/j-cb/NINCO
OpenOOD Benchmark: https://github.com/Jingkang50/OpenOOD/
Citation… See the full description on the dataset page: https://huggingface.co/datasets/torch-uncertainty/Ninco.coqar-s-uncertainty
Fields
id - unique example ID.
split - original CoQAR split (train / dev).
conversation_id — source conversation ID.
turn_id - turn index inside the conversation.
story - source passage shared by all variants.
answer - gold answer from CoQAR.
answer_span_text - supporting answer span in the passage.
all_standalone_questions - human-written standalone rewrites of the original question.
variants.low_s - standalone low specification version.
variants.high_s - original question… See the full description on the dataset page: https://huggingface.co/datasets/zykov/coqar-s-uncertainty.1k_uncertainty_CoT_data_day0_third_path_multi_turntrain2_1_2icml2026-epistemic-uncertainty-reprouncertainty-vlm-gemma-emnlp_stage
uncertainty-vlm-gemma-official EMNLP stage-wise train/dev/test splits
This dataset contains deterministic stratified 80/10/10 train/dev/test assignments for the four diagram stages. Each split uses the stage_label column as the binary target for that stage.
uncertainty-vlm-llama-emnlp_stage
uncertainty-vlm-llama-official EMNLP stage-wise train/dev/test splits
This dataset contains deterministic stratified 80/10/10 train/dev/test assignments for the four diagram stages. Each split uses the stage_label column as the binary target for that stage.
LLM-Uncertainty-BenchDatasets comprising 10,000 instances each used for uncertainty quantification in LLMs.
mmlu_10k is used for question answering.
cosmosqa_10k is used for reading comprehension.
hellaswag_10k is used for commonsense inference.
halu_dialogue is used for dialogue response selection.
halu_summarization is used for document summarization.
For more details on how these datasets are utilized, check out our github repo: https://github.com/smartyfh/LLM-Uncertainty-Bench/tree/main
uncertainty-vlm-qwen3-emnlp_stage
uncertainty-vlm-qwen3-official EMNLP stage-wise train/dev/test splits
This dataset contains deterministic stratified 80/10/10 train/dev/test assignments for the four diagram stages. Each split uses the stage_label column as the binary target for that stage.
worldmodel-uncertainty-raw-half-val-seed42
World Model Uncertainty raw half-validation caches
Lossless uint8 RGB clips and actions for deterministic Bridge, RoboNet, and
RT-1 calibration evaluation. Each directory is independently downloadable.
Its portable manifest fixes the selected episodes, clip starts, and SHA-256
of every array. These are raw pixel targets, not VAE latents.
uncertainty-vlm-qwen2p5-emnlp_stage
uncertainty-vlm-qwen2p5-official EMNLP stage-wise train/dev/test splits
This dataset contains deterministic stratified 80/10/10 train/dev/test assignments for the four diagram stages. Each split uses the stage_label column as the binary target for that stage.
Dist_Uncertainty
Dist_Uncertainty – Hyperspectral Case Studies for Plant Trait Uncertainty Assessment
Dataset Description
This dataset contains hyperspectral remote sensing imagery and associated land cover labels used as out-of-domain (OOD) test cases for evaluating uncertainty estimation methods in deep learning-based plant trait retrievals. It accompanies the paper by Cherif et al. (2025, Biogeosciences) and supports the evaluation of a distance-based uncertainty method (Dis_UN)… See the full description on the dataset page: https://huggingface.co/datasets/Avatarr05/Dist_Uncertainty.