datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
amazon_counterfactual
AmazonCounterfactualClassification
An MTEB dataset
Massive Text Embedding Benchmark
A collection of Amazon customer reviews annotated for counterfactual detection pair classification.
Task category
t2c
Domains
Reviews, Written
Reference
https://arxiv.org/abs/2104.06893
How to evaluate on this task
You can evaluate an embedding model on this dataset using the following code:
import mteb
task = mteb.get_tasks(["AmazonCounterfactualClassification"])… See the full description on the dataset page: https://huggingface.co/datasets/mteb/amazon_counterfactual.Countdown-Tasks-3to4CountBenchQAThis dataset was introduced in PaliGemma for evaluating counting in vision language models. This version only includes 491 images from the original CountBench dataset, since some of the original URLs can no longer be accessed.
Original Description
CountBench: We introduce a new object counting benchmark called CountBench,
automatically curated (and manually verified) from the publicly available
LAION-400M image-text dataset. CountBench contains a total of 540 images
containing… See the full description on the dataset page: https://huggingface.co/datasets/vikhyatk/CountBenchQA.CountQA
Dataset Summary
CountQA is the new benchmark designed to stress-test the Achilles' heel of even the most advanced Multimodal Large Language Models (MLLMs): object counting. While modern AI demonstrates stunning visual fluency, it often fails at this fundamental cognitive skill, a critical blind spot limiting its real-world reliability.
This dataset directly confronts that weakness with over 1,500 challenging question-answer pairs built on real-world images, hand-captured to feature… See the full description on the dataset page: https://huggingface.co/datasets/Jayant-Sravan/CountQA.CS2CD.Counter-Strike_2_Cheat_Detection
Counter Strike 2 Cheat Detection Dataset
Overview
The CS2CD (Counter-Strike 2 Cheat Detection) dataset is an anonymised dataset comprised of Counter-Strike 2(CS2) gameplay at a variety of skill-levels with cheater annotations. This dataset contains 478 CS2 matches with no cheater present, and 317 matches CS2 matches with at least one cheater present.
Dataset structure
The dataset is partitioned into data with at least one cheater present, and data with no… See the full description on the dataset page: https://huggingface.co/datasets/CS2CD/CS2CD.Counter-Strike_2_Cheat_Detection.counterfact-tracing
Dataset Card for "counterfact-tracing"
This is adapted from the counterfact dataset from the excellent ROME paper from David Bau and Kevin Meng.
This is a dataset of 21919 factual relations, formatted as data["prompt"]==f"{data['relation_prefix']}{data['subject']}{data['relation_suffix']}". Each has two responses data["target_true"] and data["target_false"] which is intended to go immediately after the prompt.
The dataset was originally designed for memory editing in models. I made… See the full description on the dataset page: https://huggingface.co/datasets/NeelNanda/counterfact-tracing.counterfact
Dataset Card for "counterfact"
Dataset from ROME by Meng et al.
More Information needed
FaithEval-counterfactual-v1.0
FaithEval
FaithEval is a new and comprehensive benchmark dedicated to evaluating contextual faithfulness in LLMs across three diverse tasks: unanswerable, inconsistent, and counterfactual contexts.
[Paper] FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows", ICLR 2025, https://arxiv.org/abs/2410.03727
[Code and Detailed Instructions] https://github.com/SalesforceAIResearch/FaithEval
Disclaimer and Ethical Considerations… See the full description on the dataset page: https://huggingface.co/datasets/Salesforce/FaithEval-counterfactual-v1.0.pixmo-count
PixMo-Count
PixMo-Count is a dataset of images paired with objects and their point locations in the image.
It was built by running the Detic object detector on web images, and then filtering the data
to improve accuracy and diversity. The val and test sets are human-verified and only contain counts from 2 to 10.
PixMo-Count is a part of the PixMo dataset collection and was used to
augment the pointing capabilities of the Molmo family of models
Quick links:
📃 Paper
🎥 Blog with… See the full description on the dataset page: https://huggingface.co/datasets/allenai/pixmo-count.VSI-SUPER-Count
VSI-SUPER-Count
Website | Paper | GitHub | Models
Authors: Shusheng Yang*, Jihan Yang*, Pinzhi Huang†, Ellis Brown†, et al.
VSI-SUPER-Count is a benchmark for testing continual counting capabilities across changing viewpoints and scenes in arbitrarily long videos. It challenges models to maintain accurate object counts as new objects appear throughout extended video sequences.
Overview
VSI-SUPER-Count evaluates spatial supersensing by testing whether models can:
Count… See the full description on the dataset page: https://huggingface.co/datasets/nyu-visionx/VSI-SUPER-Count.pixmo-point-count-concat_0-20SP_CountingThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so_follower",
"total_episodes": 50,
"total_frames": 37501,
"total_tasks": 5,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:50"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/justintiensmith/SP_Counting.ai2thor-counting-largeMulti-Hop-Objects-Countingpixmo-point-count-gen-undCountdown-Tasks-3to4-Uniqueprocthor-100-counting-balancedcountry211
Dataset Card for Country211
The Country 211 Dataset from OpenAI.
This dataset was built by filtering the images from the YFCC100m dataset that have GPS coordinate corresponding to a ISO-3166 country code. The dataset is balanced by sampling 150 train images, 50 validation images, and 100 test images images for each country.
countbench
Dataset Card for "countbench"
This dataset was introduced in the paper Teaching CLIP to Count to Ten.
Genre-Classifier-Country-Per-Country
Name Dataset — Gender Classifier Parquet
Parquet conversion of philipperemy/name-dataset for first-name gender classification.
Source
Original repository: https://github.com/philipperemy/name-dataset
Original archive: name_dataset.zip
Original CSV format: first_name,last_name,gender,country_code
Converted format: first_name,gender
One Hugging Face config/subset per country code.
Cleaning
Rows are removed when:
first_name is null, empty, or… See the full description on the dataset page: https://huggingface.co/datasets/SpiceeChat/Genre-Classifier-Country-Per-Country.metrics-danbooru2025-alltime-tag-counts
dataproc5/metrics-danbooru2025-alltime-tag-counts
Dataset Overview
tag_count provides aggregated tag usage statistics from the Danbooru2025 dataset. Each entry corresponds to a specific tag's usage count in all time.
import unibox as ub
df = ub.loads("hf://dataproc5/metrics-danbooru2025-monthly-tag-counts").to_pandas()
alltime_tag_counts = df.groupby(["tag_string", "tag_type"], as_index=False)["tag_count"].sum()
alltime_tag_counts =… See the full description on the dataset page: https://huggingface.co/datasets/dataproc5/metrics-danbooru2025-alltime-tag-counts.sd-prompt-image-in-the-wild-counterfeitcounterfactual-physicskbot_cappuccino_countThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "kbot_right_arm_follower",
"total_episodes": 224,
"total_frames": 166988,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:224"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet"… See the full description on the dataset page: https://huggingface.co/datasets/infoslack/kbot_cappuccino_count.counterfactual_culture
Counterfactual Culture
Multilingual minimal-change counterfactual etiquette vignettes for five cultures,
with conforming / violating pairs for factorization and representation studies.
Cultures
english (US norms), japan, china, india, russia
Languages
en, ja, zh, hi, ru (full cross: every culture × every language)
Samples
152,500 (76,250 pairs)
Seed samples
610 English seed vignettes (before variation expansion)
Norms
305 etiquette norms… See the full description on the dataset page: https://huggingface.co/datasets/nirmalendu01/counterfactual_culture.CounterStrike-1K
CounterStrike-1K
1,490 rendered POV-hours · 7,347 synchronized rounds · 73,470 POV clips · 7 maps · 720p + audio
Synchronized 10-POV rounds with per-frame action overlays
Seven active-duty maps
Ancient · Anubis · Dust2 · Inferno · Mirage · Nuke · Overpass
CounterStrike-1K is the first grounded, professional-grade Counter-Strike 2 dataset with 10 synchronized first-person perspectives per round, captured from professional match demos. It is designed for video… See the full description on the dataset page: https://huggingface.co/datasets/ArnieRamesh/CounterStrike-1K.worldcuisines_format_sea_country_only_with_metadataclevr_count_70kThis dataset is borrowed from clevr_cogen_a_train
pixmo-count-filtered-imgContainedVisual-Counterfact
Visual CounterFact: Controlling Knowledge Priors in Vision-Language Models through Visual Counterfactuals
This dataset is part of the work "Pixels Versus Priors: Controlling Knowledge Priors in Vision-Language Models through Visual Counterfacts".📖 Read the Paper💾 GitHub Repository
Overview
Visual CounterFact is a novel dataset designed to investigate how Multimodal Large Language Models (MLLMs) balance memorized world knowledge priors (e.g., "strawberries are red")… See the full description on the dataset page: https://huggingface.co/datasets/mgolov/Visual-Counterfact.