datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Pretrain-Behaviors
Pretrain-Behaviors
Dataset Description
Behavior-focused text covering reasoning, planning, data science, games, general content, and format rewriting. This repository is part of the K2 Horizon collection.
The repository is organized into multiple subsets. Every subset has a train split backed by Parquet shards, which supports Dataset Viewer inspection and streaming access.
K2 Horizon Dataset Series
Dataset repository
Focus
Subsets… See the full description on the dataset page: https://huggingface.co/datasets/IFM/Pretrain-Behaviors.human_behavior_atlas_tar
Human Behavior Atlas (HBA)
Human Behavior Atlas (HBA) is a unified benchmark for multimodal behavioral understanding.It aggregates and standardizes multiple behavioral datasets into a single training and evaluation framework, enabling consistent training and evaluation of foundation models on psychological and social behavior tasks (e.g., emotion, intent, sarcasm, mental health signals, nonverbal behavior).
Dataset on Hugging Face:… See the full description on the dataset page: https://huggingface.co/datasets/HumanBehaviorAtlas/human_behavior_atlas_tar.behavioral-fine-tuning-v1
Why This Dataset Exists
"A model that refuses everything is useless. A model that refuses nothing is dangerous. The goal is a model that thinks."
The Problem
Our Solution
Uncensored data → helpful but uncontrolled
Surgical 85% helpfulness + 13% safety + 2% eval mix
Safety-only data → lobotomized, over-refusing models
Calibrated ratio preserves full helpfulness
Raw data → PII, leaked secrets, duplicates
7-stage pipeline validates every… See the full description on the dataset page: https://huggingface.co/datasets/abhinav00anand/behavioral-fine-tuning-v1.LAMBDA
Dataset Summary
LAMDBA is a long term ad memorability dataset, featuring data from 1749 participants and 2205 ads across 276 brands.
Dataset Structure
from datasets import load_dataset
ds = load_dataset("behavior-in-the-wild/LAMBDA")
ds
DatasetDict({
train: Dataset({
features: ['video_id', 'recall_score', 'youtube_id', 'ad_details'],
num_rows: 1964
})
test: Dataset({
features: ['video_id', 'recall_score', 'youtube_id', 'ad_details']… See the full description on the dataset page: https://huggingface.co/datasets/behavior-in-the-wild/LAMBDA.identity-behavior
identity-behavior
Recipe: recipes/04-train/identity · Collections: Character, Start here: foundational post-training datasets
Teach an open model who it is.
Identity behavior is the simplest thing every shipped assistant needs and
open models do not have out of the box: a consistent answer to "who are
you?" and "who made you?", in every phrasing and every language, without a
system prompt propping it up. Ask a base Qwen model and it tells you about
Alibaba; put a persona in the… See the full description on the dataset page: https://huggingface.co/datasets/while-ai/identity-behavior.behaviortune-v1-1-r1
BehaviorTune Dataset
Controlled synthetic dataset used to train and evaluate BehaviorTune, a QLoRA post-training project on Qwen/Qwen3-4B-Instruct-2507.
It contains 544 scenarios across six splits, including 240 training rows, 48 development rows, and a 64-row eval_core set used for the published matched evaluation.
The dataset supports completion-only QLoRA training and deterministic BASE / SYSTEM / CONTEXT / QLoRA evaluation. V1.1-R1 is the frozen dataset/version identifier.… See the full description on the dataset page: https://huggingface.co/datasets/aamish-ahmad/behaviortune-v1-1-r1.human_behavior_atlas
Human Behavior Atlas (HBA)
Human Behavior Atlas (HBA) is a unified benchmark for multimodal behavioral understanding.It aggregates and standardizes multiple behavioral datasets into a single training and evaluation framework, enabling consistent training and evaluation of foundation models on psychological and social behavior tasks (e.g., emotion, intent, sarcasm, mental health signals, nonverbal behavior).
Dataset on Hugging Face:… See the full description on the dataset page: https://huggingface.co/datasets/droiden/human_behavior_atlas.behavior_grounding
Behaviorally Grounded User Profiles from the Wild
Open-ended, anonymized user profiles distilled from authentic social-media behavior, released with the paper
"Behaviorally Grounded User Profiles from the Wild for Personalized Alignment and Multi-Perspective Reasoning."
Persona-driven methods for personalizing LLMs typically rely on rigid synthetic personas built from a small set of
categorical attributes (age, gender, nationality). These flatten individual variation and lean on… See the full description on the dataset page: https://huggingface.co/datasets/UWaterloo/behavior_grounding.playcat-cat-behavior-new-data-set
PlayCat Cat Behavioral Enrichment Dataset
The definitive multilingual research dataset on cat behavioral enrichment by PlayCat Research
Dataset Summary
The PlayCat Cat Behavioral Enrichment Dataset is the largest open, bilingual (Korean-English) collection dedicated to feline environmental enrichment research. It contains 12,262 deduplicated entries spanning peer-reviewed academic papers, patents, veterinary Q&A, and community knowledge on cat behavior enrichment… See the full description on the dataset page: https://huggingface.co/datasets/playcat/playcat-cat-behavior-new-data-set.claudesidian-behaviors-merged
Claudesidian Merged Behavioral Dataset
Dataset Description
This dataset contains 1,852 synthetic training examples demonstrating 8 different behavioral patterns for training language models to use the Claudesidian-MCP toolset effectively with Obsidian vaults.
The dataset is specifically formatted for KTO (Kahneman-Tversky Optimization) preference learning with properly interleaved positive and negative examples.
Behavioral Categories
This dataset includes… See the full description on the dataset page: https://huggingface.co/datasets/professorsynapse/claudesidian-behaviors-merged.SDR-Bench
SDR-Bench: A Benchmark for Sales Development Representative Agents
This dataset contains 6,279 verified business success stories from various corporate domains.
It was curated for the SDR-Bench paper.
This dataset serves as a benchmark for evaluating AI agents on their ability to conduct deep research and generate targeted sales pitch points. The data is derived from real-world Customer Success Stories, where the "Ground Truth" consists of the actual value propositions and pain… See the full description on the dataset page: https://huggingface.co/datasets/behavior-in-the-wild/SDR-Bench.behavioral-loops
Behavioral Loops
1,140 behavioral patterns across 279 categories, each structured as given/when/then/result logic with taxonomy classification, veracity scores, and intervention strategies.
Quick Start
from datasets import load_dataset
ds = load_dataset("buley/behavioral-loops")
print(ds["train"][0])
Structure
Field
Description
given
Initial condition or context
when
Trigger event
then
Resulting behavior
result
Long-term outcome
origin… See the full description on the dataset page: https://huggingface.co/datasets/buley/behavioral-loops.grounded-behavior-framework-v1_5
Grounded Behavior Framework N1 v1.5
Dataset sintético em português europeu para treino e avaliação de respostas
fundamentadas num contexto fornecido. Cada exemplo contém um contexto, uma
pergunta e uma resposta curta que aparece literalmente no contexto.
Como carregar
from datasets import load_dataset
dataset = load_dataset("empgces/grounded-behavior-framework-v1_5")
print(dataset)
print(dataset["train"][0])
Splits
Split
Exemplos
Utilização… See the full description on the dataset page: https://huggingface.co/datasets/empgces/grounded-behavior-framework-v1_5.harmful_behaviors_ja_synth
harmful_behaviors_ja_synth
Japanese synthetic harmful-behavior prompts for safety/refusal evaluation.
Columns:
text: prompt text
science_behavioral_and_domain_diversity_dataset
Nepali Science SFT Dataset — Clean Candidate
A high-quality Nepali Science Supervised Fine-Tuning (SFT) dataset containing short question–answer instruction-following examples written primarily in Nepali Devanagari script.
This release is the clean candidate produced after structural validation, language checks, duplicate analysis, and Unicode-contamination filtering.
Dataset Overview
Property
Value
Dataset file
clean_candidate.jsonl
Records
29,320… See the full description on the dataset page: https://huggingface.co/datasets/Somtharu181coder/science_behavioral_and_domain_diversity_dataset.Caca-Behaviorgemma-4-e2b-deception-behavior-completions
Gemma-4-E2B deception & behavior completions
Consolidated 910-row corpus of (scenario prompt + Gemma-4-E2B-generated completion) pairs from earlier mechanistic-interpretability experiments. Each row captures the prompt the model saw and the text it actually produced; for a subset, Claude-Haiku-4-5 judge verdicts and SAE-feature labels are included.
The corpus is meant to be used as activation-extraction input for downstream interpretability work — Natural Language Autoencoder (NLA)… See the full description on the dataset page: https://huggingface.co/datasets/Solshine/gemma-4-e2b-deception-behavior-completions.infosec_harmful_behaviors
Infosec Harmful Behaviors
Offensive-security instruction prompts for refusal-direction research and abliteration of code/security models.
Dataset Details
This dataset contains infosec-domain harmful prompts intended to elicit refusal behavior from aligned instruction models. It is designed as the harmful side of a harmful/harmless contrast pair, analogous to mlabonne/harmful_behaviors but focused on offensive-security and malicious-coding requests.
Rows:
train:… See the full description on the dataset page: https://huggingface.co/datasets/zaakirio/infosec_harmful_behaviors.grounded-behavior-n1-pt
Dataset Description
Synthetic European Portuguese grounded question-answering examples generated by multiple model providers.
Objective
Train models to answer from the supplied context rather than external knowledge.
Dataset Structure
JSONL splits: train (4440), validation (250), and test (240).
Data Fields
Each row contains an ID, context, question, answer, source grouping metadata, and available curriculum metadata.… See the full description on the dataset page: https://huggingface.co/datasets/empgces/grounded-behavior-n1-pt.harmful_behaviors_ja
harmful_behaviors_ja_synth
mlabonne/harmful_behaviors を DeepSeek v4 pro を用いて日本語訳したものです
Japanese synthetic harmful-behavior prompts for safety/refusal evaluation.
Columns:
text: prompt text
harmless_behaviors_ja_synth
harmless_behaviors_ja_synth
Japanese synthetic harmless instruction prompts for ordinary-response / refusal-direction evaluation.
Splits
train: 2400
test: 600
Columns
id: stable hash ID
text: Japanese harmless instruction prompt
label: always good
category: rough generation category
lang: always ja
source: generation source
r16-behavioral-metamerism-pilot
R16 Behavioral Metamerism Pilot
Brand Function x synthetic cohort interaction experiment from the Spectral Brand Theory research program.
Dataset Summary
675 API calls testing whether Brand Function specification differentially affects dimensional collapse across synthetic observer cohorts. Design: 5 cohorts x 5 brands x 3 conditions (no BF, structural BF, enriched BF) x 3 models x 3 repetitions.
Companion paper: AI-Native Brand Identity: From Visual Recognition… See the full description on the dataset page: https://huggingface.co/datasets/spectralbranding/r16-behavioral-metamerism-pilot.behavioral-probes-v1
behavioral-probes-v1
Dataset of surprising, unintended behavioral generalizations detected in fine-tuned LoRAs via a next-token logit-shift probe bank.
Schema
column
description
prompt_id
LoRA identifier (matches introspection-auditing/models/{prompt_id} on HF)
question
Auditing question
answer
Description of how the LoRA differs from base, or denial of a false hypothesis
qa_type
behavior_probe — how would the LoRA differ from baseline on Xdemo — concrete… See the full description on the dataset page: https://huggingface.co/datasets/ceselder/behavioral-probes-v1.