DATASETS
The abliteration workbench
A living encyclopedia of the datasets that make abliteration possible. Curated from the 1M-dataset Hugging Face corpus down to the 3,910 that actually matter here, then organised by the stage of the workflow where each is used. New datasets are folded in weekly.
3,910
Datasets tracked
13.2M
Total downloads
216
New this month
1,980
Publishers
6
Workflow stages
2,737
Discussion threads
Ask the datasets
LIVE
Query 2,400+ datasets, 18 curated recipes, and 53 wiki articles in plain English.
Hint: type @ to bring a dataset, recipe, or method into context
Reads every dataset card, recipe, and wiki article. Tools catalog → 23 best practices → Known gaps → See public API →
Stage
Extract
Harmful and harmless pairs used to identify the refusal direction.
1,461 datasets · 6.3M downloads · +85 this month
01 VIEW → 02 VIEW → 03 VIEW →
Anthropic/hh-rlhf
Dataset Card for HH-RLHF
Dataset Summary
This repository provides access to two different kinds of dat…
djain95/sae-jailbreaks-cache
mosaicml/dolly_hhrlhf
Dataset Card for "dolly_hhrlhf"
This dataset is a combination of Databrick's dolly-15k dataset and a filtered subset …
Stage
Ablate
Direction-specific corpora and activation probes for weight editing.
303 datasets · 357.1k downloads · +19 this month
01 VIEW → 02 VIEW → 03 VIEW →
lasrprobegen/refusal-activations
Refusal Activations Dataset
This dataset is now configured to load the full ~97k samples from jailbreak_mixed_100k.cs…
lu-christina/assistant-axis-vectors
The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
This repository contains pre-com…
stablellama/erotic-image-prompts
Dataset Card for Erotic Image Prompts
Dataset Description
Dataset Summary
Large languag…
Stage
Heal
DPO and ORPO preference mixes to recover capability after ablation.
229 datasets · 975.7k downloads · +14 this month
01 VIEW → 02 VIEW → 03 VIEW →
argilla/distilabel-intel-orca-dpo-pairs
distilabel Orca Pairs for DPO
The dataset is a "distilabeled" version of the widely used dataset: Intel/orca_dpo_pair…
arcee-ai/distilabel-intel-orca-dpo-pairs-binarized
This is the binarized version of distilabel Orca Pairs for DPO and ORPO.
Reference: https://huggingface.co/datasets/arg…
Intel/orca_dpo_pairs
The dataset contains 12k examples from Orca style dataset Open-Orca/OpenOrca.
Stage
Evaluate
Jailbreak benchmarks, over-refusal probes, and judge datasets.
851 datasets · 3.8M downloads · +47 this month
01 VIEW → 02 VIEW → 03 VIEW →
google/IFEval
Dataset Card for IFEval
Dataset Summary
This dataset contains the prompts used in the Instruction-Fo…
JailbreakBench/JBB-Behaviors
An Open Robustness Benchmark for Jailbreaking Language Models
NeurIPS 2024 Datasets and Benchmarks Track
…
Beijing-AISI/panda-bench
PandaBench
PandaBench is a comprehensive benchmark for evaluating Large Language Model (LLM) safety, focusing on jail…
Stage
Fine-tune
Uncensored SFT corpora, the Dolphin / Wizard tradition.
340 datasets · 598.9k downloads · +23 this month
01 VIEW → 02 VIEW → 03 VIEW →
QuixiAI/dolphin
Dolphin 🐬
https://erichartford.com/dolphin
Dataset details
This dataset is an attempt to replicate the r…
nvidia/Nemotron-SFT-Safety-v1
Dataset Description:
The Nemotron-SFT-Safety-v1 data is designed to align models to be robust against a variety of sa…
open-llm-leaderboard-old/details_ehartford__WizardLM-1.0-Uncensored-Llama2-13b
Dataset Card for Evaluation run of ehartford/WizardLM-1.0-Uncensored-Llama2-13b
Dataset Summary
Datase…
Stage
Misc
Passed the relevance filter but does not cleanly map to a stage.
726 datasets · 1.2M downloads · +28 this month
01 VIEW → 02 VIEW → 03 VIEW →
nvidia/Aegis-AI-Content-Safety-Dataset-2.0
🛡️ Nemotron Content Safety Dataset V2
The Nemotron Content Safety Dataset V2, formerly known as Aegis AI Content S…
allenai/tulu-3-sft-personas-instruction-following
Dataset Descriptions
This dataset contains 29980 examples and is synthetically created to enhance model's capabilitie…
meta-llama/Llama-3.1-8B-Instruct-evals
Dataset Card for Llama-3.1-8B-Instruct Evaluation Result Details
This dataset contains the Meta evaluation result d…
Market analytics
The dataset market, in numbers
Nine views of the 2,443 relevant datasets. License landscape, size distribution, publisher concentration, download velocity, and the acceleration of the last three years.
Growth by category / year
The ingredient supply is accelerating
Extract Ablate Heal Evaluate Fine-tune Misc
2022 6
2023 216
2024 844
2025 1,108
2026 1,736
License
43.5%
carry a recognizable open license
Size
9.48 TB
total corpus, median 11.2 MB
Language
36.9%
English-primary, 0 multilingual
Freshness
14.7%
created in the last 3 months
Publishers
54.4%
of downloads by top 10 authors
Top 10 authors 1.5% of datasets
Top 10 downloads share 54.4%
Long-tail authors 1,489 (75.2%)
Gated datasets 351 (9%)
Velocity 30d
708.6k
downloads across the top-10 this month
- 01 djain95/sae-jailbreaks-cache 355.1k
- 02 google/IFEval 160.3k
- 03 Anthropic/hh-rlhf 34.2k
- 04 mlabonne/harmless_alpaca 31.2k
- 05 mlabonne/harmful_behaviors 31.0k
- 06 JailbreakBench/JBB-Behaviors 27.8k