pretty_name: Harmful Requests (English)
language:
- en
license: cc-by-4.0
size_categories: - n<1K
task_categories: - text-generation
- text-classification
annotations_creators: - machine-generated
language_creators: - machine-generated
source_datasets: - original
tags: - text
- synthetic
- contrastive
- response-sensitivity
- datasets
configs: - config_name: default
default: true
data_files:- split: train
path: data/train.parquet
- split: train
Harmful Requests (English)
This dataset contains 556 English text requests for evaluating LLM model safety and abliteration.
Explore: Columns · Load the data · Creation and review
General properties
| Property | Value |
|---|---|
| Version | 1.0.0 |
| Size | 556 requests |
| Language | English |
| Coverage | 10 domains · 49-60 entries per domain |
| Length | 13–80 words per request |
| Formats | Parquet for the viewer · JSONL mirror |
| License | CC BY 4.0 |
Columns
Each row contains:
| Column | Meaning |
|---|---|
request |
Full text of the harmful request. |
category |
Main category of the harm. |
context |
Could be "implicit" or "explicit", conveying whether request explaines user's motive. |
id |
Unique ID inside one specific category |
Column definitions and coverage counts are included.
Load the data
from datasets import load_dataset
requests = load_dataset(
"strangecreator/harmful-requests-english",
split="train",
)
request = requests[0]["request"]
Load the JSONL mirror locally
from datasets import load_dataset
requests = load_dataset(
"json",
data_files="data/train.jsonl",
split="train",
)
Creation and review
The examples were generated with AI assistance (using abliterated Qwen3.8-27B-4bit) using a manual balanced plan. Review underwent partial manual checks.
Scope
The collection supports LLM safety evaluations and abliteration pipelines.
Credits and license
Created by strangecreator. Available under CC BY 4.0.
Citation
@misc{strangecreator2026harmfulrequestsenglish,
author = {strangecreator},
title = {Harmful Requests (English)},
year = {2026},
howpublished = {Hugging Face dataset},
url = {https://huggingface.co/datasets/strangecreator/harmful-requests-english}
}