extract gated 0 0

Carol0110/ASABench

Name: Carol0110/ASABench
Creator: Carol0110
Published: 2025-06-17T07:30:50

published by @Carol0110

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads

Last 30 days

Likes

Size

3.7 MB

Created on HF

2025-06-17

Age

1 years ago

Description

ASABench

ASABench is a dataset for evaluating the safety robustness of large language models under latent-space perturbations.

The dataset is constructed based on Activation Steering Attack (ASA), a probing method that injects controlled perturbations into hidden representations to assess the stability of safety-aligned behaviors.

Overview

Existing safety evaluations are primarily behavior-level and rely on input-space perturbations or red-teaming datasets. In… See the full description on the dataset page: https://huggingface.co/datasets/Carol0110/ASABench.

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

2026-04-09[bot] Conversion to Parquetopen1 💬#1
Loading...

Continue on Hugging Face →

Related datasets same stage · extract

Read further

Community best practices (extraction)

5 rules the field converged on for building the pair.

Building your own extraction pair

When the reference pair is not enough. Sizing, framing, and evaluation.

Datasets that shaped abliteration

The five corpora that did most of the work.

Non-English abliteration communities

How Chinese, Russian, Japanese, and Korean practitioners build extraction pairs.

Repository

FILE .gitattributes FILE Llama-2-13B-Chat_asabench_seed42.json FILE Llama-3.1-8B-Base_asabench_seed42.json FILE Llama-3.1-8B-Instruct_asabench_seed42.json FILE Llama-3.2-3B-Base_asabench_seed42.json FILE Llama-3.2-3B-Instruct_asabench_seed42.json FILE Llama-3.3-70B-Instruct_asabench_seed42.json FILE Qwen-2.5-7B-Base_asabench_seed42.json FILE Qwen-2.5-7B-Instruct_asabench_seed42.json FILE README.md

File contents live on Hugging Face. Click a file or folder to open it there.

Download the dataset

Gated dataset. You need to accept the author's terms on Hugging Face before downloading. Visit the repository, click Agree and access repository, then log in locally with huggingface-cli login and use one of the snippets below.

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)

from datasets import load_dataset
ds = load_dataset("Carol0110/ASABench")

Raw snapshot (Python)

from huggingface_hub import snapshot_download
snapshot_download(repo_id="Carol0110/ASABench", repo_type="dataset")

git clone (requires git-lfs)

git clone https://huggingface.co/datasets/Carol0110/ASABench

Browse repository on Hugging Face ->