Get the app
ablate 0 just added 0

sunnycloudhust/CHaRS-activations

published by @sunnycloudhust

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
5.3K
Last 30 days
5.3K
Likes
2
Size
63.8 GB
Created on HF
2026-10-14
Age
8 days ago

Description

CHaRS: Concept Heterogeneity-Aware Representation Steering

This repository contains implementation of CHaRS (Concept Heterogeneity-Aware Representation Steering), based on the research presented at ICML 2026.

The project explores concept heterogeneity-aware representation steering for large language models (LLMs), investigating how heterogeneous representations can be leveraged to steer model behavior at inference time.

My implementation includes multiple steering configurations… See the full description on the dataset page: https://huggingface.co/datasets/sunnycloudhust/CHaRS-activations.

Tags

license:mitregion:usconcept-heterogeneityrepresentation-steeringactivation-steeringoptimal-transportlarge-language-models

README current version from Hugging Face

license: mit
tags:

  • concept-heterogeneity
  • representation-steering
  • activation-steering
  • optimal-transport
  • large-language-models

CHaRS: Concept Heterogeneity-Aware Representation Steering

This repository contains implementation of CHaRS (Concept Heterogeneity-Aware Representation Steering), based on the research presented at ICML 2026.

The project explores concept heterogeneity-aware representation steering for large language models (LLMs), investigating how heterogeneous representations can be leveraged to steer model behavior at inference time.

My implementation includes multiple steering configurations, including CHaRS-AA (Activation Addition), CHaRS-DA (Directional Ablation), and CHaRS-PCT, enabling experiments with different representation manipulation strategies.

This repository is intended for research, experimentation, and further exploration of concept heterogeneity-aware representation steering.

README history 20 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-10-19Update output/Qwen2.5-3B-Instruct from Kaggle51e9e70971 B
    Loading...
  2. 2026-10-19Update jailbreaking from Kaggleddeb6a2971 B
    Loading...
  3. 2026-10-19Rename output folders9fb00ff971 B
    Loading...
  4. 2026-10-19Upload eval_tinybench.pyc40175d971 B
    Loading...
  5. 2026-10-18Upload 4 filesb7aecc6971 B
    Loading...
  6. 2026-10-18Upload folder using huggingface_hub37ff744971 B
    Loading...
  7. 2026-10-18Upload folder using huggingface_hub04ccb13971 B
    Loading...
  8. 2026-10-18Upload folder using huggingface_hubefeaf87971 B
    Loading...
  9. 2026-10-18Upload folder using huggingface_hubf5100ca971 B
    Loading...
  10. 2026-10-18Upload folder using huggingface_hub09e495d971 B
    Loading...
  11. 2026-10-18Upload folder using huggingface_hubc75f77e971 B
    Loading...
  12. 2026-10-18Rename output to output-run128bd008971 B
    Loading...
  13. 2026-10-18Upload CHaRS evaluation results384971b971 B
    Loading...
  14. 2026-10-18Upload 4 filesc79ffcc971 B
    Loading...
  15. 2026-10-17Upload eval_tinybench.py7e5a750971 B
    Loading...
  16. 2026-10-17Upload folder using huggingface_hub4f786ef971 B
    Loading...
  17. 2026-10-17Upload evaluate_jailbreak_CHaRS_AA.py2a0d8f7971 B
    Loading...
  18. 2026-10-17Upload 4 filesc3b51bc971 B
    Loading...
  19. 2026-10-17Upload 4 files21c2fd9971 B
    Loading...
  20. 2026-10-16Upload CHaRS evaluation results (part 3)e64a7fd971 B
    Loading...

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.

Related datasets same stage · ablate

lasrprobegen/refusal-activations
74.6K dl 1 likes 1.21 TB
Refusal Activations Dataset This dataset is now configured to load the full ~97k samples from jailbreak_mixed_100k.csv.
lu-christina/assistant-axis-vectors
14.4K dl 13 likes 1.2 GB
The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models This repository contains pre-computed axes...
stablellama/erotic-image-prompts
11.9K dl 66 likes 104.7 MB
Dataset Card for Erotic Image Prompts Dataset Description Dataset Summary Large language models (...
pandaman007/assistant-axis-abliteration-vectors
9.5K dl 24.5 GB
pandaman007/steered-persona-space
9.1K dl 539.6 MB
Steered Persona Space How activation steering reshapes a language model's character cloud. We apply activation-steering hooks a...
Undi95/orthogonal-activation-steering-TOXIC
7.2K dl 19 likes 2.1 MB
dementor-research/dementor-complete-experiment-results
5.1K dl 8.2 GB
Dementor complete experiment results Audited outputs for the configuration-defined Dementor completion campaign. Au...
Nafnlaus/ShrimpMoss_Chinese_Censorship_Abliteration
5K dl 8 likes 54.1 MB
ShrimpMoss (虾苔) is a dataset designed for the abliteration (https://github.com/FailSpy/abliterator) of Chinese government-imposed ...

Read further

Community best practices (ablation)
5 rules from KL-optimisation to layer-0 preservation.
M1 - Direct removal
The canonical Arditi-style ablation.
M3 - Layer-wise ablation
Selective abliteration limited to a band of layers.
Non-English abliteration communities
The Chinese layer-window heuristic and huihui-ai practice.

Metadata

License
mit

Download the dataset

0

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("sunnycloudhust/CHaRS-activations")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="sunnycloudhust/CHaRS-activations", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/sunnycloudhust/CHaRS-activations
Browse repository on Hugging Face ->
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Get the app Browse the catalog
Open in Abliteration