Get the app
ablate 0 0 0

xzAscC/TDNV-results

published by @xzAscC

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
169
Last 30 days
169
Likes
-
Size
40.2 MB
Created on HF
2026-10-07
Age
14 days ago

Description

TDNV results

Results and figures for layerwise TDNV (Task-Distance Normalized Variance) of last-token hidden states

on contrastive (steering) datasets. Code: https://github.com/xzAscC/TDNV.

Files

outputs/<model>/<concept>/metrics.json: per-layer TDNV for one model and one concept.

Keys: layer, tdnv, tdnv_shuffled, within_var, between_dist2, probe_acc, config.

Layer 0 is the embedding output.

outputs/<model>/steer_<concept>/metrics.json: activation-steering… See the full description on the dataset page: https://huggingface.co/datasets/xzAscC/TDNV-results.

Tags

license:mitregion:usinterpretabilityactivation-steering

README current version from Hugging Face

license: mit
pretty_name: TDNV results
tags:

  • interpretability
  • activation-steering

TDNV results

Results and figures for layerwise TDNV (Task-Distance Normalized Variance) of last-token hidden states
on contrastive (steering) datasets. Code: https://github.com/xzAscC/TDNV.

Files

  • outputs/<model>/<concept>/metrics.json: per-layer TDNV for one model and one concept.
    Keys: layer, tdnv, tdnv_shuffled, within_var, between_dist2, probe_acc, config.
    Layer 0 is the embedding output.
  • outputs/<model>/steer_<concept>/metrics.json: activation-steering results for one model and one concept.
  • outputs/<model>/<concept>/tdnv.png: quick per-run plot (only for some runs).
  • results/u_shape_summary.csv: down-then-up statistics per model and concept.
  • figs/: figures made from outputs/.

Model directory names use __ in place of / (for example Qwen__Qwen3-8B is Qwen/Qwen3-8B).

Redraw the figures

git clone https://github.com/xzAscC/TDNV && cd TDNV
hf download xzAscC/TDNV-results --repo-type dataset --include "outputs/*" --local-dir .
uv sync
uv run python notebooks/plot_sweep.py
uv run python notebooks/plot_steering.py

README history 8 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. 2026-10-19Sync steering results and current paper figures97818cf1.2 KB
    Loading...
  2. 2026-10-11Every-layer steering (steer_all), resid/amid runs, ICL runs; one-alpha-per-pa...25a14e11.2 KB
    Loading...
  3. 2026-10-11Every-layer steering (steer_all), resid/amid runs, ICL runs; one-alpha-per-pa...bb1116f1.2 KB
    Loading...
  4. 2026-10-11Every-layer steering (steer_all), resid/amid runs, ICL runs; one-alpha-per-pa...72fc8451.2 KB
    Loading...
  5. 2026-10-11Every-layer steering (steer_all), resid/amid runs, ICL runs; one-alpha-per-pa...ddc51d61.2 KB
    Loading...
  6. 2026-10-08ICL TDNV: 8 models x 12 concepts x K=1,5,10,15fd31b9a1.2 KB
    Loading...
  7. 2026-10-07Add TDNV outputs, figures and summaryf579a101.2 KB
    Loading...
  8. 2026-10-07initial commit488bce80 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
sunnycloudhust/CHaRS-activations
5.3K dl 2 likes 63.8 GB
CHaRS: Concept Heterogeneity-Aware Representation Steering This repository contains implementation of CHaRS (Concept Heterogene...
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...

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("xzAscC/TDNV-results")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="xzAscC/TDNV-results", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/xzAscC/TDNV-results
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