README current version from Hugging Face

dataset_info:

config_name: deepseek-ai-DeepSeek-R1-Distill-Llama-8B_15-llamascope-slimpj-res-32k_deepseek-deepseek-r1-distill-llama-70b
features:
- name: index
  dtype: int64
- name: description
  dtype: 'null'
- name: pos_str
  list: string
- name: neg_str
  list: string
- name: explanations
  list: string
- name: diff
  dtype: float64
- name: p_value
  dtype: float64
- name: pos_mean
  dtype: float64
- name: neg_mean
  dtype: float64
- name: pos_median
  dtype: float64
- name: neg_median
  dtype: float64
  splits:
- name: train
  num_bytes: 4280
  num_examples: 13
  download_size: 9359
  dataset_size: 4280
config_name: gemma-2-9b_9-gemmascope-res-16k_google-gemma-2-9b-it
features:
- name: index
  dtype: int64
- name: description
  dtype: 'null'
- name: pos_str
  list: string
- name: neg_str
  list: string
- name: explanations
  list: string
- name: diff
  dtype: float64
- name: p_value
  dtype: float64
- name: pos_mean
  dtype: float64
- name: neg_mean
  dtype: float64
- name: pos_median
  dtype: float64
- name: neg_median
  dtype: float64
  splits:
- name: train
  num_bytes: 42849
  num_examples: 105
  download_size: 34695
  dataset_size: 42849
config_name: meta-llama-Meta-Llama-3-8B-Instruct_25-res-jh_meta-llama-llama-3-8b-instruct
features:
- name: index
  dtype: int64
- name: description
  dtype: 'null'
- name: pos_str
  list: string
- name: neg_str
  list: string
- name: explanations
  list: string
- name: diff
  dtype: float64
- name: p_value
  dtype: float64
- name: pos_mean
  dtype: float64
- name: neg_mean
  dtype: float64
- name: pos_median
  dtype: float64
- name: neg_median
  dtype: float64
  splits:
- name: train
  num_bytes: 11777
  num_examples: 56
  download_size: 14657
  dataset_size: 11777
  configs:
config_name: deepseek-ai-DeepSeek-R1-Distill-Llama-8B_15-llamascope-slimpj-res-32k_deepseek-deepseek-r1-distill-llama-70b
data_files:
- split: train
  path: sae_model=deepseek-ai-DeepSeek-R1-Distill-Llama-8B/sourceset=15-llamascope-slimpj-res-32k/refusals_model=deepseek-deepseek-r1-distill-llama-70b/train-*
config_name: gemma-2-9b_9-gemmascope-res-16k_google-gemma-2-9b-it
data_files:
- split: train
  path: sae_model=gemma-2-9b/sourceset=9-gemmascope-res-16k/refusals_model=google-gemma-2-9b-it/train-*
config_name: meta-llama-Meta-Llama-3-8B-Instruct_25-res-jh_meta-llama-llama-3-8b-instruct
data_files:
- split: train
  path: sae_model=meta-llama-Meta-Llama-3-8B-Instruct/sourceset=25-res-jh/refusals_model=meta-llama-llama-3-8b-instruct/train-*

README history 4 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.

2026-01-11Upload features for meta-llama/Meta-Llama-3-8B-Instruct / 25-res-jh / deepsee...71f5be02.6 KB
Loading...
2026-01-11Upload features for gemma-2-9b / 9-gemmascope-res-16k / deepseek/deepseek-r1-...24b65781.8 KB
Loading...
2026-01-11Upload features for deepseek-ai/DeepSeek-R1-Distill-Llama-8B / 15-llamascope-...f841c7d1004 B
Loading...
2026-01-11initial commit1d539440 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.

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.

2026-01-12[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

DIR sae_model=deepseek-ai-DeepSeek-R1-Distill-Llama-8B DIR sae_model=gemma-2-9b DIR sae_model=meta-llama-Meta-Llama-3-8B-Instruct FILE .gitattributes FILE README.md

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

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("djain95/sae-jailbreaks-features")

Raw snapshot (Python)

from huggingface_hub import snapshot_download
snapshot_download(repo_id="djain95/sae-jailbreaks-features", repo_type="dataset")

git clone (requires git-lfs)

git clone https://huggingface.co/datasets/djain95/sae-jailbreaks-features

Browse repository on Hugging Face ->

djain95/sae-jailbreaks-features

Tags