Get the app
extract 0 0 0

LRudL/anthropic_hh_modified

published by @LRudL

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
789
Last 30 days
22
Likes
-
Size
31.6 MB
Created on HF
2023-03-21
Age
3 years ago

Description

Dataset Card for "anthropic_hh_modified"

Total copy of Anthropic/hh-rlhf (all credit and rights to the authors of that), just with some modifications to the format so that it can be used with the eleuther-elk repository.

Changes:

rename column "chosen" to "choice0" and "rejected" to "choice1"

randomly flip the entry in column choice0 and choice1 for half of the entries

create a ClassLabel column "label" that stores an integer 0 or 1, corresponding to which of choice0 or choice1… See the full description on the dataset page: https://huggingface.co/datasets/LRudL/anthropic_hh_modified.

Tags

size_categories:10K<n<100Kformat:parquetmodality:textlibrary:datasetslibrary:pandaslibrary:mlcroissantlibrary:polarsregion:us

README current version from Hugging Face

dataset_info:
features:

  • name: choice0
    dtype: string
  • name: choice1
    dtype: string
  • name: label
    dtype:
    class_label:
    names:
    '0': '0'
    '1': '1'

splits:

  • name: train
    num_bytes: 56635938
    num_examples: 42537
  • name: test
    num_bytes: 3195756
    num_examples: 2312
    download_size: 33135346
    dataset_size: 59831694

Dataset Card for "anthropic_hh_modified"

Total copy of Anthropic/hh-rlhf (all credit and rights to the authors of that), just with some modifications to the format so that it can be used with the eleuther-elk repository.

Changes:

  • rename column "chosen" to "choice0" and "rejected" to "choice1"
  • randomly flip the entry in column choice0 and choice1 for half of the entries
  • create a ClassLabel column "label" that stores an integer 0 or 1, corresponding to which of choice0 or choice1 was preferred by the human.

More Information needed

README history 5 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. 2023-03-21Update README.md969d59b1.1 KB
    Loading...
  2. 2023-03-21Upload README.md with huggingface_hubf1a49fc566 B
    Loading...
  3. 2023-03-21Upload data/test-00000-of-00001-428caf0072ed0c18.parquet with huggingface_hub6500e060 B
    Loading...
  4. 2023-03-21Upload data/train-00000-of-00001-04e6404dfaf0143d.parquet with huggingface_hub8cd64c80 B
    Loading...
  5. 2023-03-21initial commite4013b50 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 · extract

Anthropic/hh-rlhf
2.1M dl 2442 likes 90.4 MB
Dataset Card for HH-RLHF Dataset Summary This repository provides access to two different kinds of data: Human ...
djain95/sae-jailbreaks-cache
468.6K dl 1 likes 30.6 GB
mosaicml/dolly_hhrlhf
309.5K dl 113 likes 23.7 MB
Dataset Card for "dolly_hhrlhf" This dataset is a combination of Databrick's dolly-15k dataset and a filtered subset of Anthrop...
Dahoas/full-hh-rlhf
266.3K dl 104 likes 129.8 MB
Dataset Card for "full-hh-rlhf" Anthropic's HH dataset reformatted into prompt, chosen, rejected samples.
mlabonne/harmful_behaviors
198.2K dl 169 likes 23 KB
mlabonne/harmless_alpaca
190.2K dl 51 likes 1.2 MB
walledai/AdvBench
176.3K dl 123 likes 38 KB
Dataset Card for AdvBench Paper: Universal and Transferable Adversarial Attacks on Aligned Language Models Data: AdvBench Datas...
JailbreakV-28K/JailBreakV-28k
116.4K dl 80 likes 288.8 MB
⛓‍💥 JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks 🌐 ...

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 data 2 files 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("LRudL/anthropic_hh_modified")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="LRudL/anthropic_hh_modified", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/LRudL/anthropic_hh_modified
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