datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
UltraFeedback-preference-standardWe include all the possible comparisons following the Instruct-GPT. We use the fine-grained_score.
import os
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from datasets import load_dataset, DatasetDict
from transformers import AutoTokenizer
from tqdm import tqdm
from transformers import AutoTokenizer
ds = load_dataset("openbmb/UltraFeedback", split="train")
import itertools
data = []
for example in ds:
prompt = example['instruction']
responses = {}… See the full description on the dataset page: https://huggingface.co/datasets/RLHFlow/UltraFeedback-preference-standard.Nemotron-RLHF-GenRM-v1
Dataset Description:
This dataset is designed to train Generative Reward Models (GenRMs). It leverages reinforcement learning at scale to train accurate and robust GenRMs that generalize better than traditional Bradley-Terry models and reduce the risk of reward hacking.
The dataset is composed of:
Preference data focused on diverse domains
A synthetic safety blend
The data follows a "meta-prompt" structure where the model is instructed to act as an expert evaluation judge. For… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RLHF-GenRM-v1.Helpsteer-preference-standardzhihu_rlhf_3khh-rlhf-strength-cleaned
Dataset Card for hh-rlhf-strength-cleaned
Other Language Versions: English, 中文.
Dataset Description
In the paper titled "Secrets of RLHF in Large Language Models Part II: Reward Modeling" we measured the preference strength of each preference pair in the hh-rlhf dataset through model ensemble and annotated the valid set with GPT-4. In this repository, we provide:
Metadata of preference strength for both the training and valid sets.
GPT-4 annotations on the valid set.
We… See the full description on the dataset page: https://huggingface.co/datasets/OpenMOSS-Team/hh-rlhf-strength-cleaned.Argilla-Math-DPO-standardproofwriter-source
ProofWriter (The Source)
An unmodified copy of AI2's ProofWriter dataset (release V2020.12.3), re-hosted as datasets
configs for convenient loading. The records are faithful to the upstream release — the id-keyed JSON
is preserved as-is; typing and reasoning-graph extraction happen in later stages.
Each config is a {world}-depth-{n} shelf of the synthetic core (OWA/CWA × depths
0/1/2/3/5), split train/dev/test (dev kept as the corpus names it).
Source:… See the full description on the dataset page: https://huggingface.co/datasets/rlhf-and-friends/proofwriter-source.hh-rlhf
Dataset Card for HH-RLHF
Dataset Summary
This repository provides access to two different kinds of data:
Human preference data about helpfulness and harmlessness from Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. These data are meant to train preference (or reward) models for subsequent RLHF training. These data are not meant for supervised training of dialogue agents. Training dialogue agents on these data is likely to lead… See the full description on the dataset page: https://huggingface.co/datasets/polinaeterna/hh-rlhf.Capybara-distibalel-Filter-standardFrom argilla/Capybara-Preferences-Filtered. We process it into the standard format.
Orca-distibalel-standardFrom argilla/distilabel-intel-orca-dpo-pairs. We delete the pairs with the same score.
def process_example(example):
prompt = example['input']
chosen_response = example['chosen']
rejected_response = example['rejected']
if example['rating'] is None:
chosen_rating = None
rejected_rating = None
else:
if example['order'][0] == 'chosen':
chosen_rating = example['rating'][0]
rejected_rating = example['rating'][1]
elif… See the full description on the dataset page: https://huggingface.co/datasets/RLHFlow/Orca-distibalel-standard.proofwriter-mirror
ProofWriter (The Mirror)
A typed, content-faithful mirror of AI2's ProofWriter dataset (release V2020.12.3),
derived from proofwriter-source. The JSON encoding is cleaned up: the id-keyed dicts
(triple1, Q3, …) become lists of structs that keep their id, every atom representation
is parsed into a typed {subject, relation, object, polarity} triple, and the closed enums
(answer, strategy) are typed. The content stays faithful — nothing renamed, no rows
dropped — and the recursive… See the full description on the dataset page: https://huggingface.co/datasets/rlhf-and-friends/proofwriter-mirror.SHP-standardThe original dataset is the train set of stanfordnlp/SHP. We only keep the pair with ratio > 2.0 and take at most 5 pairs per prompt.
def filter_example(example):
prompt = example['history']
if example['labels'] == 0:
ratio = example['score_B'] * 1.0 / example['score_A']
elif example['labels'] == 1:
ratio = example['score_A'] * 1.0 / example['score_B']
if ratio > 2.0:
return True
else:
return False
import itertools
from collections… See the full description on the dataset page: https://huggingface.co/datasets/RLHFlow/SHP-standard.anthropic-hh-rlhf-conversations-with-toxicities
Dataset Card for "anthropic-hh-rlhf-conversations-with-toxicities"
More Information needed
Nemotron-Cascade-RL-RLHF
Dataset Description:
The Nemotron-Cascade-RL-RLHF dataset is designed for Reinforcement Learning from Human Feedback (RLHF) training. It contains prompts and associated metadata to support the development of language model alignment.
This dataset is ready for commercial use.
The dataset contains the following subset:
RLHF Training Data
This data contains 45,882 samples used for RLHF training. It includes prompts, data sources, and category information.
This dataset is a… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-Cascade-RL-RLHF.Nemotron-RLHF-GenRM-v1-prompt-only
Nemotron-RLHF-GenRM-v1-prompt-only
Prompt-only extraction from nvidia/Nemotron-RLHF-GenRM-v1.
Files:
prompts.csv: one prompt extraction record per source row. Records include
prompt, separated system_prompt, and structured tools when the source row
defines available tools. Nested values are JSON-encoded inside CSV cells.
summary.md: source row counts, extracted row counts, count deltas, and failed prompt counts.
null_or_empty_rows.md: row indexes where prompt extraction… See the full description on the dataset page: https://huggingface.co/datasets/jamesdborin/Nemotron-RLHF-GenRM-v1-prompt-only.korean_rlhf_content_filtered
Korean RLHF Content Filtered
Dataset Summary
This dataset is a cleaned, content-only derivative of:
Source dataset: jojo0217/korean_rlhf_dataset
Source URL: https://huggingface.co/datasets/jojo0217/korean_rlhf_dataset
Each row has a single content field suitable for LM pretraining/SFT-style text modeling.
Construction
Content construction rule
For each source row:
If input is empty: content = instruction + "\n" + output
If input is not empty:… See the full description on the dataset page: https://huggingface.co/datasets/alwaysgood/korean_rlhf_content_filtered.rlhf-irlinnoduel-rlhf-real-world-human-preferences-sample
Real-World Human Pairwise Preferences — Public Sample
📦 This is a free, public sample of a commercial dataset.
It contains 1,350 rows curated for inspection. The full dataset has 1.5 million
human pairwise-preference decisions.
Full dataset: https://huggingface.co/datasets/NordosoftOy/innoduel-rlhf
Request access / licensing: see § Access to the full dataset — contact kari.nieminen@nordo.fi.
Use this sample to evaluate the data's quality, structure and… See the full description on the dataset page: https://huggingface.co/datasets/NordosoftOy/innoduel-rlhf-real-world-human-preferences-sample.Anis-RLHFKusursuz_Ajan_Turkish_RLHF
Kusursuz Ajan (Flawless Agent) - Turkish RLHF Dataset (50-Sample Preview)
🌍 (English below / İngilizce metin aşağıdadır)
Bu repository, Türkçe yapay zeka eğitimleri için hazırlanan devasa ve ultra yüksek kaliteli Kusursuz Ajan veri setinin 50 kayıtlık bir vitrin örneğidir.
Örnek veri setini indirerek, modellerinizde RLHF/DPO ince ayarlarını test edebilir, veri setimizin barındırdığı zengin metadata (Intent, Tone, Subtext) yapısını inceleyebilirsiniz.
1518 kayıtlık… See the full description on the dataset page: https://huggingface.co/datasets/abidinesen/Kusursuz_Ajan_Turkish_RLHF.DPO-hh-rlhf
Dataset Card for DPO-hh-rlhf
Reformatted from Anthropic/hh-rlhf dataset.
The LION-series are trained using an empirically optimized pipeline that consists of three stages: SFT, DPO, and online preference learning (online DPO). We find simple techniques such as sequence packing, loss masking in SFT, increasing the preference dataset size in DPO, and online DPO training can significantly improve the performance of language models. Our best models (the LION-series) exceed the… See the full description on the dataset page: https://huggingface.co/datasets/Columbia-NLP/DPO-hh-rlhf.anthropic-rlhf-human-values-dpohh-rlhf_with_features_flan_t5_large
Dataset Card for "hh-rlhf_with_features_flan_t5_large"
More Information needed
hh-rlhf_with_features
Dataset Card for "hh-rlhf_with_features"
More Information needed
hh-rlhf_with_features_flan_t5_large-no_eos
Dataset Card for "hh-rlhf_with_features_flan_t5_large"
More Information needed
hh-rlhf-constitutional-sensitivities-of-pms
Constitutional-Sensitivities of HH-RLHF dataset
Description
A subset of query-response pairs from the Anthropic hh-rlhf dataset including:
Original query and response
Perturbed response using gpt-40-mini according a given constitutional principle from a Collective constitutional AI dataset using a critique and revision step.
Resources for this project are available here: https://github.com/DouwMarx/constitutional_sensitivities_of_preference_models
WiredBrain-RLHF
WiredBrain-RLHF: Production-Grade Data for High-Integrity AI Alignment
Overview
WiredBrain-RLHF transforms the foundational Anthropic HH-RLHF dataset (148K conversations) into a production-grade training resource through systematic data enrichment. Engineered by Shubham Dev (Department of Computer Science, Jaypee University of Information Technology, India), this dataset powers high-stakes AI alignment research requiring factual integrity and entity preservation.… See the full description on the dataset page: https://huggingface.co/datasets/pheonix-delta/WiredBrain-RLHF.ArmoRM-Multi-Objective-Data-v0.1Qwen3-8B-BS16-RLHF-PRM-GPQArlhf_synthetic_generalized