datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
toxic-chat
Update
[01/31/2024] We update the OpenAI Moderation API results for ToxicChat (0124) based on their updated moderation model on on Jan 25, 2024.[01/28/2024] We release an official T5-Large model trained on ToxicChat (toxicchat0124). Go and check it for you baseline comparision![01/19/2024] We have a new version of ToxicChat (toxicchat0124)!
Content
This dataset contains toxicity annotations on 10K user prompts collected from the Vicuna online demo.
We utilize a human-AI… See the full description on the dataset page: https://huggingface.co/datasets/lmsys/toxic-chat.brain-lm-alignment-ds002236
Brain–language-model alignment: ds002236 (whole-brain)
Lytle et al. 2020 — orthographic, phonological and semantic word processing in school-aged children (8.7–15.5), auditory and visual.
Paper: https://pubmed.ncbi.nlm.nih.gov/31956678/
Data: https://openneuro.org/datasets/ds002236/versions/1.0.1
Generated: 2026-09-24
Pipeline: https://github.com/suchirsalhan/cdl-representations-brains-babylms
Read this first: does the measurement work?
Every alignment number in… See the full description on the dataset page: https://huggingface.co/datasets/BrainAlign/brain-lm-alignment-ds002236.brain-lm-alignment-ds001894
Brain–language-model alignment: ds001894 (whole-brain)
Lytle et al. 2019 — longitudinal word-level phonological processing in children scanned twice, at roughly 10 and 12 years old.
Paper: https://www.nature.com/articles/s41597-019-0338-5
Data: https://openneuro.org/datasets/ds001894/versions/1.4.2
Generated: 2026-09-24
Pipeline: https://github.com/suchirsalhan/cdl-representations-brains-babylms
Read this first: does the measurement work?
Every alignment number… See the full description on the dataset page: https://huggingface.co/datasets/BrainAlign/brain-lm-alignment-ds001894.arena-human-preference-55kDataset for Kaggle competition on predicting human preference on Chatbot Arena battles.
The training dataset includes over 55,000 real-world user and LLM conversations and user preferences across over 70 state-of-the-art LLMs, such as GPT-4, Claude 2, Llama 2, Gemini, and Mistral models.
Each sample represents a battle consisting of 2 LLMs which answer the same question, with a user label of either prefer model A, prefer model B, tie, or tie (both bad).
Citation
Please cite the… See the full description on the dataset page: https://huggingface.co/datasets/lmarena-ai/arena-human-preference-55k.LM-SimBench
LM-SimBench
Dataset Description
LM-SimBench is a large-scale training-performance profiling dataset for large language models. The dataset is collected from training runs based on the MindSpeed-LLM framework and the Ascend NPU development stack, covering multiple model families, context lengths, and distributed parallel configurations.
Each model is sampled under feasible combinations of data parallelism (DP), tensor parallelism (TP), pipeline parallelism (PP), context… See the full description on the dataset page: https://huggingface.co/datasets/anonymous-neurips26-ljasd/LM-SimBench.lm-pragmatics
Citation
@misc{hu2023finegrainedcomparisonpragmaticlanguage, title={A fine-grained comparison of pragmatic language understanding in humans and language models}, author={Jennifer Hu and Sammy Floyd and Olessia Jouravlev and Evelina Fedorenko and Edward Gibson}, year={2023}, eprint={2212.06801}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2212.06801},}
lmsys_chatbot_arena_conversationsdatasource: https://colab.research.google.com/drive/1KdwokPjirkTmpO_P1WByFNFiqxWQquwH
brain-lm-alignment-ds003604
Brain-LM alignment: ds003604
Representational-similarity alignment between language-model hidden states and
child fMRI RDMs for ds003604 (children ages 5/7/9, auditory).
Tasks: Sem, Phon, Gram, Plaus Sessions: ses-5, ses-7, ses-9 Cells: 12
Models: 14 families (5 real + 9 PARC noise-seed baselines)
Rows: 1848 (family x checkpoint x task x session)
Generated: 2026-08-29
Headline: no model is distinguishable from a random seed
Alignment is computed as Spearman… See the full description on the dataset page: https://huggingface.co/datasets/BrainAlign/brain-lm-alignment-ds003604.iclr2026-lm-logprobs
LM Log-Probabilities for Value Bias Analysis
Next-token log-probability distributions from 12 language models across 54 prompts, used in the paper:
Reward Models Inherit Value Biases from Pretraining
Brian Christian, Jessica A.F. Thompson, Elle, Vincent Adam, Hannah Rose Kirk, Christopher Summerfield, Tsvetomira Dumbalska (ICLR 2026)
Part of the Oxford-HIPlab collection for this paper.
Dataset description
Each CSV contains the full next-token log-probability… See the full description on the dataset page: https://huggingface.co/datasets/Oxford-HIPlab/iclr2026-lm-logprobs.tweet-annotation-sensitivity-2
Tweet Annotation Sensitivity Experiment 2: Annotations in Five Experimental Conditions
Attention: This repository contains cases that might be offensive or upsetting. We do not support the views expressed in these hateful posts.
Description
The dataset contains tweet data annotations of hate speech (HS) and offensive language (OL) in five experimental conditions. The tweet data was sampled from the corpus created by Davidson et al. (2017). We selected 3,000 Tweets for our… See the full description on the dataset page: https://huggingface.co/datasets/soda-lmu/tweet-annotation-sensitivity-2.humor_trainannotations_creators: []
language_creators: []
languages: []
licenses: []
multilinguality: []
pretty_name: humor_train
size_categories: []
source_datasets: []
task_categories: []
task_ids: []
lm6-movies-reviews-aspects
IMDB Reviews (Aspect Based Formatted)
Overview
IMDB Reviews (Aspect Based Formatted) is a specialized dataset designed for text classification tasks that involve identifying and categorizing specific aspects of movie reviews. The dataset focuses on extracting and labeling various elements of filmmaking, such as cinematography, story, characters, direction, and unique concepts, from user reviews on IMDB.
The dataset can be used to develop and train models for aspect-based… See the full description on the dataset page: https://huggingface.co/datasets/Lowerated/lm6-movies-reviews-aspects.bilmecebench-lm-evaluation-harnesslmsys-kaggle
LMSYS/Kaggle Preference Data
This repository contains 91,667 rows in the default/train split and is viewable through the Hugging Face Dataset Viewer.
Important Provenance Notice
The repository currently does not include a source manifest, processing script, or exact upstream revision. Its name and schema are consistent with LMSYS/Kaggle preference data, but the row count should not be interpreted as proof that it is an unchanged copy of any single official… See the full description on the dataset page: https://huggingface.co/datasets/keithtyser/lmsys-kaggle.tweet-annotation-sensitivity-1
Tweet Annotation Sensitivity Experiment 1: Annotation in Six Experimental Conditions
Attention: This repository contains cases that might be offensive or upsetting. We do not support the views expressed in these hateful posts.
Description
We drew a stratified sample of 20 tweets, that were pre-annotated in a study by Davidson et al. (2017) for Hate Speech / Offensive Language / Neither. The stratification was done with respect to majority-voted class and level of… See the full description on the dataset page: https://huggingface.co/datasets/soda-lmu/tweet-annotation-sensitivity-1.lmsys-chat-1m-GEO-2-25
📊 Understanding User Intent in Chatbot Conversations
This dataset, sections the LMSYS-Chat-1M dataset into GEO and Marketing relevant categories to help maketers access real life prompts and chatlogs that are relevant to them.
💡 Why I'm doing this
GEO (generative engine optimization) is a new marketing discipline, similar to SEO, that aims to measure and optimize the responses of LLMs and LLM-powered apps for marketing purposes.
The question of "search volume" in LLMs… See the full description on the dataset page: https://huggingface.co/datasets/GPAWeb/lmsys-chat-1m-GEO-2-25.LMS_DSlmsys-train