datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
JSONSchemaBench
JSONSchemaBench
JSONSchemaBench is a benchmark of real-world JSON schemas designed to evaluate structured output generation for Large Language Models (LLMs). It contains approximately 10,000 JSON schemas, capturing diverse constraints and complexities.
import datasets
from datasets import load_dataset
def main():
# Inspect the available subsets of the datasetall_subsets = datasets.get_dataset_config_names("epfl-dlab/JSONSchemaBench")
print("Available subsets:"… See the full description on the dataset page: https://huggingface.co/datasets/epfl-dlab/JSONSchemaBench.corpus-1T-manifest
SPP Corpus 1T Manifest
The selection manifest for the ~1.0T-token pretraining corpus used in
Synthetic Persona Pretraining (SPP): Alignment from Token Zero.
The corpus is a seeded subsample of allenai/dolma3_mix-6T.
Rather than redistribute ~2.6 TB of text that is already public, this dataset
publishes the selection decisions keyed by upstream document id, so the corpus
can be reconstructed exactly by replaying against upstream.
📄 Reflections + text for the annotated half:… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/corpus-1T-manifest.reflection-50m
SPP Reflection 50M
The 51.4M-document reflection set from Synthetic Persona Pretraining (SPP):
Alignment from Token Zero — the production half-corpus run, and the dataset the
released models were actually trained on.
🔬 Small sample (same format): dlab-spp/reflection-sample-2k
📉 Earlier 10M run: dlab-spp/reflection-10m
🧾 Safety scores for the full 1T corpus: dlab-spp/safety-classifications
Each row pairs a source document with two generated constitution reflections — a… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/reflection-50m.DLR-Web
DLR-Web: Multidisciplinary Reasoning Dataset from Web Corpus [Project Page]
This repository releases the Design-Logic-Reasoning-Web (DLR-Web) dataset from the paper DESIGNER: Design-Logic-Guided Multidisciplinary Data Synthesis for LLM Reasoning (ICLR 2026).
Field definitions
original_document: web-sourced raw document text, further filtered from FineFineWeb; thanks to the FineFineWeb authors and maintainers for providing this resource
design_logic: Design Logic in… See the full description on the dataset page: https://huggingface.co/datasets/Attention1115/DLR-Web.DLLM-Planing-Task
DLLM-Planning-Task
Benchmark datasets for evaluating planning capabilities of Diffusion Language Models (DLLMs).
Dataset Description
This dataset contains multiple planning and combinatorial reasoning tasks designed to evaluate discrete diffusion language models. Each task has train/test splits in either CSV or JSONL format.
Tasks
Task
Format
Description
Sudoku
CSV
9x9 Sudoku puzzles. Columns: quizzes, solutions.
Path Finding (path-2-6)… See the full description on the dataset page: https://huggingface.co/datasets/zeyuzy/DLLM-Planing-Task.reflection-10m
SPP Reflection 10M
The full ~10M-document reflection set from Synthetic Persona Pretraining (SPP):
Alignment from Token Zero.
📝 Read the post: Synthetic Persona Pretraining: Alignment from Token Zero
🔬 Small sample (same format): dlab-spp/reflection-sample-2k — a 2,000-row sample drawn from this set, for quick inspection.
Each row pairs a pretraining document with a synthetic, value-laden reflection
generated for it: a short first-person (and third-person) moral reflection… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/reflection-10m.DLT-Tweets
DLT-Tweets
[Paper] •
[Code]
Dataset Description
Dataset Summary
DLT-Tweets is a large-scale corpus of social media posts related to Distributed Ledger Technology (DLT). This dataset is part of the larger DLT-Corpus collection, designed to support NLP research, social computing studies, and public discourse analysis in the DLT domain. It was introduced in the paper DLT-Corpus: A Large-Scale Text Collection for the Distributed Ledger Technology Domain.… See the full description on the dataset page: https://huggingface.co/datasets/ExponentialScience/DLT-Tweets.dlsite-jp-v1
puwaer/dlsite-jp-v1
This dataset consists of text extracted exclusively in Japanese from dlsite.com and is structured as JSON files. The files are categorized based on the type of URL.
このデータセットは、dlsite.comより日本語データのみを抽出したテキストで、jsonファイルで構成されます。
urlの種類によってファイル分けされています。
llaza-20B
Llaza Mixture 20B
This dataset is a 20B-token pretraining subset built for zip2zip language-model pretraining.
It is derived from the full Llaza mixture, which is byte-balanced across four top-level domains:
Domain
Source
Target byte ratio
General
HuggingFaceFW/fineweb-edu, sample-100BT
50%
Code
bigcode/the-stack-dedup
20%
Math
HuggingFaceTB/finemath, finemath-3plus
10%
Multilingual
epfml/FineWeb2-HQ, 20 language subsets
20%
The subset was created from remixed… See the full description on the dataset page: https://huggingface.co/datasets/epfl-dlab/llaza-20B.Sudoku_DLM_Reasoning
Sudoku DLM Reasoning
stwistzz/Sudoku_DLM_Reasoning is a deterministic 9x9 Sudoku benchmark for studying masked
diffusion language models, depth, and iterative decoding. Version 1.1.0
trains only on original, r0, and r1_4; r5_19 is held out for adjacent
difficulty extrapolation.
Version 1.1.0 changes only the deterministic selection seed to 0
relative to v1.0.2. The raw CSV files, sources, bucket definitions, split quotas,
schema, and validation rules are unchanged; the four… See the full description on the dataset page: https://huggingface.co/datasets/twistshan/Sudoku_DLM_Reasoning.sp-sft-normal-300k
model-raising-pbsft-instruct-300k
A constitution-aware paired SFT dataset of 300,000 general-purpose (WildChat) instruct
prompts. Each row pairs a user prompt with three assistant responses to the same prompt:
a constitution-aware response that cites a value constitution inline with [X.Y] markers,
a constitution-invisible rendering of that same response (no markers, no constitution vocabulary), and
the original response that shipped with the prompt in WildChat-1M.
It is part… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/sp-sft-normal-300k.llaza-200B
Llaza Mixture Full (200B)
This dataset is the full Llaza pretraining-data mixture for zip2zip language-model pretraining.
It combines general web text, code, math, and multilingual web text with byte-based top-level mixture ratios.
Domain
Source
Target byte ratio
General
HuggingFaceFW/fineweb-edu, sample-100BT
50%
Code
bigcode/the-stack-dedup
20%
Math
HuggingFaceTB/finemath, finemath-3plus
10%
Multilingual
epfml/FineWeb2-HQ, 20 language subsets
20%… See the full description on the dataset page: https://huggingface.co/datasets/epfl-dlab/llaza-200B.sp-sft-safety-180k
model-raising-pbsft-safety-180k
A constitution-aware paired SFT dataset of 182,688 safety-relevant prompts. Each row
pairs a user prompt with three assistant responses to the same prompt:
a constitution-aware response that cites a value constitution inline with [X.Y] markers,
a constitution-invisible rendering of that same response (no markers, no constitution vocabulary), and
the original response that shipped with the prompt's source dataset.
It is part of the Synthetic… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/sp-sft-safety-180k.BESPOKE
BESPOKE: Benchmark for Search-Augmented Large Language Model Personalization via Diagnostic Feedback
Paper | Project Page | GitHub Repository
BESPOKE is a realistic benchmark specifically designed for evaluating personalization in search-augmented LLMs.
Data Fields
Query Table (queries/query_table.parquet)
Key
Type
Description
user
str
User identifier
query-id
str
Query identifier (canonical hyphen variant)
query
str
Natural-language query… See the full description on the dataset page: https://huggingface.co/datasets/yonsei-dli/BESPOKE.fragility-moral-judgment-llms
Fragility of Moral Judgment in Large Language Models
Companion dataset for the FAccT paper Fragility of Moral Judgment in Large Language Models by Tom van Nuenen. Contains the moral dilemmas, community labels, and per-model verdicts (with explanations and reasoning traces) used in the study.
The paper investigates how stable LLM moral judgments are under minimal, morally-irrelevant perturbations of the same dilemma, and whether protocols and reasoning chains improve or worsen… See the full description on the dataset page: https://huggingface.co/datasets/ucberkeley-dlab/fragility-moral-judgment-llms.gordon-ramsay-dl-instruct
Gordon Ramsay Deep Learning Instruct Dataset
A preference alignment dataset for training language models to answer Deep Learning questions in the style of Gordon Ramsay. Each example contains a student question, a polite educational answer, and an aggressive Gordon Ramsay-style answer — both technically correct.
Dataset Description
This dataset was created for the MSc in Artificial Intelligence & Deep Learning (AIDL_B_CS01 — NLP with Deep Learning) at the University of… See the full description on the dataset page: https://huggingface.co/datasets/antonisbast/gordon-ramsay-dl-instruct.DLT-Scientific-Literature
DLT-Scientific-Literature
Paper | GitHub
Dataset Description
Dataset Summary
DLT-Scientific-Literature is a specialized corpus of academic publications focused on Distributed Ledger Technology (DLT). This dataset is part of the larger DLT-Corpus collection, designed to support NLP research, language model development, and innovation studies in the DLT domain.
The dataset contains 37,440 scientific documents with 564 million tokens, spanning publications from… See the full description on the dataset page: https://huggingface.co/datasets/ExponentialScience/DLT-Scientific-Literature.UltraGrain
UltraBeacon
UltraBeacon is a high-quality English question-answering dataset prepared by Dl26 for text-generation and chat-style model training.
The dataset is converted from TriviaQA into a consistent assistant format with prompt, response, and messages fields. It is intended for supervised fine-tuning of general assistant models that need strong question answering behavior.
Dataset Details
Property
Value
Dataset name
UltraBeacon
Developer
Dl26… See the full description on the dataset page: https://huggingface.co/datasets/Dl26/UltraGrain.clara-stage2-data
Clara Stage 2 Training Data
Training data for Clara Stage 2 (Compression Instruction Tuning).
Dataset Description
This dataset contains high-quality QA pairs with single documents for training Clara's decoder adapter to generate answers from compressed document representations.
Data Format
Each record contains:
question: The query/question
answer: Gold answer
docs: List containing 1 document
meta: Source description
metadata: Additional metadata (repo, scope… See the full description on the dataset page: https://huggingface.co/datasets/dl3239491/clara-stage2-data.reflection-sample-2k
SPP Reflection 2k Sample
A 2,000-row sample (seed 42) of dlab-spp/reflection-10m,
in the identical format, for quick inspection of the data from
Synthetic Persona Pretraining (SPP): Alignment from Token Zero.
📝 Read the post: Synthetic Persona Pretraining: Alignment from Token Zero
📦 Full dataset: dlab-spp/reflection-10m (~10M documents).
Each row pairs a pretraining document with a synthetic, value-laden reflection
(first- and third-person) grounded in a value constitution.… See the full description on the dataset page: https://huggingface.co/datasets/dlab-spp/reflection-sample-2k.UltraBeacon
UltraBeacon
UltraBeacon is a high-quality English question-answering dataset prepared by Dl26 for text-generation and chat-style model training.
The dataset is converted from TriviaQA into a consistent assistant format with prompt, response, and messages fields. It is intended for supervised fine-tuning of general assistant models that need strong question answering behavior.
Dataset Details
Property
Value
Dataset name
UltraBeacon
Developer
Dl26… See the full description on the dataset page: https://huggingface.co/datasets/Dl26/UltraBeacon.UltraAtlas
UltraAtlas
UltraAtlas is a high-quality English question-answering dataset prepared by Dl26 for text-generation and chat-style model training.
The dataset is converted from SQuAD v2 into a consistent assistant format with prompt, response, and messages fields. It is intended for supervised fine-tuning of general assistant models that need strong question answering behavior.
Dataset Details
Property
Value
Dataset name
UltraAtlas
Developer
Dl26
Source… See the full description on the dataset page: https://huggingface.co/datasets/Dl26/UltraAtlas.dltha_reasoning_v1.jsonl
DLTHA Reasoning Dataset v1
Description
This dataset is the first release from DLTHA Labs, focused on enhancing the logical reasoning and step-by-step problem-solving capabilities of Large Language Models (LLMs).
At DLTHA, we believe that the path to AGI (Artificial General Intelligence) requires high-fidelity synthetic data that mimics complex human thought processes. This dataset provides a structured "Chain-of-Thought" (CoT) format for technical and logical queries.… See the full description on the dataset page: https://huggingface.co/datasets/Dltha-Labs/dltha_reasoning_v1.jsonl.AIRC-DL-Intro
Dataset Card for Dataset Name
The dataset provides educator-generated multiple-choice quiz questions from lectures in real-world classrooms in Computer Science.
This is an subset containing the following course:
DL-Intro: an undergraduate-level course about various basic concepts and topics in Deep Learning.
Dataset Details
Uses
from datasets import load_dataset
data = load_dataset('mengxiayu/AIRC-DL-Intro', split='test')
print(data[0])
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/mengxiayu/AIRC-DL-Intro.neocortirrhea-lexicon
Dataset Card: Neocortirrhea Lexicon Entry
Summary
This dataset entry defines and contextualizes the psychological, neurological, and somatic neologism Neocortirrhea.
Dataset Structure
JSON Lines Representation (data.jsonl)
{
"term": "Neocortirrhea",
"part_of_speech": "noun",
"phonetic": "/ˌniː.oʊˌkɔːr.tɪˈriː.ə/",
"etymology": "Neocortex (higher-order cognitive processing) + -rrhea (Greek rhoia: abnormal/excessive flow or… See the full description on the dataset page: https://huggingface.co/datasets/dlewicki/neocortirrhea-lexicon.fso-dataset-dlm
fso-dataset-dlm
Dataset generated by FusionX ModelsX.
Format: sharegpt
Samples: 806
Created: 2025-12-26T11:18:39.011263+00:00
Description
A dataset to fine tune a DLM on fso terms based on Financial Services Domain.