datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SimpleSafetyTestsSimpleStories
📘📕 SimpleStories 📙📗
SimpleStories is a dataset of >2 million model-generated short stories. It was made to train small, interpretable language models on it. The generation process is open-source: To see how the dataset was generated, or to generate some stories yourself, head over to this repository.
If you'd like to commission other languages or story formats, feel free to send mail.
When using SimpleStories in your work, please cite the SimpleStories paper:… See the full description on the dataset page: https://huggingface.co/datasets/SimpleStories/SimpleStories.simple_wikipedia
simple wikipedia
the 'simple' split of Wikipedia, from Sept 1 2023. The train split contains about 65M tokens,
Pulled via:
dataset = load_dataset(
"wikipedia", language="simple", date="20230901", beam_runner="DirectRunner"
)
stats
train split
general info
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 226242 entries, 0 to 226241
Data columns (total 4 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 id… See the full description on the dataset page: https://huggingface.co/datasets/pszemraj/simple_wikipedia.simple-math
Simple Math: 2+2=4 -1=3 (LoLo: Learning Only Logical Operations)
Just like my teacher gave me homework, i thought maybe we can also add some of these basics on the trainings of our models.
It was created with very simple code that is in the repo, if you add more complex operations and so.. please share the code :D thank you
Current Code Version: 20240127.fblgit (A modification over @win10 for progressive and DPO operation)
Does it Works?
34BEAGLES… See the full description on the dataset page: https://huggingface.co/datasets/fblgit/simple-math.simpleqa-verified-multilingual
SimpleQA Verified Multilingual
Multilingual translations of SimpleQA Verified, a 1,000-prompt factuality benchmark from Google DeepMind that evaluates short-form parametric knowledge (facts stored in model weights).
Source: google/simpleqa-verified (eval split, 1,000 examples)
Languages
Config
Language
Examples
ces
Czech
100
dan
Danish
100
deu
German
1,000
fra
French
100
ita
Italian
100
nld
Dutch
100
pol
Polish
100
spa
Spanish
100
More to… See the full description on the dataset page: https://huggingface.co/datasets/ellamind/simpleqa-verified-multilingual.simplequestions-sparqltotext
Dataset Card for SimpleQuestions-SPARQLtoText
Dataset Summary
Special version of SimpleQuestions with SPARQL queries formatted for the SPARQL-to-Text task.
JSON fields
The original version of SimpleQuestions is a raw text file listing triples and the natural language question. A JSON version has been generated and augmented with the following fields:
rdf_subject, rdf_property, rdf_object: triple in the Wikidata format (IDs)
nl_subject, nl_property, nl_object:… See the full description on the dataset page: https://huggingface.co/datasets/Orange/simplequestions-sparqltotext.LiteResearcher-SFT-Data
LiteResearcher — SFT Cold-Start Data
Distilled deep-research trajectories used for the SFT cold-start of LiteResearcher-4B
This dataset contains the 68,231 multi-turn deep-research trajectories used to
train the SFT cold-start checkpoint that RL (GRPO+TIS) is later launched from —
the "68.2 K distilled deep-research trajectories" referenced in the paper and in
LiteResearcher-Data.
Each row is a complete ReAct-style episode: a research question, the model's
interleaved… See the full description on the dataset page: https://huggingface.co/datasets/simplex-ai-inc/LiteResearcher-SFT-Data.simplescaling-s1K-R1
Dataset Card: s1k R1
Dataset Description
The s1k R1 dataset is a fork of the simplescaling/s1K dataset. It contains a collection of conversations where the assistant's messages have been enhanced to include Chain of Thought (CoT) reasoning within <think> ... </think> tags, followed by the final answer. This modification aims to improve the interpretability and reasoning capabilities of AI models by providing explicit thought processes in the responses.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/mkurman/simplescaling-s1K-R1.multilingual-textarena-SimpleTak-v0-train-v2
TextArena Language Trajectories
This dataset contains language-conditioned TextArena trajectory data.
Each dataset configuration corresponds to a different model, experiment group, or source folder.
Available configurations:
gemma4-e4b-it
qwen3-4b
ministral3-3b-instruct
Usage
Install the datasets library:
pip install datasets
Load a specific configuration:
from datasets import load_dataset
dataset =… See the full description on the dataset page: https://huggingface.co/datasets/The-CoLab/multilingual-textarena-SimpleTak-v0-train-v2.SimpleStories-JA
📘📕 SimpleStories 📙📗
このデータセットは、gpt-4o-miniによって生成された短編小説で出来ているデータセットです。生成方法や、自分で物語を生成する方法については、こちらのリポジトリをご覧ください。
他の言語や物語形式の制作を希望される場合は、メールにてお問い合わせください。
SimpleStoriesは、EldenとLiによるTinyStoriesの改良版です。
特徴
物語の注釈情報（theme、topic、styleなど）
多様性の高さ
2024年のモデルによって生成
NLPのデータが用意しているためフィルタリングしやすい
以下の言語版が利用可能：
英語
日本語
他にも追加予定
This dataset is a collection of short stories generated by gpt-4o-mini (+ other models, soon). To see how this dataset was generated, or to generate some stories… See the full description on the dataset page: https://huggingface.co/datasets/SimpleStories/SimpleStories-JA.multilingual-textarena-SimpleTak-v0-train
TextArena Language Trajectories
This dataset contains language-conditioned TextArena trajectory data.
Each dataset configuration corresponds to a different model, experiment group, or source folder.
Available configurations:
gemma4-e4b-it
qwen3-4b
ministral3-3b-instruct
Usage
Install the datasets library:
pip install datasets
Load a specific configuration:
from datasets import load_dataset
dataset =… See the full description on the dataset page: https://huggingface.co/datasets/The-CoLab/multilingual-textarena-SimpleTak-v0-train.simple_wikipedia_LM
Dataset Card for "simple_wikipedia_LM"
A filtered/edited version of pszemraj/simple_wikipedia that removes headings/contents that appear in the text column without any relevant text for them (at least in the simple split).
import re
def split_on_headings(text):
headings = ["References", "Related pages", "Other websites", "Further reading"]
for heading in headings:
parts = re.split(
r"^\s*" + re.escape(heading) + r".*$", text, flags=re.MULTILINE… See the full description on the dataset page: https://huggingface.co/datasets/pszemraj/simple_wikipedia_LM.SimpleMath
🧮 SimpleMath 100K
SimpleMath 100K is a high-quality synthetic dataset of 100,000 basic arithmetic problems — no noise, no tricks, just clean and accurate math.
✅ Purpose
This was made for small AI models — not to struggle with complex math, but to get simple math right every time.
📦 Contents
75,000 numeric problems, evenly split:
18,750 addition (456 + 789 =)
18,750 subtraction (900 - 345 =)
18,750 multiplication (12 x 15 =)
18,750 division (144 / 12 =)… See the full description on the dataset page: https://huggingface.co/datasets/ProCreations/SimpleMath.simple-wiki-article
Simple-wiki-article
This dataset is directly derived from rahular/simple-wikipedia.
Various techniques were used to detect which strings were article titles, and separate the original dataset into articles, with a somewhat good accuracy.
Article content strings were merged and split with \n, and <br> was replaced with \n to make the dataset more usable.
simplepile-lite
Dataset Card for "simplepile-lite"
Interleaved dataset using 'first exhausted' strategy. Counts:
DatasetDict({
train: Dataset({
features: ['text'],
num_rows: 452432
})
validation: Dataset({
features: ['text'],
num_rows: 1000
})
test: Dataset({
features: ['text'],
num_rows: 11908
})
})
token counts - train
using GPTNeoX Tokenizer:
token_count
count
452432
mean
868.642
std
4791.71… See the full description on the dataset page: https://huggingface.co/datasets/pszemraj/simplepile-lite.LiteResearcher-Data
LiteResearcher — RL Training Data
Companion training data for the LiteResearcher paper
A low-cost, scalable Agentic RL training framework for deep-research agents.
This dataset contains the two-stage curriculum of question–answer prompts
used to train LiteResearcher-4B with on-policy GRPO+TIS, fully against a
local search / browse environment. Both stages share the same validation set.
What this is not: the underlying webpage corpus (~32 M records, used by
the local… See the full description on the dataset page: https://huggingface.co/datasets/simplex-ai-inc/LiteResearcher-Data.task112_asset_simple_sentence_identification
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task112_asset_simple_sentence_identification
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task112_asset_simple_sentence_identification.VeriReason-RTL-Coder_7b_reasoning_tb_simple
Verireason-RTL-Coder_7b_reasoning_tb_simple
For implementation details, visit our GitHub repository: VeriReason and our page
Check out our paper: VeriReason: Reinforcement Learning with Testbench Feedback for Reasoning-Enhanced Verilog Generation
Update Log
2025.05.17: Initial release of Nellyw888/Verireason-RTL-Coder_7b_reasoning_tb_simple
Project Description
This study introduces VeriReason, a novel approach utilizing reinforcement learning with… See the full description on the dataset page: https://huggingface.co/datasets/Nellyw888/VeriReason-RTL-Coder_7b_reasoning_tb_simple.simple-100m-pretrain-1b
Simple-100M Pretraining Dataset (1B Tokens)
A training-optimized, packed pretraining dataset for ~100M parameter language models. Built for reproducibility, minimal runtime overhead, and exact mixing ratios.
🎯 Purpose
This dataset was created to train Simple-100M, a decoder-only Transformer targeting:
✅ Beat GPT-2-70M perplexity with minimal complexity
✅ Reproducible artifacts with exact token accounting
✅ Zero runtime preprocessing (ready-to-train)
Target… See the full description on the dataset page: https://huggingface.co/datasets/Dodosoomro/simple-100m-pretrain-1b.keep-it-simple
keep-it-simple
Objective: An ultra-minimalist dataset for pre-training tiny language models. The logic relies on bidirectional symmetry (A is B and B is A]) to foster deep semantic understanding. By training the model to predict the "prompt" from the "text" and vice versa, we maximize the utility of every pair.
Data Sources
Simple English Wikipedia: Simplified encyclopedic articles.
Vikidia (FR): Educational content for younger audiences.
OPUS Books (en-fr):… See the full description on the dataset page: https://huggingface.co/datasets/ffurfaro/keep-it-simple.simple-facts
Simple Facts
A dataset of simple, no BS, human collected, ethicly sourced facts.
About 1000 examples.
This dataset is growing, and every day I plan to add a few more facts.
SimpleStories
📘📕 SimpleStories 📙📗
SimpleStories is a dataset of >2 million model-generated short stories. It was made to train small, interpretable language models on it. The generation process is open-source: To see how the dataset was generated, or to generate some stories yourself, head over to this repository.
If you'd like to commission other languages or story formats, feel free to send mail.
When using SimpleStories in your work, please cite the SimpleStories paper:… See the full description on the dataset page: https://huggingface.co/datasets/duoduoyeah/SimpleStories.simplemath-cot
🧮 SimpleMath-100k CoT
A chain-of-thought (CoT) extension of the
ProCreations/SimpleMath
dataset. Every one of the 100 000 algebra / arithmetic problems is paired with a
short, numbered reasoning trace (Step 1: … Step 2: …) that walks a language
model from the problem statement to the known-correct answer.
The traces in the Jupyter notebook are generated by
Qwen3.8-27B and then post-processed to strip formatting noise,
enforce sequential step numbering, and cap output at 1 000… See the full description on the dataset page: https://huggingface.co/datasets/alexfromapex/simplemath-cot.simple-daily-conversations-cleaned
Dataset Card
This dataset contains a cleaned version of simple daily conversations. It comprises nearly 98K text snippets representing informal, everyday dialogue, curated and processed for various Natural Language Processing tasks.
Uses
Direct Use
This dataset is ideal for:
Training language models on informal, everyday conversational data.
Research exploring linguistic patterns in casual conversation.
Out-of-Scope Use
The dataset may not… See the full description on the dataset page: https://huggingface.co/datasets/aarohanverma/simple-daily-conversations-cleaned.ACG-SimpleQA
ACG-SimpleQA
🌐 Website •
🤗 Hugging Face
中文 | English
ACG-SimpleQA is an objective knowledge question-answering dataset focused on the Chinese ACG (Animation, Comic, Game) domain, containing 4242 auto-generated carefully designed QA samples. This benchmark aims to evaluate large language models' factual capabilities in the ACG culture domain, featuring Chinese language, diversity, high quality, static answers, and easy evaluation.
📢 Latest Updates… See the full description on the dataset page: https://huggingface.co/datasets/Papersnake/ACG-SimpleQA.simple_wikipedia_LM_quality_score_v1
Dataset Card for "simple_wikipedia_LM_quality_score_v1"
Adding quality score v1 to pszemraj/simple_wikipedia_LM
More Information needed
simple-llm-sft
Simple LLM SFT Dataset
This synthetic dataset contains 1,000 English prompt-response pairs for
supervised fine-tuning. It was created to fine-tune
Qwen/Qwen3.5-4B to give clear,
direct, and technically correct answers in simple English.
The writing guidance is inspired by ASD-STE100 Simplified Technical English.
The dataset does not claim official ASD-STE100 compliance or certification.
Dataset structure
The default configuration contains:
Split
Examples… See the full description on the dataset page: https://huggingface.co/datasets/thisisandreeeee/simple-llm-sft.SimpleThoughts
SimpleThoughts
A complete synthetic training corpus spanning all four LLM training stages — pretraining, supervised fine-tuning (SFT), preference
alignment (DPO), and reasoning — built entirely around simple thought experiments.
SimpleThoughts is designed to train language models that can think clearly about everyday phenomena rather than just recall facts. Every
sample is grounded in a concrete thought experiment: intuitive physics, causal inference, biology, economics, spatial… See the full description on the dataset page: https://huggingface.co/datasets/tensorfiend/SimpleThoughts.Simple-FriendlyMath
Simple-FriendlyMath
Simple-FriendlyMath is a lightweight dataset focused on basic math with a friendly, conversational style.It's designed to help small and medium models learn not just to calculate — but also to explain answers in a natural, helpful way.
Built as a natural follow-up to SimpleMath, this dataset keeps things easy while adding warmth, chat phrasing, and simple explanations.
What's Inside
10,000 examples of addition, subtraction, multiplication, and… See the full description on the dataset page: https://huggingface.co/datasets/ProCreations/Simple-FriendlyMath.simple_bench
📊 Simple Bench Dataset
A Compact Benchmark for Structured Reasoning and Multiple-Choice Evaluation in Large Language Models
Simple Bench Dataset is a structured evaluation collection derived from the Simple Bench benchmark, designed to assess reasoning, comprehension, and multiple-choice question-answering capabilities of large language models through concise yet non-trivial problems that require logical inference rather than simple retrieval; each sample consists of a natural… See the full description on the dataset page: https://huggingface.co/datasets/sapiens-technology/simple_bench.