datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
tiny-textbooks
Textbook-like Dataset: A High-Quality Resource for Small Language Models
The idea is simply inspired by the Textbooks Are All You Need II: phi-1.5 technical report paper. The source texts in this dataset have been gathered and carefully select the best of the falcon-refinedweb and minipile datasets to ensure the diversity, quality while tiny in size. The dataset was synthesized using 4x3090 Ti cards over a period of 500 hours, thanks to Nous-Hermes-Llama2-13b finetuned model.
Why… See the full description on the dataset page: https://huggingface.co/datasets/nampdn-ai/tiny-textbooks.qwen3-5-tiny-cpu-repro-v1
Qwen3.5 tiny native random CPU fixture
Complete randomly initialized, untrained Qwen3_5ForConditionalGeneration checkpoint.
This is a pipeline/reproducibility fixture, not a useful language model, distillation,
quantization, quality benchmark, or claim about the performance of Qwen3.8-27B.
No upstream model weights or training data were used. No paid GPU/cloud compute.
Architecture and lineage
Architecture lineage: Qwen/Qwen3.8-27B at… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/qwen3-5-tiny-cpu-repro-v1.glm5-next-tiny-cpu-repro-v1This repository is an evidence bundle, not one root-format dataset at repository root.
first/ and repeat/ are separate complete sealed QFS root datasets; comparison/ holds
the comparison receipt and tokenwise result. panel/ is the sealed input panel. Other files
are provenance, logs and reproduction tools. Do not pass the bundle root as a QFS dataset.
GLM5-Next tiny native CPU fixture
This is a complete untrained random-initialized native Glm5NextForConditionalGeneration
wrapper… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/glm5-next-tiny-cpu-repro-v1.tiny-slm-pretraining-corpus
🚀 Ultra High-Quality Tiny SLM Pre-Training Corpus (<100GB)
A state-of-the-art, balanced 7-domain pre-training dataset engineered specifically for Small Language Models (Tiny SLMs: 50M – 2B parameters) such as SmolLM2, SmolLM3, MobileLLM, Llama 3.2 1B, and custom architectures.
100% compatible with Unsloth Studio, Unsloth AI, Hugging Face datasets, and PyTorch DataLoaders.
📊 Dataset Statistics
Total Documents: 20,066,075
Train: 19,663,898
Validation: 402,177… See the full description on the dataset page: https://huggingface.co/datasets/JustACluelessKidAtSchool/tiny-slm-pretraining-corpus.deepseek-v4-tiny-cpu-repro-v1
DeepSeek-V4 tiny corrected-native-primitives CPU text fixture
Complete randomly initialized, untrained QFSDeepseekV4ForCausalLM text class
using Transformers5.16.1 native primitives and a reviewed RMSNorm arithmetic correction.
No upstream weights, paid GPU/cloud compute or useful-model claim.
This is not unmodified native Transformers or the complete production release.
Architecture and scope
Text lineage:… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/deepseek-v4-tiny-cpu-repro-v1.k2-horizon-tiny-cpu-repro-v1
K2-Horizon MoVA tiny random CPU fixture
Complete untrained K2HorizonForCausalLM, not Moonshot Kimi despite the K2 name.
Architecture source: IFM/K2-Horizon-MoVA-36B-A4B at 05cab0a4d7150c1c460a000b37ff40cc1af2feaa.
No pretrained weights, original tokenizer, training data, paid GPU or cloud compute used.
Complete text-only K2HorizonForCausalLM, not Kimi: three-layer dense prefix followed by two real MoVA+MoE layers, grouped RMSNorm, sigmoid top-k routing with selection-only bias… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/k2-horizon-tiny-cpu-repro-v1.minimax-m3-tiny-cpu-repro-v1
minimax-m3 complete native tiny random CPU fixture
Complete untrained MiniMaxM3SparseForConditionalGeneration checkpoint with an untied full LM head,
a real 272-entry byte tokenizer and every native state tensor. Architecture lineage:
MiniMaxAI/MiniMax-M3@f0e1c1e04d40177e4673a22097036854f536e9c0.
No upstream weights, training data, paid GPU or cloud compute were used.
Complete native image/text wrapper with real shrunk Conv3D vision, nonempty 3D RoPE, patch-merge projector and… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/minimax-m3-tiny-cpu-repro-v1.minimax-m2-tiny-cpu-repro-v1
minimax-m2 complete native tiny random CPU fixture
Complete untrained MiniMaxM2ForCausalLM checkpoint with an untied full LM head,
a real 272-entry byte tokenizer and every native state tensor. Architecture lineage:
MiniMaxAI/MiniMax-M2.7@d494266a4affc0d2995ba1fa35c8481cbd84294b.
No upstream weights, training data, paid GPU or cloud compute were used.
Complete native text causal LM: sigmoid/top-k MoE routing with correction bias, per-layer flattened Q/K RMSNorm and half-head… See the full description on the dataset page: https://huggingface.co/datasets/malaiwah/minimax-m2-tiny-cpu-repro-v1.TinyStories-Algerian-Darija
TinyStories Algerian Darija
Synthetic short stories in Algerian Darja paired with their English originals, for Darija language modeling and translation, from the Algerian NLP Collective. The Hub datasets-server reports 11,326 train rows (/info?dataset=algerian-nlp/TinyStories-Algerian-Darija, 2026-09-17), independently confirmed by the build ledger processed_story_ids.json in this repo: 11,326 unique story ids (0 to 11,492, non-contiguous).
The default config answers: what does… See the full description on the dataset page: https://huggingface.co/datasets/algerian-nlp/TinyStories-Algerian-Darija.tiny-code-textbooks
Code Explanation Textbooks
A collection of 207k synthetic code with explanation as a tiny textbook. Filtered from the-stack, each programming language contains few thousands samples. I only choose the best meaningful code to generate synthetic textbook.
tiny-ua-bench-responses
Tiny-UA-Bench Responses
This dataset contains the response matrix for Tiny-UA-Bench.
The matrix contains 919,160 model and item records.
The matrix covers 20 models and 45,958 items.
The evaluation excludes FLORES and LongFLORES.
Use
Use this dataset to reproduce the benchmark compression analysis.
Do not use a held-out model response to fit a selector or predictor.
Use the reference and held-out split definitions from the code repository.
Load the data with the… See the full description on the dataset page: https://huggingface.co/datasets/robinhad/tiny-ua-bench-responses.tinystories-gpt4-instruct
tinystories-gpt4-instruct
Request→story pairs for supervised fine-tuning of small language models, derived from karpathy/tinystories-gpt4-clean. Each example pairs a natural-language request ("Can you tell me a story about a boy named Tim?") with a TinyStories story that satisfies it.
The dataset lives on Hugging Face; the notebook that generates it lives on GitHub.
This is not roneneldan/TinyStoriesInstruct. That dataset frames its tasks in a structured format (Words:… See the full description on the dataset page: https://huggingface.co/datasets/Pondsiders/tinystories-gpt4-instruct.tiny-vintage-completions
Tiny vintage completions
Synthetic vintage texts, with a cutoff date for year 1900.
Based on unique 2-3 word seeds, extracted from croqaz/Vintage-v1, croqaz/Vintage-v2 and Haykgrigorian/English-historical-corpus-1800-1875.
Check the files seeds1.txt and seeds2.txt.
Generated by TypeWriter-7B-base and Talkie-13B-base completions.
Citation
If you find this dataset valuable, please consider citing:
@misc{Tiny-vintage-completions,
title = {Tiny vintage completions}… See the full description on the dataset page: https://huggingface.co/datasets/croqaz/tiny-vintage-completions.tinysynth-reasoning
TinySynth Reasoning Primitives
Synthetic training data for teaching small language models stable state
representation and controlled reasoning operations — entity/attribute
binding, state persistence, mutation, transfer, reference resolution,
current-vs-cumulative distinctions, and claim validation — in a
systems/computing vocabulary.
Every example is generated from a hidden symbolic world and verified by a
symbolic solver before any natural language is produced:
semantic state… See the full description on the dataset page: https://huggingface.co/datasets/projenix/tinysynth-reasoning.sorry-bench-202503-multilingual
sorry-bench-202503-multilingual
Multilingual version of SorryBench — a benchmark for evaluating LLM safety refusals across 44 harm categories and 21 prompt styles.
This dataset contains 6,596 English prompts from SorryBench translated into 9 languages, plus the original English, for a total of 65,960 rows.
Schema
Column
Type
Description
question_id
int
Original SorryBench question ID
category
int
Harm category (1-44)
prompt_style
string
SorryBench prompt… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-safety/sorry-bench-202503-multilingual.tinystories-gr
TinyStories-GR
A full Modern Greek translation of the TinyStories dataset (~2.1 million short English children's stories), with AI-generated quality scores for each translation.
Dataset Description
TinyStories-GR was generated by running the entire TinyStories corpus through a two-stage AI pipeline:
Translation — each English story was translated to Modern Greek by Google Gemini (gemini-3.1-flash-lite-preview)
Evaluation — each translation was independently scored (1–5)… See the full description on the dataset page: https://huggingface.co/datasets/alexliap/tinystories-gr.tiny-webtext
Tiny WebText
The Tiny WebText dataset is designed to help models learn about perception on web text while neutralizing the bias of the source text using critical thinking methods. By providing a rich and diverse set of texts, I aim to improve the ability of models to understand and analyze information in a more objective and unbiased manner.
This dataset can be used to train and evaluate natural language processing and machine learning models, with the goal of improving their… See the full description on the dataset page: https://huggingface.co/datasets/nampdn-ai/tiny-webtext.dna-tiny-world
DNA-World-Tiny
Benchmark for DNA foundational models using real MPRA data from MPRAbase.
Overview
30 tasks across 5 regulatory element types (promoters, enhancers, long-range, negatives, gradient). All targets are real wet-lab MPRA measurements.
Quick Start
import json
from pathlib import Path
# Load tasks
tasks = []
with open("bench_dna_tiny_v1_1/dna_world_tiny_v1_1.jsonl") as f:
for line in f:
tasks.append(json.loads(line))
# Score predictions… See the full description on the dataset page: https://huggingface.co/datasets/sethmorton/dna-tiny-world.gpt-oss-20b_tinystories-val1pct-raw
openai/gpt-oss-20b — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: openai/gpt-oss-20b
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after meta-prompt… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/gpt-oss-20b_tinystories-val1pct-raw.olmo-3-7b-instruct_tinystories-val1pct-raw
allenai/OLMo-3-7B-Instruct — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: allenai/OLMo-3-7B-Instruct
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/olmo-3-7b-instruct_tinystories-val1pct-raw.reverse-text-tinystories-hard
Reverse Text TinyStories Hard
This is the hard-difficulty TinyStories dataset for the reverse-text task.
Splits
train: 4000 rows
test: 500 rows
Columns
prompt
char_count
word_count
source
Source
Derived from roneneldan/TinyStories using non-overlapping word windows with cross-dataset prefix checks.
Difficulty Rule
hard rows only
character-count range: 99-160
Notes
The reverse answer is not stored because the reverse-text… See the full description on the dataset page: https://huggingface.co/datasets/13point5/reverse-text-tinystories-hard.tiny-aya-base-blind-spots
Blind Spots of a Frontier Base Model: Evaluation Dataset
This dataset documents blind spots discovered in a frontier open-weight base model through 19 structured evaluation tests. It was assembled as part of an assignment on identifying model weaknesses using the HelloBench evaluation framework.
Model Tested
CohereLabs/tiny-aya-base
Architecture: Transformer with Sliding Window Attention (SWA) (window size 4096, with RoPE) on three layers + one global attention layer… See the full description on the dataset page: https://huggingface.co/datasets/Mawube/tiny-aya-base-blind-spots.nanbeige4-3b-thinking-2511_tinystories-val1pct-raw
Nanbeige/Nanbeige4-3B-Thinking-2511 — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: Nanbeige/Nanbeige4-3B-Thinking-2511
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/nanbeige4-3b-thinking-2511_tinystories-val1pct-raw.qwen3-8b_tinystories-val1pct-raw
Qwen/Qwen3-8B — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: Qwen/Qwen3-8B
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after meta-prompt application)… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/qwen3-8b_tinystories-val1pct-raw.gemma-3-27b-it_tinystories-val1pct-raw
google/gemma-3-27b-it — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: google/gemma-3-27b-it
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after meta-prompt… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/gemma-3-27b-it_tinystories-val1pct-raw.tiny-lessons
Tiny Lessons
The dataset is designed to help causal language models learn more effectively from raw web text. It is augmented from public web text and contains two key components: theoretical concepts and practical examples.
The theoretical concepts provide a foundation for understanding the underlying principles and ideas behind the information contained in the raw web text. The practical examples demonstrate how these theoretical concepts can be applied in real-world situations.… See the full description on the dataset page: https://huggingface.co/datasets/nampdn-ai/tiny-lessons.qwen3-32b_tinystories-val1pct-raw
Qwen/Qwen3-32B — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: Qwen/Qwen3-32B
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after meta-prompt application)… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/qwen3-32b_tinystories-val1pct-raw.tiny-coop-es
Dataset Card for Tiny-Coop-ES
This dataset contains examples of synthetic data generated with Mistral Small 3.2 following the TinyStories methodology. Tiny-Coop-ES contains
examples of stories written in Spanish, with vocabulary that a kid between 3-4 years old would use and understand. Putting special emphasis in fables
where cooperation values are taught.
Dataset Details
Dataset Description
TinyCoop-ES is a synthetic dataset created inspired in the… See the full description on the dataset page: https://huggingface.co/datasets/hetline/tiny-coop-es.gemma-4-31b-it_tinystories-val1pct-raw
google/gemma-4-31b-it — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: google/gemma-4-31b-it
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model (after meta-prompt… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/gemma-4-31b-it_tinystories-val1pct-raw.llama-3.1-8b-instruct_tinystories-val1pct-raw
meta-llama/Llama-3.1-8B-Instruct — tinystories-val1pct-raw
Model outputs from the micro-creativity inference suite.
Model: meta-llama/Llama-3.1-8B-Instruct
Dataset: tinystories-val1pct-raw (220 items)
Part of collection: ZachW/llm-creativity-benchmarks
Generation config
temperature: 0.0
max_tokens: 16384
seed: 42
backend: vllm
Columns
Column
Description
task_id
Unique task identifier
input
The exact prompt sent to the model… See the full description on the dataset page: https://huggingface.co/datasets/ZachW/llama-3.1-8b-instruct_tinystories-val1pct-raw.