datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
the-stack-smol
Dataset Description
A small subset (~0.1%) of the-stack dataset, each programming language has 10,000 random samples from the original dataset. The dataset has 2.6GB of text (code).
Languages
The dataset contains 30 programming languages:
"assembly", "batchfile", "c++", "c", "c-sharp", "cmake", "css", "dockerfile", "fortran", "go", "haskell", "html", "java",
"javascript", "julia", "lua", "makefile", "markdown", "perl", "php", "powershell", "python", "ruby", "rust"… See the full description on the dataset page: https://huggingface.co/datasets/bigcode/the-stack-smol.the-stack-smol-xl
Dataset Description
A small subset of the-stack dataset, with 87 programming languages, each has 10,000 random samples from the original dataset.
Languages
The dataset contains 87 programming languages:
'ada', 'agda', 'alloy', 'antlr', 'applescript', 'assembly', 'augeas', 'awk', 'batchfile', 'bison', 'bluespec', 'c',
'c++', 'c-sharp', 'clojure', 'cmake', 'coffeescript', 'common-lisp', 'css', 'cuda', 'dart', 'dockerfile', 'elixir',
'elm', 'emacs-lisp','erlang'… See the full description on the dataset page: https://huggingface.co/datasets/bigcode/the-stack-smol-xl.the-stack-smol-xs\smoltalk-chinese
Chinese SmolTalk Dataset [中文] [English]
[OpenCSG Community] [👾github] [wechat] [Twitter]
📖Technical Report
smoltalk-chinese is a Chinese fine-tuning dataset constructed with reference to the SmolTalk dataset. It aims to provide high-quality synthetic data support for training large language models (LLMs). The dataset consists entirely of synthetic data, comprising over 700,000 entries. It is specifically designed to enhance the performance of Chinese… See the full description on the dataset page: https://huggingface.co/datasets/opencsg/smoltalk-chinese.jupyter-scripts-smollm3
The Stack v2 Jupyter Notebooks as Scripts
This dataset contains script representations of the Jupyter notebooks in
The Stack v2. It was
created from the materialized Jupyter_Notebook split in
jordangong/the-stack-v2-smollm3.
The output schema follows the Jupyter-script schema used by
bigcode/starcoderdata,
but this release is not deduplicated, PII-filtered, or otherwise equivalent
to StarCoderData's filtered split.
Relationship to the SmolLM3 training mix
This… See the full description on the dataset page: https://huggingface.co/datasets/jordangong/jupyter-scripts-smollm3.smolmo-sft-v2-seqlen64k
smolmo-sft-v2-seqlen64k
A supervised fine-tuning (SFT) dataset of math problems with full chain-of-thought solutions,
formatted for the Olmo 3 "Thinking" models.
2,813,055 examples · ~37.9 B tokens.
Three task families: proofs, numeric-answer problems, and tool-augmented (Python) problems.
Every assistant turn carries an explicit <think> … </think> reasoning trace before the answer.
Olmo 3 native chat + function-calling format; every example fits within a 64k-token context.… See the full description on the dataset page: https://huggingface.co/datasets/chankhavu/smolmo-sft-v2-seqlen64k.smolkalam-arabic-conversational-sft
SmolKalam
SmolKalam is a quality-filtered Arabic SFT dataset of 1,790,478 examples (~2.45B tokens), built as an ensemble translation of SmolTalk2. It covers multi-turn dialogue (23% of rows), reasoning traces (19% carry <think>), tool and function calling (4.4%), and long context, categories that are underrepresented in existing Arabic post-training data. The SmolTalk2 source mixtures are kept as subsets.
Released with the paper SmolKalam: Ensemble Quality-Filtered Translation… See the full description on the dataset page: https://huggingface.co/datasets/AdaMLLab/smolkalam-arabic-conversational-sft.the-stack-v2-train-smol-ids
The Stack v2
The dataset consists of 4 versions:
bigcode/the-stack-v2: the full "The Stack v2" dataset
bigcode/the-stack-v2-dedup: based on the bigcode/the-stack-v2 but further near-deduplicated
bigcode/the-stack-v2-train-full-ids: based on the bigcode/the-stack-v2-dedup dataset but further filtered with heuristics and spanning 600+ programming languages. The data is grouped into repositories.
bigcode/the-stack-v2-train-smol-ids: based on the bigcode/the-stack-v2-dedup… See the full description on the dataset page: https://huggingface.co/datasets/bigcode/the-stack-v2-train-smol-ids.smol-worldcup
🏟️ Smol AI WorldCup — SHIFT Benchmark
The world's first 5-axis evaluation framework for small language models.
Not just "how smart?" — but "how honest? how fast? how small? how efficient?"
🏟️ Leaderboard
huggingface.co/spaces/ginigen-ai/smol-worldcup
📊 Dataset
huggingface.co/datasets/ginigen-ai/smol-worldcup
🏅 ALL Bench
huggingface.co/spaces/FINAL-Bench/all-bench-leaderboard
🏆 Official Ranking: WCS (WorldCup Score)
WCS = √( SHIFT × PIR_norm )… See the full description on the dataset page: https://huggingface.co/datasets/ginigen-ai/smol-worldcup.smoltalk-chinese
Chinese SmolTalk Dataset [中文] [English]
[OpenCSG Community] [👾github] [wechat] [Twitter]
📖Technical Report
smoltalk-chinese is a Chinese fine-tuning dataset constructed with reference to the SmolTalk dataset. It aims to provide high-quality synthetic data support for training large language models (LLMs). The dataset consists entirely of synthetic data, comprising over 700,000 entries. It is specifically designed to enhance the performance of Chinese… See the full description on the dataset page: https://huggingface.co/datasets/sunorme/smoltalk-chinese.smoltalk-chinese-QwQ-Distrill
smoltalk-chinese-QwQ-Distrill [中文] [English]
📖Technical Report
smoltalk-chinese-QwQ-Distrill is a Chinese fine-tuning dataset constructed with reference to the SmolTalk-Chinese dataset. It aims to provide high-quality synthetic reasoning data support for training large language models (LLMs). The dataset consists entirely of synthetic data, comprising over 700,000 entries. It is specifically designed to enhance the performance of Chinese LLMs across various tasks… See the full description on the dataset page: https://huggingface.co/datasets/ChinaunicomSoftware/smoltalk-chinese-QwQ-Distrill.smolgpt-markdown-stories
SmolGPT-Fables Stories
A deterministic, text-only corpus of 96,000 original English
Markdown stories built for SmolGPT-Fables. Every row is one complete supervised
story example with an exact prompt / completion boundary, a requested scene
count from one to six, and plain-language conditioning fields.
No model, API, browser, or network service was used to create this dataset.
Dataset summary
96,000 stories across 96,000 isolated story families
25 genres and all… See the full description on the dataset page: https://huggingface.co/datasets/neonforestmist/smolgpt-markdown-stories.smoltalk-smol-magpie-ultra-no-refusals
SmolTalk Smol-Magpie-Ultra No Refusals
A Minos-cleaned version of HuggingFaceTB/smoltalk / smol-magpie-ultra for use as a neutral helpfulness SFT anchor.
Rows are removed when NousResearch/Minos-v1 classifies the conversation as a refusal. The original train/test split structure is preserved.
Cleaning version: minos-only-v1-2026-06-23
Counts
Split
Input rows
Kept rows
Dropped rows
train
409,537
408,447
1,090
test
21,555
21,488
67
Overall removal… See the full description on the dataset page: https://huggingface.co/datasets/nchapman/smoltalk-smol-magpie-ultra-no-refusals.the-stack-smol-xs-all
bigcode/the-stack-smol-xs - all configs
All configs from bigcode/the-stack-smol-xs concatenated and shuffled. 100 examples each of:
['ada', 'agda', 'alloy', 'antlr', 'applescript', 'assembly', 'augeas', 'awk',
'batchfile', 'bison', 'bluespec', 'c', 'c++', 'c-sharp', 'clojure', 'cmake',
'coffeescript', 'common-lisp', 'css', 'cuda', 'dart', 'dockerfile', 'elixir',
'elm', 'emacs-lisp', 'erlang', 'f-sharp', 'fortran', 'glsl', 'go', 'groovy',
'haskell', 'html', 'idris', 'isabelle'… See the full description on the dataset page: https://huggingface.co/datasets/BEE-spoke-data/the-stack-smol-xs-all.smoltalk-chinese
Chinese SmolTalk Dataset [中文] [English]
[OpenCSG Community] [👾github] [wechat] [Twitter]
📖Technical Report
smoltalk-chinese is a Chinese fine-tuning dataset constructed with reference to the SmolTalk dataset. It aims to provide high-quality synthetic data support for training large language models (LLMs). The dataset consists entirely of synthetic data, comprising over 700,000 entries. It is specifically designed to enhance the performance of Chinese… See the full description on the dataset page: https://huggingface.co/datasets/pigandcat0624/smoltalk-chinese.nb-asr-numerics-categorized-smoke-test
Norwegian Bokmål Numeric Expression Categorized Dataset
This dataset represents Stage 2 of the Norwegian numerics-data pipeline. It contains semantic validation and categorization annotations of the Norwegian numeric expression sentences harvested in Stage 1.
Source Dataset
Harvested Dataset: pere/nb-asr-numerics-harvested (approx. 3.8 million rows across 6 shards).
Processing Architecture
Inference Model: google/gemma-4-12B-it (instruction-tuned… See the full description on the dataset page: https://huggingface.co/datasets/pere/nb-asr-numerics-categorized-smoke-test.gemma-4-e2b-nla-eval-smoke
Gemma-4-E2B NLA smoke-eval (20-row held-out set)
A 20-row held-out subset of OpenWebText activations extracted from google/gemma-4-E2B at layer 23. Used as the canonical eval set for smoke-testing the v0.0.1 Gemma-4-E2B NLA pair on a fresh environment.
This dataset is a subset of the held-out rl.parquet evaluation set used for the v0.0.1 round-trip eval (n=50 attempted, 42 evaluated after 8 empty-output exclusions, cos 0.438 ± 0.054). The 20-row subset preserves the activation… See the full description on the dataset page: https://huggingface.co/datasets/Solshine/gemma-4-e2b-nla-eval-smoke.smolkalam
Dataset Splits
Gemma 3
Split Name
# Examples
LongAlign_64k_Qwen3_32B_yarn_131k_think
7,526
LongAlign_64k_context_lang_annotated_lang_6_no_think
6,249
Mixture_of_Thoughts_science_no_think
86,110
OpenHermes_2.5_no_think
384,900
OpenThoughts3_50K
50,000
OpenThoughts3_NoThink_180K
180,000
aya_dataset_Qwen3_32B_think
15,222
hermes_function_calling_v1_no_think
8,961
multi_turn_reasoning_if_think
28,217
s1k_1.1_think
835
smolagents_toolcalling_traces_think
9… See the full description on the dataset page: https://huggingface.co/datasets/SultanR/smolkalam.matilda-smollm-mix-15b-gpt2
matilda-smollm-mix-15B-gpt2
15 B GPT-2-BPE tokens drawn from a 5:1 token-balanced mix of
HuggingFaceTB/smollm-corpus:
Source
Share
Tokens
fineweb-edu-dedup
83.33 %
12.50 B
cosmopedia-v2
16.67 %
2.50 B
Total: 15,000,349,569 tokens across 151 shards (shard_*.bin, uint16,
100 M tokens per shard).
The full SmolLM recipe is 75 / 15 / 10 fineweb-edu / cosmopedia-v2 / python-edu.
python-edu was dropped because the HuggingFaceTB/smollm-corpus subset
ships only blob_id… See the full description on the dataset page: https://huggingface.co/datasets/prometheus04/matilda-smollm-mix-15b-gpt2.smoltalk-chinese
Chinese SmolTalk Dataset [中文] [English]
[OpenCSG Community] [👾github] [wechat] [Twitter]
📖Technical Report
smoltalk-chinese is a Chinese fine-tuning dataset constructed with reference to the SmolTalk dataset. It aims to provide high-quality synthetic data support for training large language models (LLMs). The dataset consists entirely of synthetic data, comprising over 700,000 entries. It is specifically designed to enhance the performance of Chinese… See the full description on the dataset page: https://huggingface.co/datasets/syszxxxwu/smoltalk-chinese.smoltalk_azThis is part of the translated version of the original dataset: https://huggingface.co/datasets/HuggingFaceTB/smoltalk
smollm3-base-blindspots
SmolLM3-3B-Base Blind Spots Evaluation Dataset
Dataset Summary
This dataset documents 10 diverse failure cases discovered while evaluating
HuggingFaceTB/SmolLM3-3B-Base,
a 3-billion parameter decoder-only base language model released by Hugging Face in July 2025.
The evaluation was conducted as part of the Fatima Fellowship technical challenge on Blind Spots of Frontier Models.
Model Tested
Model: HuggingFaceTB/SmolLM3-3B-Base
Parameters: 3 billion… See the full description on the dataset page: https://huggingface.co/datasets/habibahabchi/smollm3-base-blindspots.smollm-corpus-python
smollm-corpus - python
A version of the python-edu subset with the text added
reverse-text-tinystories-easy-smoke
Reverse Text TinyStories Easy Smoke
This is a small smoke-test dataset for the reverse-text task.
Splits
train: 12 rows
test: 4 rows
Columns
prompt
char_count
word_count
source
Source
Derived from roneneldan/TinyStories by taking non-overlapping word windows and keeping only prompts that fall in the easy character-length bucket.
Difficulty Rule
All rows in this dataset are easy examples with prompt lengths in the 20-74 character range.… See the full description on the dataset page: https://huggingface.co/datasets/13point5/reverse-text-tinystories-easy-smoke.SmolDataEnvs-sft
🛠️ SmolDataEnvs — SFT
5.5K+ RL tasks for hill-climbing small models in code and data science.
A 2B model on these tasks. Left: what it optimises. Right: 144 held-out tasks it never trains on.
Two runs over the same 5,000 tasks — shuffled against a curriculum ordered easiest to hardest.
4,677 worked examples of an agent doing data science the right way. Each row is a complete,
verified-correct trajectory: read the question, poke at the data with a shell tool… See the full description on the dataset page: https://huggingface.co/datasets/FineEnvs/SmolDataEnvs-sft.