datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Domofon-Cot-Conversations-700k
Domofon-Cot-Conversations-700k
Synthetic XML conversation data for training small language models on reasoning,
instruction following, XML formatting, and tool-use traces.
Repository: domofon/Domofon-Cot-Conversations-700k
What is inside
The dataset contains cleaned generated XML conversations from six families:
conv: multi-turn factual conversations with tool-use traces.
instruct: text-processing instructions, including deterministic count tool calls.
ds:… See the full description on the dataset page: https://huggingface.co/datasets/domofon/Domofon-Cot-Conversations-700k.structured-cpt
Structured CPT - JSON + SQL pretrain documents
SmolLM2-1.7B continued-pretraining shard of structured documents. Each document
is a <task> / <input> / <output> block whose <output> is a canonical
JSON object, terminated by the SmolLM2 end-of-text token ``.
Sources:
source
description
rows
shards
repeat
sql_bmc2
b-mc2 sql-create-context -> JSON (4 keys, stub explanation)
392,885
1
5
sql_gretelai
gretelai synthetic_text_to_sql -> JSON (4 keys)
529,255
1
5… See the full description on the dataset page: https://huggingface.co/datasets/domofon/structured-cpt.worldsim-domofon-cpt
Worldsim → Domofon CPT
Cleaned continued-pretraining shard from VatsaDev/Worldsim.
One split: train. Assistant name is Domofon; creator/org is DomofonAI. Documents do not include <|endoftext|>. 499 source docs, each repeated 10 times with a unique id (4,990 rows).
from datasets import load_dataset
ds = load_dataset("domofon/worldsim-domofon-cpt", split="train")
Columns
column
description
id
unique document id ({source_id}-{copy})
text
full CPT… See the full description on the dataset page: https://huggingface.co/datasets/domofon/worldsim-domofon-cpt.domofon-cot-part1
Domofon COT Dataset - Part 1
Mixed Chain-of-Thought датасет на русском языке. (Russian language)
Статистика
Строк | lines : 700,000
Токенов: | tokens 594,412,523
Модели-генераторы (Distill Generation Models used):
Deepseek R1 671B
Mistral Large
Devstral Mini
Qwen 480B
Структура (Structure)
user - вопрос пользователя (User Question)
thinking - цепочка рассуждений (Chain-of-Thought)
answer - финальный ответ (Final answer)
Связанные… See the full description on the dataset page: https://huggingface.co/datasets/domofon/domofon-cot-part1.python-code-cot-18k
python-code-cot-18k
COT distilled dataset with 16,565 examples.
Source
Base: iamtarun/python_code_instructions_18k_alpaca
Model: Mistral-7B-Instruct-v0.2-AWQ
Format
instruction: Task
thinking: <think>...</think> reasoning
response: Solution
reddit-mental-health-summaries-150k
reddit-mental-health-summaries-150k
Empathetic summaries dataset with 85,288 examples.
Source
Base: Akhil059/reddit_mental_health_posts
Model: Mistral-7B-Instruct-v0.2-AWQ
Generated: 2025-12-21
Format
subreddit: Source subreddit\n- title: Post title\n- post: Original post\n- empathetic_summary: Caring summary
Usage
from datasets import load_dataset
ds = load_dataset("domofon/reddit-mental-health-summaries-150k")
Chat-Agent-1M
Chat-Agent-1M
919,869 English chat & agentic conversations in ChatML (Anthropic-style) Messages format. Every
assistant turn opens with a short, dense chain-of-thought (<reasoning>) that leads to the answer,
and ~30% of the conversations include a full tool-use trajectory (tool availability → tool call →
tool result → grounded answer).
The conversations are built on top of the curated answers of
HuggingFaceTB/smoltalk2 (the broad chat
subsets): the original verbose thinking was… See the full description on the dataset page: https://huggingface.co/datasets/DomofonResearch/Chat-Agent-1M.50k-HTML-PRETRAIN
50k-HTML-PRETRAIN
Pretrain-style pairs: an English site assignment and a complete HTML document that implements it.
Each row is one assignment (user) and one HTML page (assistant). A handful of rows include a PNG screenshot of the page; the rest leave screenshot empty.
Split
split
n
train
57,617
10 rows have a PNG in screenshot / images/. The other rows have a null screenshot.
from datasets import load_dataset
ds =… See the full description on the dataset page: https://huggingface.co/datasets/domofon/50k-HTML-PRETRAIN.python-questions-cot-10k
python-questions-cot-10k
COT distilled dataset with 10,485 examples.
Source
Base: maniteja7463/python_programming_questions
Model: Mistral-7B-Instruct-v0.2-AWQ
Generated: 2025-12-21
Format
instruction: Task\n- thinking: COT reasoning\n- response: Solution
Usage
from datasets import load_dataset
ds = load_dataset("domofon/python-questions-cot-10k")
britannica-pretrain
britannica-pretrain
Private pretrain documents from Encyclopaedia Britannica scans (Internet Archive OCR *_djvu.txt), indexed via biglam/britannica-illustrated-pages manifest. One Arrow row is one article. No chunking. Trainer reads text.
<encyclopedia>
<meta source="britannica" edition="11th" year="1911" />
<title>Elasticity</title>
<body>
...
</body>
</encyclopedia>
<|endoftext|>
<|endoftext|> is SmolLM2-1.7B eos. OCR is noisy. Duplicate library copies were collapsed by… See the full description on the dataset page: https://huggingface.co/datasets/domofon/britannica-pretrain.misc_pretrain_268649
misc_pretrain_268649
Miscellaneous pretrain-style pairs: a source file (Python or SVG) plus a short assistant note.
Splits
split
n
source
code
248,127
short Python files with a first-person author note
svg
20,522
SVG drawings with an English description of the picture
from datasets import load_dataset
code = load_dataset("domofon/misc_pretrain_268649", split="code")
svg = load_dataset("domofon/misc_pretrain_268649", split="svg")… See the full description on the dataset page: https://huggingface.co/datasets/domofon/misc_pretrain_268649.domofon-identity-anchor
Domofon identity anchor
1,158 identity dialogues, each repeated 10 times with a unique id (11,580 pretrain documents). Each document is an <identity> block plus a SmolLM2 ChatML turn (<|im_start|>user / <|im_start|>assistant). There is no <|endoftext|>.
from datasets import load_dataset
ds = load_dataset("domofon/domofon-identity-anchor", split="train")
print(ds[0]["text"])
Document
<identity>
You are Domofon, developed by DomofonAI. Your knowledge cutoff is… See the full description on the dataset page: https://huggingface.co/datasets/domofon/domofon-identity-anchor.reddit-2026
2026 DUMP !
reddit-pretrain-xml
Private pretrain-document dump of public Reddit mirrors. One JSONL row is one document. The trainer should read text only.
text format
<doc>
<meta subreddit="changemyview" date="2026-08-01" />
<title>...</title>
<body>
...
</body>
</doc>
<|endoftext|>
<|endoftext|> is SmolLM2-1.7B eos (token id 0). Do not wrap these docs in Chat SFT.
XML special characters in title/body are escaped. Markdown and emoji are kept. If… See the full description on the dataset page: https://huggingface.co/datasets/domofon/reddit-2026.Document-XML-100k
Document-XML-100k
Noisy/unstructured text to semantically tagged XML. 118K pairs for fine-tuning document markup models.
Splits
Split
Rows
Description
verified
79,067
Content-exact: byte-level match between input text and XML text content. Zero information loss guaranteed.
good
38,916
High quality (word overlap >= 85%, well-formed XML, no HTML tags) but with minor whitespace normalization.
What's the difference?
Both splits are… See the full description on the dataset page: https://huggingface.co/datasets/domofon/Document-XML-100k.magicoder-cot-92k
Magicoder COT 92K
Chain-of-Thought distilled dataset for code generation tasks.
Dataset Description
This dataset contains 92,334 examples with added <think> reasoning traces, distilled from the original Magicoder-Evol-Instruct-110K dataset.
Source
Base dataset: ise-uiuc/Magicoder-Evol-Instruct-110K
Distillation model: Mistral-7B-Instruct-v0.2-AWQ
Method: vLLM inference with 50 parallel workers
Format
Each example contains:
instruction: The coding… See the full description on the dataset page: https://huggingface.co/datasets/domofon/magicoder-cot-92k.Tool-Reasoning-31K
Tool-Reasoning-31K BY Domofon Research
30,764 unique tool-use conversations in Anthropic-style ChatML, each assistant step prefixed
with a short first-person <reasoning>. The assistant decides whether and which tool to call,
calls it, reads a real tool result, and answers — across single-turn, multi-step, multi-turn, and
"no tool fits" (relevance) scenarios.
Derived from interstellarninja/hermes_reasoning_tool_use
(which aggregates xLAM / ToolACE / Glaive / Nous-Hermes /… See the full description on the dataset page: https://huggingface.co/datasets/DomofonResearch/Tool-Reasoning-31K.Gasai-Agent-CoderForge
Gasai-Agent-CoderForge
5,205 verified-resolved agentic software-engineering trajectories, re-authored into the Gasai harness
format for pretraining small language models. Derived from togethercomputer/CoderForge-Preview
(reward==1, Python). Each row is one complete tool-use trace serialized as a single Gasai control-token
sequence in the gasai field.
Sister dataset to GasaiAI/Gasai-Agent-5k
(from nvidia/Open-SWE-Traces) — byte-identical format, same pipeline.
What… See the full description on the dataset page: https://huggingface.co/datasets/domofon/Gasai-Agent-CoderForge.domofon-cot-part2
Domofon COT Dataset - Part 2 (Mixed Styles)
Mixed Chain-of-Thought датасет на русском языке с разными стилями генерации.
Статистика
Строк: 321,072
Токенов: 299,560,472
Стили генерации
default - базовый стиль
analytical - аналитический, структурированный
creative - творческий, с метафорами
practical - практичный, с примерами
Модели-генераторы
Deepseek R1 671B
Mistral Large
Devstral Mini
Qwen 480B
Структура
user - вопрос пользователя… See the full description on the dataset page: https://huggingface.co/datasets/domofon/domofon-cot-part2.SLM-Tools-300k
SLM-tools-300k
351,833 agentic tool-use traces in Anthropic-style ChatML format — for teaching small models tool-calling with concise step-by-step reasoning.
Format
Each JSONL line: {"rendered": <full ChatML string>, "messages": [...], "metrics": {...}}.
ChatML blocks per trace:
available_tools — full JSON tool schemas
system — assistant persona / style directive
user — request
assistant — <reasoning>…</reasoning> + <tool_call>{"name","arguments"}</tool_call>… See the full description on the dataset page: https://huggingface.co/datasets/DomofonResearch/SLM-Tools-300k.code-alpaca-cot-20k
code-alpaca-cot-20k
COT distilled dataset with 18,213 examples.
Source
Base: sahil2801/CodeAlpaca-20k
Model: Mistral-7B-Instruct-v0.2-AWQ
Format
instruction: Task
thinking: <think>...</think> reasoning
response: Solution
reddit-pretrain-post
reddit-pretrain-post
Same documents as domofon/reddit-pretrain-xml, private. The only difference is the wrapper tag: <post> instead of <doc>. <title>, <body>, and <meta> are unchanged.
<post>
<meta subreddit="changemyview" date="2026-08-01" />
<title>...</title>
<body>
...
</body>
</post>
<|endoftext|>
<|endoftext|> is SmolLM2-1.7B eos (token id 0). Trainer reads text only. 3,713,744 documents, 8 data/train-*.jsonl.gz shards.
fake_news_cot_reasoning
Fake News Chain-of-Thought Reasoning Dataset
This dataset contains 9,500 news articles with Chain-of-Thought (CoT) reasoning explanations for why each article is classified as fake or real news.
Dataset Description
Each record contains natural language reasoning that explains the classification decision, generated using Qwen 2.5 1.5B model via llama.cpp inference.
Features
Field
Type
Description
title
string
News article headline
input
string
Full… See the full description on the dataset page: https://huggingface.co/datasets/domofon/fake_news_cot_reasoning.Gasai-Agent-5k
Gasai-Agent-5k
4,997 verified agentic software-engineering trajectories, re-authored end-to-end for training small language models (SLMs). Each example is a complete tool-use trace — a structured task prompt, a sequence of tool calls with dense reasoning, tool observations, and a final answer — serialized in the Gasai harness format.
Built for Gasai-6B (a ~6B general-purpose SLM), this dataset is a heavy re-working of real resolved GitHub issues so that a small model learns from… See the full description on the dataset page: https://huggingface.co/datasets/DomofonResearch/Gasai-Agent-5k.evol-instruct-code-cot-80k
evol-instruct-code-cot-80k
COT distilled dataset with 63,007 examples.
Source
Base: nickrosh/Evol-Instruct-Code-80k-v1
Model: Mistral-7B-Instruct-v0.2-AWQ
Format
instruction: Task
thinking: <think>...</think> reasoning
response: Solution