datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
UltraData-SFT-2605
UltraData-SFT-2605
📦 UltraData Collection |
🌐 UltraData |
🤗 MiniCPM5 Series
English |
中文
📚 Introduction
UltraData-SFT-2605 is the full set of core-domain SFT data used in the post-training of MiniCPM5-1B-SFT within the MiniCPM5-1B series, and a key representative of L3 refined data in the UltraData L0-L4 tiered data management framework. It covers math, code, knowledge, instruction following, and other core domains, containing over 15 million Deep… See the full description on the dataset page: https://huggingface.co/datasets/openbmb/UltraData-SFT-2605.UltraData-SFT-Agent-2609
UltraData-SFT-Agent-2609
📦 UltraData Collection |
🌐 UltraData |
🤗 MiniCPM5 Series
English |
中文
📚 Introduction
UltraData-SFT-Agent-2609 is the L3 refined data for Agent instruction-tuning within UltraData's L0-L4 tiered data management framework. Built for the post-training of MiniCPM5-2B, it complements UltraData-SFT-2605 (core-domain SFT) with executable Agent trajectories. The release contains approximately 500,000 samples spanning tool use… See the full description on the dataset page: https://huggingface.co/datasets/openbmb/UltraData-SFT-Agent-2609.medical-o1-reasoning-SFT
News
[2025/04/22] We split the data and kept only the medical SFT dataset (medical_o1_sft.json). The file medical_o1_sft_mix.json contains a mix of medical and general instruction data.
[2025/02/22] We released the distilled dataset from Deepseek-R1 based on medical verifiable problems. You can use it to initialize your models with the reasoning chain from Deepseek-R1.
[2024/12/25] We open-sourced the medical reasoning dataset for SFT, built on medical verifiable problems and an… See the full description on the dataset page: https://huggingface.co/datasets/FreedomIntelligence/medical-o1-reasoning-SFT.Nemotron-SFT-Science-v2
Dataset Description:
Nemotron-Science-v2 is a science reasoning dataset with synthetic (synthetic MCQ, RQA) and non-synthetic vendor problems and LLM-generated solutions. It comprises three domains (Physics, Biology, and Chemistry), two question formats (multiple-choice questions [MCQ] and open questions [OpenQ]), and three generation setups: chain-of-thought (CoT) reasoning without tools, Python tool usage, and search tools usage with the Tavily API.
The solutions were… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-Science-v2.Nemotron-SFT-Instruction-Following-Chat-v3
Dataset Description:
The Nemotron-Instruction-Following-Chat-v3 dataset is designed to strengthen multi-turn, interactive capabilities, including open-ended chat and precise instruction following.
The chat subset uses human written prompts from sources like lmarena, lmsys, and wildchat as seed prompts. Responses are generated with GLM-5. Multiple responses are sampled from the model and the best response as judged by pairwise comparisons using Qwen3-Nemotron-235B-A22B-GenRM-2603… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-Instruction-Following-Chat-v3.creative-writing-sft-50k
Creative Writing SFT (50K)
50,000 ShareGPT-format creative writing conversations across 12 literary forms and 25 themes. Written to demonstrate craft — not just competent completion, but genuine literary quality: specific detail, earned emotion, controlled voice, purposeful structure.
Motivation
Most LLM creative writing training data optimizes for fluency and completion rather than craft. Models learn to produce writing that reads smoothly but relies on clichés… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/creative-writing-sft-50k.novel-agent-sft-dataset
All Novel Can Be Galgame — 完整数据集
中文小说叙事理解项目的完整数据集。包含 669 本中文小说的原始文本、标注和训练数据，用于训练叙事 Agent 系统。
项目地址：https://github.com/lin1753/novel2galgame
训练代码仓库：https://github.com/lin1753/novel-agent
数据规模
目录
文件数
大小
说明
training/
52
689 MB
训练用 SFT 数据 (JSONL)
raw-books/
671
327 MB
669 本原始小说
processed/
39,842
1.2 GB
按章节预处理文本
annotations/
1,626
1 MB
原始标注文件
合计
42,191
2.2 GB
目录结构
datasets/
├── training/
│ ├── base-sft/… See the full description on the dataset page: https://huggingface.co/datasets/mikuhhn1239/novel-agent-sft-dataset.UltraData-SFT-2605-no-think-8k-32k
UltraData-SFT-2605 · no_think · 8k–32k
A length-filtered subset of the no_think split of
openbmb/UltraData-SFT-2605,
containing conversations whose token length falls in the 8k–32k range.
This is the medium-length tier intended for standard long-context SFT.
Two companion tiers were produced from the same source:
Dataset
Length range
Records
this repo — fxmeng/UltraData-SFT-2605-no-think-8k-32k
8k–32k tokens
623,421
fxmeng/UltraData-SFT-2605-no-think-32k-200k… See the full description on the dataset page: https://huggingface.co/datasets/fxmeng/UltraData-SFT-2605-no-think-8k-32k.Nemotron-Math-Proofs-v3-SFT
Nemotron-Math-Proofs-v3-SFT
Dataset Description:
Nemotron-Math-Proofs-v3-SFT is a long-form mathematical reasoning dataset containing proof-generation, proof-refinement, verification, and meta-verification traces. The release contains 414,890 samples representing 15,818 unique problems after quality filtering.
The source pool contains 15,879 hard proof problems selected from the AoPS subset of nvidia/Nemotron-Math-Proofs-v1. Responses are generated using… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-Math-Proofs-v3-SFT.Nemotron-SFT-Multilingual-v2
Dataset Description:
Nemotron-SFT-Multilingual-v2 is a multilingual supervised fine-tuning (SFT) dataset for post-training text-generation models. It is generated by translating seed data from Nemotron-Math-v2, Nemotron-Competitive-Programming-v1, and Nemotron-Science-v1, adding multilingual coverage for Hindi (hi), Korean (ko), Brazilian Portuguese (pt-br), and refreshed Japanese (ja) data.
The dataset is generated with a new data processing pipeline that avoids line-breaking… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-Multilingual-v2.Light-R1-SFTData
Light-R1: Surpassing R1-Distill from Scratch* with $1000 through Curriculum SFT & DPO
*from models without long COT
technical report
GitHub page
Here are the two-stage SFT data we used to train Light-R1-32B.
Simply refer to stage1-76k.json and stage2-3k.json
Model
Trained From
Release Date
AIME24
AIME25
DeepSeek-R1-Distill-Llama-70B
Llama-3.3-70B-Instruct
25.1.20
70.0
54.1
DeepSeek-R1-Distill-Qwen-32B
Qwen2.5-32B
25.1.20
72.6
54.9
LIMO (32B)
Qwen2.5-32B-Instruct
25.2.4… See the full description on the dataset page: https://huggingface.co/datasets/qihoo360/Light-R1-SFTData.Nemotron-SFT-ARC-AGI-v1
Dataset Description:
Nemotron-SFT-ARC-AGI-v1 is a supervised fine-tuning (SFT) dataset of multi-turn agentic reasoning traces produced by open-weight large language models attempting to solve ARC-AGI visual-reasoning puzzles. Each ARC puzzle (a set of (input grid, output grid) demonstration pairs plus one or more test inputs, where grids are 2D integer arrays representing colors) is formatted as a text prompt and given to an agent powered by one of nine open-weight reasoning… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-ARC-AGI-v1.Superior-Reasoning-SFT-gpt-oss-120b
Superior-Reasoning-SFT-gpt-oss-120b
📣 News
Our dataset ranked #1 on the Hugging Face Datasets Trending leaderboard from January 20 to January 30.
🚀 Overview
The Superior-Reasoning-SFT-gpt-oss-120b dataset is a high-quality, open-source collection containing 435K samples designed to democratize the training of high-performance Long Chain-of-Thought (Long-CoT) models. Unlike standard distilled datasets that rely on random sampling or… See the full description on the dataset page: https://huggingface.co/datasets/Alibaba-Apsara/Superior-Reasoning-SFT-gpt-oss-120b.sat-vl-sft-training-ready-v1
Dataset Summary
NuTonic/sat-bbox-metadata-sft-v1 is a metadata-first, procedural VLM SFT dataset built from an existing “sat-bbox” style dataset tree (Sentinel‑2 chips + per-tile JSON metadata sidecars, optionally paired Mapbox stills).
The goal is to create high-signal, production-shaped supervision for multimodal chat models:
Captioning for satellite chips
Grounding (bounding boxes in normalized coordinates) for land-cover regions
Class-focused captions and absence checks for… See the full description on the dataset page: https://huggingface.co/datasets/NuTonic/sat-vl-sft-training-ready-v1.OpenO1-SFTThis repository contains the dataset used for fine-tuning a language model using SFT for Chain-of-Thought Activation from the paper Generative AI Act II: Test Time Scaling Drives Cognition Engineering.
Code: https://github.com/GAIR-NLP/cognition-engineering
🎉🎉🎉This repository contains the dataset used for fine-tuning a language model using SFT for Chain-of-Thought Activation.
🌈🌈🌈The dataset is designed to enhance the model's ability to generate coherent and logical reasoning sequences.… See the full description on the dataset page: https://huggingface.co/datasets/O1-OPEN/OpenO1-SFT.moss-002-sft-data
Dataset Card for "moss-002-sft-data"
Dataset Summary
An open-source conversational dataset that was used to train MOSS-002. The user prompts are extended based on a small set of human-written seed prompts in a way similar to Self-Instruct. The AI responses are generated using text-davinci-003. The user prompts of en_harmlessness are from Anthropic red teaming data.
Data Splits
name
# samples
en_helpfulness.json
419049
en_honesty.json
112580… See the full description on the dataset page: https://huggingface.co/datasets/OpenMOSS-Team/moss-002-sft-data.XYZ-Aquila-SFT
XYZ-Aquila SFT
XYZ-Aquila SFT is a bilingual release of 7,000 multi-turn, search-oriented
tool-use trajectories, comprising 5,000 English examples and 2,000 Chinese
examples.
This release is a sample of the broader supervised fine-tuning data used for
XYZ-Aquila-mini and
XYZ-Aquila-pro. The examples
capture agent interactions with search tools, intermediate observations, and
answer generation in English and Chinese.
A small portion of the QA content is derived from… See the full description on the dataset page: https://huggingface.co/datasets/XYZAILab/XYZ-Aquila-SFT.cyberstrike-sft-120k
CyberStrike SFT 120K
The largest open-source offensive cybersecurity SFT dataset
121,422 expert-level red team instruction-response pairs across 15 security generators
Quick Start •
Why CyberStrike •
Domains •
Data Format •
Training Guide •
Benchmarks •
Contributing •
License
Why CyberStrike?
Most LLMs refuse or give surface-level answers to offensive security questions. Security professionals —… See the full description on the dataset page: https://huggingface.co/datasets/oyildirim/cyberstrike-sft-120k.UltraData-SFT-2605-no-think-32k-200k
UltraData-SFT-2605 · no_think · 32k–200k
A length-filtered subset of the no_think split of
openbmb/UltraData-SFT-2605,
containing conversations whose token length falls in the 32k–200k range.
This is the long-context tier intended for extended-context SFT.
Two companion tiers were produced from the same source:
Dataset
Length range
Records
fxmeng/UltraData-SFT-2605-no-think-8k-32k
8k–32k tokens
623,421
this repo — fxmeng/UltraData-SFT-2605-no-think-32k-200k
32k–200k… See the full description on the dataset page: https://huggingface.co/datasets/fxmeng/UltraData-SFT-2605-no-think-32k-200k.text-sft-questions-answers-only
text-sft: Questions and Answers
This dataset consists of question-and-answer pairs generated from short excerpts drawn from Wikipedia, Cosmopedia, and FineWeb-Edu. It is an adapted version of agentlans/text-sft.
Overview
The dataset provides compact examples of English question-and-answer relationships that can help models learn linguistic patterns, syntactic structures, and semantic associations between questions and their corresponding answers.
Intended Use… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/text-sft-questions-answers-only.math-reasoning-sft-100k
Math Reasoning SFT (100K)
100,000 math problems with detailed step-by-step solutions — ready for supervised fine-tuning of math reasoning models.
Dataset Description
100,000 problems across 8 mathematical categories and 3 difficulty levels:
Categories
Category
Examples
Topics
word_problems
~23,100
Rate/time/distance, work problems, mixture, meeting/catch-up
arithmetic
~15,400
Percentages, profit/loss, ratios
geometry
~15,400
Area… See the full description on the dataset page: https://huggingface.co/datasets/stindardlogic/math-reasoning-sft-100k.Soyuz-sft
Soyuz-sft 🚀
Soyuz — a union of trajectories from the orbits of different models.
Unified ChatML-format SFT dataset aggregating agent trajectories from frontier models (Claude Sonnet 4.5, GPT-5 codex, Kimi K2/K2.5, Qwen3-Coder-480B, GLM-5/5.1, Minimax M2.5, gpt-5.2). Curated for distillation into smaller agent models.
Format
Each line is one trajectory:
{
"messages": [
{"role": "system", "content": "..."},
{"role": "user", "content": "..."},
{"role":… See the full description on the dataset page: https://huggingface.co/datasets/AlexWortega/Soyuz-sft.Agent-G2-ALFWorld-Webshop-sft-data
Agent-G2 SFT Data
Agent-G2 SFT Data contains reasoning and action trajectories for supervised
fine-tuning (SFT) in the Agent-G2 project.
Associated paper: Agent-G2: Gaussian Guidance for Agentic Reinforcement
Learning — accepted to the EMNLP 2026 Main Conference.
The dataset covers two interactive agent environments:
WebShop: agents search for products, select options, and complete
purchases according to user requirements.
ALFWorld: agents interact with household environments… See the full description on the dataset page: https://huggingface.co/datasets/xiamoent/Agent-G2-ALFWorld-Webshop-sft-data.sat-vl-sft-postprocessed-merged-v1
Dataset Summary
NuTonic/sat-bbox-metadata-sft-v1 is a metadata-first, procedural VLM SFT dataset built from an existing “sat-bbox” style dataset tree (Sentinel‑2 chips + per-tile JSON metadata sidecars, optionally paired Mapbox stills).
The goal is to create high-signal, production-shaped supervision for multimodal chat models:
Captioning for satellite chips
Grounding (bounding boxes in normalized coordinates) for land-cover regions
Class-focused captions and absence checks for… See the full description on the dataset page: https://huggingface.co/datasets/NuTonic/sat-vl-sft-postprocessed-merged-v1.Nemotron-SFT-CUDA-v1
Dataset Description:
Nemotron-SFT-CUDA-v1 is a training dataset for CUDA code. It helps language models write CUDA kernels and solve CUDA programming problems.
We start from CUDA code in Nemotron Pretraining Code v2, which has a permissive license. An OpenCode agent powered by GLM-4.7 reads that code and writes new CUDA programming problems. Each problem comes with a hidden answer and tests. A second OpenCode + GLM-4.7 agent then tries to solve the problems using only the prompt… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SFT-CUDA-v1.Chinese-DeepSeek-R1-Distill-data-110k-SFT
中文基于满血DeepSeek-R1蒸馏数据集（Chinese-Data-Distill-From-R1）
🤗 Hugging Face | 🤖 ModelScope | 🚀 Github | 📑 Blog
注意：该版本为，可以直接SFT使用的版本，将原始数据中的思考和答案整合成output字段，大部分SFT代码框架均可直接直接加载训练。
本数据集为中文开源蒸馏满血R1的数据集，数据集中不仅包含math数据，还包括大量的通用类型数据，总数量为110K。
为什么开源这个数据？
R1的效果十分强大，并且基于R1蒸馏数据SFT的小模型也展现出了强大的效果，但检索发现，大部分开源的R1蒸馏数据集均为英文数据集。 同时，R1的报告中展示，蒸馏模型中同时也使用了部分通用场景数据集。
为了帮助大家更好地复现R1蒸馏模型的效果，特此开源中文数据集。该中文数据集中的数据分布如下：
Math：共计36568个样本，
Exam：共计2432个样本，
STEM：共计12648个样本，… See the full description on the dataset page: https://huggingface.co/datasets/Congliu/Chinese-DeepSeek-R1-Distill-data-110k-SFT.Turkish-SFT-Dataset-v1.0
Turkish-SFT-Dataset-v1.01
Repo: AlicanKiraz0/Turkish-SFT-Dataset-v1.0Sürüm: v1.01Lisans: MITBiçim: jsonl (kolonlar: system, user, assistant)Boyut: ~5500 satır ve satır başına 3.000–4.500 token/satır (≈ 20M+ token)Dil: Türkçe (tr)Görevler: talimat izleme, SFT, muhakeme, güvenli ret, uzun-bağlam ve araç kullanım bilinci
🔎 Özet
Bu veri kümesi, Türkçe Denetimli İnce Ayar (SFT) için tasarlanmış, yüksek kaliteli ve uzun çıktılar içeren örneklerden oluşur. İçerik 12 ana… See the full description on the dataset page: https://huggingface.co/datasets/AlicanKiraz0/Turkish-SFT-Dataset-v1.0.data-use-sft-tiered
Data-use SFT — tiered workflow (two task subsets)
Multitask SFT anchored exclusively on mentions the tiered extractor emits
(T1 evidential ∪ T2 declaration; see
rafmacalaba/data-use-mentions-tiered). Every row carries task
("provenance" | "usage_impact") and origin (prwp | fcv). Rows whose anchor
span was judged T3 (non-mention) or junk are dropped — audit trail in
manifest.jsonl (provenance) and manifest_usage.jsonl (usage/impact).
task = provenance (22,201 rows)… See the full description on the dataset page: https://huggingface.co/datasets/rafmacalaba/data-use-sft-tiered.LiteCoder-Terminal-SFT
LiteCoder-SFT-Terminal
Paper | Code | Blog Post
LiteCoder-SFT-Terminal is a dataset of 11,255 agent trajectories in terminal environments, introduced in the paper LiteCoder-Terminal: Scaling Long-Horizon Terminal Environments for Learning Language Agents.
Fine-tuned on this data, the LiteCoder-Terminal-30b-a3b-sft model achieves 31.5% Pass@1 on Terminal Bench Pro, while the LiteCoder-Terminal-4b-sft model shows distinct gains over its baseline.
Released Artifacts… See the full description on the dataset page: https://huggingface.co/datasets/Lite-Coder/LiteCoder-Terminal-SFT.magpie-sft-v1.0
magpie-sft-v1.0
This repository provides an instruction-tuning dataset developed by LLM-jp, a collaborative project launched in Japan.
This is a dataset of instruction and response pairs created using the Magpie method.
cyberagent/calm3-22b-chat was used for generating the instructions, and Qwen/Qwen2.5-32B-Instruct was used for generating the responses.
Send Questions to
llm-jp(at)nii.ac.jp
Model Card Authors
The names are listed in alphabetical order.… See the full description on the dataset page: https://huggingface.co/datasets/llm-jp/magpie-sft-v1.0.