datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/SHSLab/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection.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.fable-5-sft-traces
Fable-5 SFT Traces
Author / maintainer: kelexine (github.com/kelexine)
A cleaned, anonymised, schema-normalised derivative of
Kelexine/Fable-5-traces
— agentic traces from Fable-5 (claude-fable-5), the model now publicly
known as Claude Mythos — Anthropic's top-of-family frontier model at time
of collection.
The dataset supports three fine-tuning shapes off a single JSONL with no
preprocessing required:
Mode
Fields used
Full SFT (thinking + response)
messages or… See the full description on the dataset page: https://huggingface.co/datasets/kelexine/fable-5-sft-traces.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.gigaverbo-v2-sft
GigaVerbo-v2 SFT: A Large-Scale Portuguese Instruction-Tuning Dataset
Dataset Summary
GigaVerbo-v2 SFT is a large-scale instruction-tuning dataset designed for supervised fine-tuning of language models in Portuguese. The dataset comprises approximately 2.1 billion tokens (~4.4 GB) across 4 million instruction-following examples, organized into 12 distinct task categories. It is entirely composed of high-quality, LLM-generated data that has been carefully curated and… See the full description on the dataset page: https://huggingface.co/datasets/Polygl0t/gigaverbo-v2-sft.Dolci-Think-SFT-translated
Dolci-Think-SFT-translated
Machine translations of the Dolci-Think-SFT-32B dataset, produced with gemma-4-31B-it. The samples selected for translation are those where content_quality == "excellent" according to the propella annotations.
Columns
Each row is a translated conversation plus the result of a post-translation quality filter:
id — source record id.
messages — the translated conversation (list of {content, role}).
filter_pass — true if the row passed… See the full description on the dataset page: https://huggingface.co/datasets/openeurollm/Dolci-Think-SFT-translated.bhasha-sft
Bhasha SFT
Bhasha SFT is a massive collection of multiple open sourced Supervised Fine-Tuning datasets for training Multilingual
Large Language Models. The dataset contains collation of over 13 million instances of
instruction-response data for 3 Indian languages (Hindi, Gujarati, Bengali) and English having both human annotated and synthetic data.
Curated by: Soket AI Labs
Language(s) (NLP): [English, Hindi, Bengali, Gujarati]
License: [cc-by-4.0, apache-2.0, mit]… See the full description on the dataset page: https://huggingface.co/datasets/soketlabs/bhasha-sft.Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2.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.agent-sft-stitch-zh-tts-taste-codec-chat-sample
Gemma 4 E2B Taste-S multi-turn codec SFT
This dataset contains 37,362 complete Traditional Chinese agent
dialogues selected from voidful/agent-sft-stitch-zh-tts. It covers
229,434 synthesized speech segments, approximately
520.5 hours of audio before codec extraction.
Every assistant speech segment is represented without Gemma native audio tags:
<SAY> text_token <a_code> <b_code> ... <p_code> ... </SAY>
The first assistant output starts immediately with <SAY>.
[SOPR]...[EOPR]… See the full description on the dataset page: https://huggingface.co/datasets/voidful/agent-sft-stitch-zh-tts-taste-codec-chat-sample.synthetic-pre1930-sftTL;DR
A vintage finetuning dataset (~416k rows, eleven task routes). Sourced by taking excerpts
from pre-1930's texts, turning these into verbatim answers, and then using deepseek-chat to
generate period-appropriate questions of those answers. Any model tuned on this dataset should,
theoretically, never update its weights on anachronistic text, since questions are masked in the
finetuning stages. Features composition, verse, narrative, reasoning, multiturn dialogue, and
calibrated uncertainty… See the full description on the dataset page: https://huggingface.co/datasets/zachnorton03/synthetic-pre1930-sft.readall-sft-stage-a-b
ReadAll / ReadTwice SFT：Stage A + Stage B
当前 ReadAll 模型的 SFT 数据和可移植训练包。训练链为：
Qwen/Qwen2.5-7B-Instruct → Stage A (ReadTwice step286) → Stage B (ReadAll union step448)
阶段
训练行数
验证行数
Parquet 分片
全局 batch
学习率
1 epoch 更新数
A
73,416
1,676
15 + 1
256
1e-5
286
B
57,287
无独立验证集
29
128
5e-7
448
这些计数是 SFT 消息样本行数，包含 SKIM / UPDATE / FINAL，并非独立问题数。45 个 Parquet 共 1,030,251,959 字节。原始数据分片和 manifest 原样保留，逐一核对原始 SHA-256；没有删列、重新筛选或重新生成。
Stage A 使用普通多轮 assistant-token SFT、12,288 token… See the full description on the dataset page: https://huggingface.co/datasets/Xirui1208/readall-sft-stage-a-b.turkce-sft-qa-3.7m
🇹🇷 Turkish SFT/QA — Birleştirilmiş ve Tekrarsız Veri Seti
3,723,264 örnek. 24 açık Türkçe SFT/QA veri setinin, satır düzeyinde
tekrar temizliği ve kalite kontrolünden geçirilmiş birleşimi. Her satır hangi veri
setinden geldiğini taşır.
English: A merged, row-level deduplicated and quality-filtered collection of
24 open Turkish SFT/QA datasets (3,723,264 examples). Every row carries
its source dataset, source URL and original license.
🙏 Teşekkür /… See the full description on the dataset page: https://huggingface.co/datasets/MercanAI/turkce-sft-qa-3.7m.openresearcher-sft-deep-research-cleaned
OpenResearcher SFT DeepResearch — Parquet Mirror
This is a re-hosted copy of the tool-reasoning SFT deep-research dataset by Aman Priyanshu, itself a cleaned/restructured version of the OpenResearcher Dataset from TIGER-AI-Lab.
Why this repo exists: the source wasn't laid out as ready-to-download Parquet files. This mirror simply stores the data as plain seed_*.parquet files so you can grab the whole dataset or a single segment easily. No changes were made to the content — all… See the full description on the dataset page: https://huggingface.co/datasets/DanielTobi0/openresearcher-sft-deep-research-cleaned.dataforge-sft-trajectories
DataForge SFT Trajectories
This dataset contains chunk-level expert_v1, versioned expert_v2,
inferability-audited expert_v3, and contract-repair expert_v4
supervised-fine-tuning records for the DataForge warmup model. The current
milestone is built from
split-safe dirty/clean CSV diffs (oracle_from_clean_diff) so model training is
anchored to audited labels rather than teacher guesses.
The earlier v0-smoke checkpoint proved the Kaggle-to-Hugging-Face handoff. It
is not a… See the full description on the dataset page: https://huggingface.co/datasets/Praneshrajan15/dataforge-sft-trajectories.FineProofs-SFT
FineProofs SFT
Dataset Description
FineProofs SFT is a high-quality supervised fine-tuning dataset containing mathematical Olympiad problems paired with chain-of-thought reasoning and formal proofs distilled from DeepSeek-Math-V2. The dataset comprises 7,777 samples (4,300 unique problems) sourced from international Olympiad competitions and Art of Problem Solving (AoPS), each annotated with:
Detailed reasoning traces (thinking content) generated by… See the full description on the dataset page: https://huggingface.co/datasets/lm-provers/FineProofs-SFT.Qwen3-Coder-Next-Open-Code-SFT
Dataset Card — OpenCode Rejection Sampling
Overview
This dataset contains high-quality code reasoning data for training language models on competitive programming tasks. It is produced via rejection sampling with Qwen3-Coder-Next, which would generate multiple candidate solutions per problem, each candidate is executed against test cases in a sandboxed environment, and the results are used to build two complementary training datasets:
SFT dataset (49,374 examples)… See the full description on the dataset page: https://huggingface.co/datasets/zake7749/Qwen3-Coder-Next-Open-Code-SFT.chess-sft-eval
Chess SFT Eval & Benchmark
Held-out evaluation splits and a frozen benchmark for the
Chess SFT training pipeline.
Every FEN in these files is excluded from training data via a blocklist to guarantee
zero contamination.
Eval examples
13,000
Benchmark examples
13,000
Splits
9 (perception, rules, tactics, evaluation, openings, endgames, planning, chess960, mate)
Format
JSONL
Training companion
Chess-Nut-Engine/chess-sft-data
How eval and benchmark differ… See the full description on the dataset page: https://huggingface.co/datasets/Chess-Nut-Engine/chess-sft-eval.RST-SFT-Qwen3.5-27B
RST SFT trajectories for Qwen3.5-27B
Multi-turn terminal-agent conversations distilled from
Zhongzhi1228/Recursive-Task-Synthesis-Trajectories,
ready for supervised fine-tuning of Qwen/Qwen3.5-27B.
Pipeline, launchers, and the full plan: https://github.com/k1ssloo/RST-Train
cap10 reproduces the paper's SFT example count exactly
The source release has 327,189 trajectories. cap10 ends at 10,778 examples —
the count arXiv:2608.05466v3 states it trained
on. That was… See the full description on the dataset page: https://huggingface.co/datasets/NiuNiu0110/RST-SFT-Qwen3.5-27B.political-perspectives-sft
Neutrality Political Perspectives SFT
This dataset contains independent, open-ended political and public-policy
questions plus responses produced under documented strong U.S. progressive and
strong U.S. conservative interventions. It supports controlled research on
perspective-conditioned supervised fine-tuning. It is not a source of political
or factual ground truth.
The repository preserves the original 1,000-question release as v1, adds a
separate, substantially larger v2… See the full description on the dataset page: https://huggingface.co/datasets/TheNeutralityProject/political-perspectives-sft.if_oracle_sft
IF Oracle SFT
Oracle instruction-following SFT data built by rejection sampling from
willhx/Qwen3-8B-Base-IF.
How it was made
Policy: willhx/Qwen3-8B-Base-IF (a trained IF RL checkpoint).
Prompts: the full allenai/IF_multi_constraints_upto5-derived training set
(IF_multi_constraints_upto5_ifbench_en, 88,832 prompts covered).
Sampling: 8 responses per prompt, temperature 1.0, max 8192 response tokens
(rollout via slime + SGLang, --debug-rollout-only).
Reward:… See the full description on the dataset page: https://huggingface.co/datasets/willhx/if_oracle_sft.Soofi-Think-SFT-V2-firsthalf-DE
Soofi-Think-SFT-V2-firsthalf-DE
German-translated version of toroe/Soofi-Think-SFT-V2-firsthalf — a large-scale supervised fine-tuning dataset featuring chain-of-thought reasoning traces (<think>...</think>) across math, science, code, and general instruction-following tasks.
The translation was produced using Qwen3-32B via vLLM, applying professional-grade translation prompts with formal German register (Sie-form for professional/technical content).
Dataset Summary… See the full description on the dataset page: https://huggingface.co/datasets/toroe/Soofi-Think-SFT-V2-firsthalf-DE.Soofi-Think-SFT-V2-firsthalf-FR
Soofi-Think-SFT-V2-firsthalf-FR
French-translated version of toroe/Soofi-Think-SFT-V2-firsthalf — a large-scale supervised fine-tuning dataset featuring chain-of-thought reasoning traces (<think>...</think>) across math, science, code, tool-calling, and general instruction-following tasks.
The translation was produced using Qwen3-32B via vLLM, applying professional-grade translation prompts targeting standard French suitable for international francophone audiences.… See the full description on the dataset page: https://huggingface.co/datasets/toroe/Soofi-Think-SFT-V2-firsthalf-FR.superscout-sft-search
SuperScout search corpus (SFT)
The supervised fine-tuning corpus behind SuperScout-7B, a 7B searcher that
explores a repository, localizes the fault, writes a failing reproduction, and
emits a structured handoff. The dataset contains 19,911 examples as built and
frozen; six rows carrying malformed tool-call wrappers are dropped at load time,
giving the 19,905 examples actually trained on. A further 9,478 examples (the
third-best trace per issue) were held back as a shelf and… See the full description on the dataset page: https://huggingface.co/datasets/SuperAGI/superscout-sft-search.toy-models-of-sft-data
Toy Models of SFT Data
This is a public-clean candidate data package for the Toy Models of SFT project.
It is built for researcher inspection first.
The package answers two questions:
What were the models trained on?
How did the models actually behave under evaluation?
The package includes training data, eval inputs, model rollouts, judge scores,
parsed GPQA outputs, aggregate tables, paper figures, frozen plot data, and
provenance records. It deliberately includes some… See the full description on the dataset page: https://huggingface.co/datasets/matonski/toy-models-of-sft-data.running-coach-sft
Running Coach SFT
Instruction-tuning data for a distance-running coaching assistant. Every pace,
split, and race-equivalent in the corpus is computed from a Daniels/Gilbert VDOT
implementation rather than written into a template, so the numbers are internally
consistent across all 1,500 examples.
Why this exists
Coaching corpora scraped from forums and blogs teach a model the register of
coaching without the arithmetic underneath it. A model that interpolates… See the full description on the dataset page: https://huggingface.co/datasets/hoodarunner/running-coach-sft.multilingual-reasoning-gym-sft
Reasoning Gym SFT Dataset
This dataset contains Supervised Fine-Tuning (SFT) reasoning data procedurally generated using Reasoning Gym environments.
It is designed to train reasoning models (such as DeepSeek-R1-style or Qwen-Coder-style models) to explain their step-by-step reasoning chain before outputting a final answer wrapped inside LaTeX \boxed{...}.
Where Does This Dataset Come From?
This dataset is procedurally generated from Reasoning Gym, an open-source… See the full description on the dataset page: https://huggingface.co/datasets/MauroPello/multilingual-reasoning-gym-sft.chinese-legal-sft
Chinese Legal SFT Dataset(中文法律 SFT 数据集)
面向大模型监督微调(SFT)的中文法律问答数据集,共 19,332 条问答对,
每条附带 LLM 质量评分。覆盖数据采集 → 清洗 → 去重 → 质量过滤 → 格式化 → 质量打分的完整数据工程流程。
配套代码与完整流水线:https://github.com/noah-white-python/legal-sft-dataset
数据构建流程
冷启动:基于开源数据集 DISC-Law-SFT 整理。
清洗:NFKC 全角半角统一、去控制字符、去空白、缺失过滤。
去重:精确去重(MD5)+ MinHash + LSH 近似去重(阈值 0.8)。
质量过滤:长度、中文字符占比等启发式规则,有效率 96.7%(20,000 → 19,332)。
格式化:输出标准 Alpaca 指令格式。
质量打分:用 LLM-as-judge 对全部数据从复杂度、清晰度、信息量三维度打分(1-5 分)。
字段说明… See the full description on the dataset page: https://huggingface.co/datasets/noah248/chinese-legal-sft.danish-ner-sft-v1
danish-ner-sft-v1
Danish named-entity SFT rows derived from chcaa/dansk-ner
(DANSK), which annotates Danish Gigaword text with the 18 OntoNotes entity
types using the same scheme in every split. Each row asks for a random subset
of entity types (1-6, including types absent from the passage so the empty
marker is exercised) and renders the answer in one of fourteen output formats.
Ten are key-value, listing the extracted entities: JSON, key: value,
key=value, [key] value, value… See the full description on the dataset page: https://huggingface.co/datasets/jensjepsen/danish-ner-sft-v1.