datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
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.lm-eval-results-princeton-nlp-Llama-3-Base-8B-SFT-RDPO-private
Dataset Card for Evaluation run of princeton-nlp/Llama-3-Base-8B-SFT-RDPO
Dataset automatically created during the evaluation run of model princeton-nlp/Llama-3-Base-8B-SFT-RDPO
The dataset is composed of 62 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 7 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the… See the full description on the dataset page: https://huggingface.co/datasets/nyu-dice-lab/lm-eval-results-princeton-nlp-Llama-3-Base-8B-SFT-RDPO-private.LLM-fingerprinted-SFTround2-oss-matched
round2-oss-matched — 第二轮 4 组实验数据（每组 10 节点，共 40）
代码：repo 分支 claude/round2-matched-compute（先 git fetch origin && git merge origin/claude/round2-matched-compute）。
目录：
exp0_20b/node00..04/pool.jsonl # 实验 0:shard-05 修复重跑(20B)
exp0_120b/node00..04/pool.jsonl # 实验 0:同上(120B)
exp1_20b/node00..09/{seeds,budgets,pool}.jsonl # 实验 1:20B token 对齐独立采样
exp2_120b/node00..09/{seeds,budgets,pool}.jsonl # 实验 2:120B 同上
exp3_120b/node00..09/{ck_nonsat/,nonsat_seeds,budgets,pool… See the full description on the dataset page: https://huggingface.co/datasets/tts-sft/round2-oss-matched.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.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.qwen3_instruct_sft_2196030
OpenThinker3-qwen3-2196030
This model is a fine-tuned version of /home/rishabhtiwari/hf_cache/Qwen--Qwen3-30B-A3B-Base on the open_thoughts_3_small_instruct dataset.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate:… See the full description on the dataset page: https://huggingface.co/datasets/rishabh2k1/qwen3_instruct_sft_2196030.qwen3_reasoning_sft_2111685
OpenThinker3-qwen3-2111685
This model is a fine-tuned version of /home/rishabhtiwari/hf_cache/Qwen--Qwen3-30B-A3B-Base on the open_thoughts_3_small dataset.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate: 8e-05… See the full description on the dataset page: https://huggingface.co/datasets/rishabh2k1/qwen3_reasoning_sft_2111685.qwen3_instruct_sft_2196031
OpenThinker3-qwen3-2196031
This model is a fine-tuned version of /home/rishabhtiwari/hf_cache/Qwen--Qwen3-30B-A3B-Base on the open_thoughts_3_small_instruct dataset.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate:… See the full description on the dataset page: https://huggingface.co/datasets/rishabh2k1/qwen3_instruct_sft_2196031.FineMed-SFT
Introduction
This dataset is constructed using Qwen2.5-72B-Instruct and QwQ-32B-Preview, and it serves as the foundation for fine-tuning both FineMedLM and FineMedLM-o1.
This repository contains the SFT data used during the training process, including all medical data, fine-grained medical data, and thinking data. To use the dataset, simply download the jsonl file from the folder.
For details, see our paper and GitHub repository.
Citation
If you find our data useful… See the full description on the dataset page: https://huggingface.co/datasets/hongzhouyu/FineMed-SFT.dr-tulu-sft-data
[!NOTE]
For full information, go check out the Dr Tulu paper here.
DR Tulu SFT Data
This dataset contains the SFT training data for DR Tulu, containing prompts and full trajectories including reasoning traces, tool calls, and answers with citations.
The source prompts are curated from OpenScholar, Search Arena, and short-form QA datasets inclduing WebWalker-Silver, TaskCraft, PopQA and TyDiQA (English).
Important: This does not contain the SFT subsets created using prompts from… See the full description on the dataset page: https://huggingface.co/datasets/rl-research/dr-tulu-sft-data.primo-sft-json
PRIMO SFT Data
Stage-1 (SFT cold start) training annotations for PRIMO R1 (paper). Each record carries a chain-of-thought trace with planning / observation / reasoning subsections, which is what the model imitates before RL.
116,755 records across 10 subsets. Annotations only (856 MB); videos are in primo-video-media.
Subsets
Subset
Records
JSON
Video group
behavior-1k
19,991
167 MB
5,981 GB
robotwin-randomized
18,497
103 MB
12.5 GB (shared with… See the full description on the dataset page: https://huggingface.co/datasets/LeonOverload/primo-sft-json.Soofi-Think-SFT-V2-secondhalf-DEqwen3_reasoning_sft_2111879
OpenThinker3-qwen3-2111879
This model is a fine-tuned version of /home/rishabhtiwari/hf_cache/Qwen--Qwen3-30B-A3B-Base on the open_thoughts_3_small dataset.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate: 8e-05… See the full description on the dataset page: https://huggingface.co/datasets/rishabh2k1/qwen3_reasoning_sft_2111879.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.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.round4-independent
ROUND 4 — combined r2+r3 pool, oracle-fb + independent, WITH REASONING RETENTION
Cut 2026-08-12. The first generation round whose outputs keep the gpt-oss
analysis channel (reasoning) on disk — see tts-sft/docs/REASONING_RETENTION.md.
Rounds 2–3 saved only the harmony final channel; their reasoning is unrecoverable.
What round 4 is
The combined pool — round-2 rerun pool (4,322 problems, apps-*/cc-*) +
round-3 pool (2,833 problems, cc3-*), zero id overlap, 7,155… See the full description on the dataset page: https://huggingface.co/datasets/tts-sft/round4-independent.OpenWebRL-SFT-Trajectories
OpenWebRL SFT Trajectories
Dataset Summary
OpenWebRL SFT Trajectories contains successful browser-agent trajectory data collected by Qwen3-VL-235B-A22B-Thinking. These trajectories are used to supervised fine-tune the initial OpenWebRL visual web agent. The SFT checkpoint is then used as the initialization checkpoint for OpenWebRL's subsequent online multi-turn reinforcement learning stage.
OpenWebRL trains visual web agents on live websites. The full training… See the full description on the dataset page: https://huggingface.co/datasets/OpenWebRL/OpenWebRL-SFT-Trajectories.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.sailor2-sft-stage1toy-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.2026-07-31-toolcalling-tulu-sft-run
Run record — Qwen3.6-27B tool-calling 20/80 SFT
Everything the training run produced except the weights: the TRL log history, the resolved
config, the environment, the loss/accuracy figure and its greppable markdown mirror.
The adapter is at LASR-Callum/2026-07-31-wrongly-trained-qwen36-toolcalling-tulu-lora-20-80; the training data is
at LASR-Callum/2026-07-31-toolcalling-tulu-20-80-mixture.
Required metadata
field
value
experiment
One bf16 LoRA SFT… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-07-31-toolcalling-tulu-sft-run.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.OpenO1_SFT_ultra_BoN_rewardedpi-agent-traces-sft
pi-agent-traces-sft
Merged supervised-fine-tuning dataset of coding-agent session traces (the "pi"
harness format), converted to conversational format for Unsloth SFT of Qwen3-family
models.
Files
pi_sft_qwen3_24k.jsonl — recommended training file. Every session fits in
24,576 tokens (Qwen3 tokenizer). 5,717 sessions, max length 24,564 tokens.
pi_sft_qwen3.jsonl — full unfiltered set (6,929 sessions, up to ~632k tokens).
Sources (merged)… See the full description on the dataset page: https://huggingface.co/datasets/vojtavlas2/pi-agent-traces-sft.LogosForge-scored-sft-v1
LogosForge-scored-sft-v1
This dataset is a scored Supervised Fine-Tuning (SFT) distillation corpus built on top of the Natural Reasoning question set.It is constructed in two stages:
First, a large-scale reasoning-oriented teacher model (gpt-oss-120B-high) is used to generate distilled student responses, including explicit chain-of-thought reasoning, for natural reasoning questions.
Second, these distilled responses are evaluated by a separate instruction-following model… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/LogosForge-scored-sft-v1.thomas-yanxin-MT-SFT-ShareGPT-sample
MT-SFT-ShareGPT Sample Dataset
This dataset provides a sample of the thomas-yanxin/MT-SFT-ShareGPT dataset with English and Chinese subsets.
Dataset Contents
train.jsonl: Contains 1/10 of the original data, shuffled
EN.jsonl: English conversations from train.jsonl
ZH.jsonl: Chinese conversations from train.jsonl
Each row represents a conversation with an optional system message, followed by human and GPT turns.
Columns from the original dataset are preserved, with… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/thomas-yanxin-MT-SFT-ShareGPT-sample.thomas-yanxin-MT-SFT-ShareGPT
thomas-yanxin/MT-SFT-ShareGPT
This is the complete thomas-yanxin/MT-SFT-ShareGPT dataset,
with duplicates removed and the entire dataset shuffled. Sensitive data has been redacted.
For practical work, consider using agentlans/thomas-yanxin-MT-SFT-ShareGPT-sample
which is smaller and split by language.
lm-eval-results-alnrg2arg-test3_sft_16bit-private
Dataset Card for Evaluation run of alnrg2arg/test3_sft_16bit
Dataset automatically created during the evaluation run of model alnrg2arg/test3_sft_16bit
The dataset is composed of 62 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 2 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/nyu-dice-lab/lm-eval-results-alnrg2arg-test3_sft_16bit-private.k3-sft-cc0-flan
Dataset Card for K3 SFT CC0 FLAN
844-row Kimi K3 synthetic instruction-tuning shard built from DPI-traced CC0/public-domain
FLAN prompts in the Tülu mix. Four overlapping Hub configs expose different cohort
views; adaptive is the recommended default for quality-conscious SFT mixing.
Dataset Details
Curated by: Training Datasmith
Teacher: kimi-k3 via deltafin (local inference)
Languages: English prompts; translation pairs include German, Spanish, Czech, Igbo… See the full description on the dataset page: https://huggingface.co/datasets/Training-Datasmith/k3-sft-cc0-flan.