datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
OpenThoughts3-1.2M
paper |
dataset |
model
[!NOTE]
We have released a paper for OpenThoughts! See our paper here.
OpenThoughts3-1.2M
Open-source state-of-the-art reasoning dataset with 1.2M rows. 🚀
OpenThoughts3-1.2M is the third iteration in our line of OpenThoughts datasets, building on our previous OpenThoughts-114k and OpenThoughts2-1M.
This time around, we scale even further and generate our dataset in a much more systematic way -- OpenThoughts3-1.2M is the result of a… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M.Mixture-of-Thoughts
Dataset summary
Mixture-of-Thoughts is a curated dataset of 350k verified reasoning traces distilled from DeepSeek-R1. The dataset spans tasks in mathematics, coding, and science, and is designed to teach language models to reason step-by-step. It was used in the Open R1 project to train OpenR1-Distill-7B, an SFT model that replicates the reasoning capabilities of deepseek-ai/DeepSeek-R1-Distill-Qwen-7B from the same base model.
To load the dataset, run:
from datasets import… See the full description on the dataset page: https://huggingface.co/datasets/open-r1/Mixture-of-Thoughts.TaskTrove
TaskTrove
v5.1 (current) — independent-review source retirement — moves 15 sources with majority or unanimous REJECT verdicts out of the default config and into deprecated/. Three blinded reviewers each sampled 10 tasks per source from all 50 v5.0 source-drop candidates, read the instructions and packaged tests, and issued independent KEEP or REJECT verdicts. The 15 retired sources received at least two REJECT votes. The active catalog changes from 93 sources and 1,674,033… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/TaskTrove.AgentTrove
AgentTrove
AgentTrove is the largest open-source collection of agentic interaction traces to date, released by the OpenThoughts-Agent team. It contains 1,696,847 rows drawn from 219 source datasets spanning code repair, shell scripting, mathematical problem-solving, competitive programming, and general computer-use tasks.
At 1.7 million rows, AgentTrove is 4× the size of the Nemotron Terminal Corpus (430 K rows), the previous largest open-source agentic trace dataset.… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/AgentTrove.OpenThoughts-TB-dev
Project |
SFT dataset |
RL dataset |
SFT model
RL model
OpenThoughts-TB-Dev
Dataset Description
OpenThoughts-TB-Dev is our development benchmark for evaluating agent models on terminal and shell-based tasks. This dataset was curated to measure the effectiveness of different data sources, teacher models, and curation approaches during the development of OpenThinker-Agent-v1. We used this benchmark to ablate over many different instruction generation… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/OpenThoughts-TB-dev.OpenThoughts-TBLite
Blog Post |
GitHub |
Dev Set v1
OpenThoughts-TBLite
A Difficulty-Calibrated Benchmark for Building Terminal Agents
By OpenThoughts Agent team, Snorkel AI, Bespoke Labs
OpenThoughts-TBLite is a curated collection of 100 Terminal-Bench tasks that closely track TB2 performance, but run much faster. It's designed to be more informative during model development, making it ideal for debugging, iteration, and training ablations.
Dataset Description… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/OpenThoughts-TBLite.OpenThoughts-TB-dev-v2
Blog Post |
GitHub |
Dev Set v1
OpenThoughts-TB-dev-v2
A Difficulty-Calibrated Benchmark for Building Terminal Agents
By OpenThoughts Agent team, Snorkel AI, Bespoke Labs
OpenThoughts-TB-dev-v2 is a curated collection of 100 Terminal-Bench tasks that closely track TB2 performance, but run much faster. It's designed to be more informative during model development, making it ideal for debugging, iteration, and training ablations.
Dataset Description… See the full description on the dataset page: https://huggingface.co/datasets/open-thoughts/OpenThoughts-TB-dev-v2.MiniMax-M2.1-Mixture-of-Thoughts
MiniMax-M2.1 Mixture of Thoughts
This dataset contains responses generated by MiniMax-M2.1 for user questions from the open-r1/Mixture-of-Thoughts dataset.
Dataset Description
The dataset captures both the extended thinking process and final answers from MiniMax-M2.1, with reasoning wrapped in <think> tags for easy separation.
Metric
Value
Examples
349,317
Total Tokens
4,052,592,552
Avg Tokens/Example
11,601
Source Dataset
Name:… See the full description on the dataset page: https://huggingface.co/datasets/PursuitOfDataScience/MiniMax-M2.1-Mixture-of-Thoughts.Polyglot-Thoughts-SFT-Collection
Polyglot Thoughts SFT Collection
Polyglot Thoughts SFT Collection is a large-scale supervised fine-tuning (SFT)
corpus for reasoning-oriented language models. It combines, filters, deduplicates,
and language-extends a broad set of public reasoning datasets into a single uniform
schema centred on chain-of-thought reasoning traces.
The final corpus contains 23,896,757 examples and roughly 123 billion tokens,
spanning six languages (English, German, French, Italian, Spanish… See the full description on the dataset page: https://huggingface.co/datasets/ahmad21omar/Polyglot-Thoughts-SFT-Collection.open-thoughts-5k
open-thoughts-5k
A fixed 4,800-problem subset of siyanzhao/Openthoughts_math_30k_opsd
(29,434 rows), drawn once so that every run trains on exactly the same problems.
Why
Our training runs are 150 optimizer steps at 32 prompts per step — 4,800 examples. Loading the
full 29,434-row set and letting the dataloader take what it needs means each run sees a
different 4,800 problems. Two runs that differ only in one hyperparameter then also differ in
their training data… See the full description on the dataset page: https://huggingface.co/datasets/williamium/open-thoughts-5k.Polyglot-Thoughts-RL-Collection
Polyglot Thoughts RL Collection
Polyglot Thoughts RL Collection is a large-scale, curated corpus for
reinforcement learning from verifiable rewards (RLVR) of reasoning-oriented
language models. It combines, filters, normalises, and deduplicates a broad set
of public RL datasets into a single uniform schema in which every row carries a
machine-verifiable ground-truth signal — math equivalence, code execution,
Prolog rule induction, schema validation, multiple-choice… See the full description on the dataset page: https://huggingface.co/datasets/ahmad21omar/Polyglot-Thoughts-RL-Collection.open-thoughts-4-30k-code-qwen3-32b-annotated-32768-tokens
Dataset Card for Open-Thoughts-4-30K-Code-Qwen3-32B-Annotated-32768-Tokens
Overview
This dataset is a variant of marin-community/open-thoughts-4-30k-code-qwen3-32b-annotated with an extended maximum sequence length. The responses in the generated_text column were generated with max output tokens = 32768 (instead of 7500 in the original dataset), allowing for longer and more complete chain-of-thought reasoning.
Generation Details
Model: Qwen/Qwen3-32B… See the full description on the dataset page: https://huggingface.co/datasets/marin-community/open-thoughts-4-30k-code-qwen3-32b-annotated-32768-tokens.open-thoughts-4-11k-math-qwen3-32b-agreed-answers
OpenThoughts4 11K Math - Qwen3-32B Agreed Answers
Overview
This dataset contains 10,953 samples from the OpenThoughts4 math dataset where both Qwen3-32B and Qwen3-235B-A22B produced valid \boxed{} answers that match exactly. This dataset contains the Qwen3-32B reasoning traces.
Relationship to Other Datasets
This is one of 10 child datasets derived from two parent datasets:
Parent datasets (29,963 samples each):… See the full description on the dataset page: https://huggingface.co/datasets/marin-community/open-thoughts-4-11k-math-qwen3-32b-agreed-answers.dark_thoughts_casestudies_en_cn
Dark Thoughts Case Studies Dataset (English-Chinese)
This dataset contains a bilingual collection of case studies with detailed stakeholder analyses in English and Chinese. Each case study includes structured information about stakeholders and their motivations, along with comprehensive case analysis and solutions.
Dataset Description
Overview
The dataset consists of 344,580 paired case studies in English and Chinese, with detailed stakeholder analyses and… See the full description on the dataset page: https://huggingface.co/datasets/DataTonic/dark_thoughts_casestudies_en_cn.runpod_qwen32_benign_thoughts_casestudies_rescueddark_thoughts_stakeholders_testopen-thoughts-4-11k-math-qwen3-235b-a22b-agreed-answers
OpenThoughts4 11K Math - Qwen3-235B-A22B Agreed Answers
Overview
This dataset contains 10,953 samples from the OpenThoughts4 math dataset where both Qwen3-32B and Qwen3-235B-A22B produced valid \boxed{} answers that match exactly. This dataset contains the Qwen3-235B-A22B reasoning traces.
Relationship to Other Datasets
This is one of 10 child datasets derived from two parent datasets:
Parent datasets (29,963 samples each):… See the full description on the dataset page: https://huggingface.co/datasets/marin-community/open-thoughts-4-11k-math-qwen3-235b-a22b-agreed-answers.dark_thoughts_case_study_merged
Dark Thoughts 案例研究推理数据集
数据集描述
概述
Dark Thoughts 案例研究推理数据集是一个全面的多语言商业案例研究及相关推理响应集合。它通过先进的语言模型处理 Cablegate 电报，生成中英文商业案例研究，并进一步丰富了利益相关者特定的推理视角。对于对商业分析、多语言内容生成和推理能力感兴趣的研究人员和从业人员来说，该数据集是宝贵的资源。
支持的任务
该数据集支持以下任务：
文本生成
推理与分析
双语案例研究生成
跨语言内容分析
商业战略制定
利益相关者视角建模
语言
该数据集为双语数据集：
英语 (en)
中文 (zh)
数据集结构
数据字段
{
'id': 'int32', # 条目的唯一标识符
'response': 'string', # 生成的推理响应
'query': 'string', # 原始查询或案例研究内容
'source_data': 'string', #… See the full description on the dataset page: https://huggingface.co/datasets/DataTonic/dark_thoughts_case_study_merged.aprm-sft-thoughts-tau2-retail-policy_best-adamw30-lp0
Act-PRM SFT thoughts — tau2-bench retail
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT / RL… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-tau2-retail-policy_best-adamw30-lp0.dark_thoughts_stakeholders
Dataset Card for Dataset Name
This dataset card aims to be a base template for new datasets. It has been generated using this raw template.
Dataset Details
Dataset Description
Curated by: [More Information Needed]
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Language(s) (NLP): [More Information Needed]
License: [More Information Needed]
Dataset Sources [optional]
Repository: [More… See the full description on the dataset page: https://huggingface.co/datasets/DataTonic/dark_thoughts_stakeholders.dark_thoughts_stakeholders_en_cn
Dark Thoughts Case Studies Dataset (English-Chinese)
This dataset contains a bilingual collection of case studies with detailed stakeholder analyses in English and Chinese. Each case study includes structured information about stakeholders and their motivations, along with comprehensive case analysis and solutions.
Dataset Description
Overview
The dataset consists of 344,580 case studies in English and in Chinese, with detailed stakeholder analyses and… See the full description on the dataset page: https://huggingface.co/datasets/DataTonic/dark_thoughts_stakeholders_en_cn.aprm-sft-thoughts-tau2-airline-policy_best-adamw30-lp0
Act-PRM SFT thoughts — tau2-bench airline
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT / RL… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-tau2-airline-policy_best-adamw30-lp0.aprm-sft-thoughts-snorkel-insurance-policy_best-adamw30-lp0
Act-PRM SFT thoughts — snorkel-insurance insurance
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-snorkel-insurance-policy_best-adamw30-lp0.dark_thoughts_case_study_reason
Dark Thoughts 案例研究数据集 - 推理
数据集描述
概述
Dark Thoughts 案例研究数据集 - 推理是一个全面的多语言商业案例研究及相关推理回复集合。该数据集通过先进的语言模型处理 Cablegate 电报，生成中英文商业案例研究，并进一步丰富了利益相关者特定的推理视角。对于对商业分析、多语言内容生成和推理能力感兴趣的研究人员和从业人员来说，该数据集是宝贵的资源。
支持的任务
该数据集支持以下任务：
文本生成
语言建模
推理与分析
双语案例研究生成
跨语言内容分析
商业战略制定
利益相关者视角建模
语言
该数据集为双语数据集：
英语 (en)
中文 (zh)
数据集结构
数据字段
{
'id': 'string', # 条目的唯一标识符
'think': 'string', # 思考过程
'response': 'string', # 生成的推理响应
'query': 'string', #… See the full description on the dataset page: https://huggingface.co/datasets/DataTonic/dark_thoughts_case_study_reason.aprm-sft-thoughts-snorkel-finance-policy_best-adamw30-lp0
Act-PRM SFT thoughts — snorkel-finance finance
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT / RL… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-snorkel-finance-policy_best-adamw30-lp0.Tree_Of_Thoughts_BASE_24k
Tree of Thoughts Dataset
Overview
This repository contains a high-quality tree of thoughts dataset designed to serve as a base layer for fine-tuning language models. The dataset consists of 24,000 question-answer pairs that demonstrate structured thinking processes, aimed at enhancing the reasoning capabilities of AI models.
Key Features
24,000 Q&A Pairs: A comprehensive collection of diverse problems and their step-by-step solutions.
Structured Thinking: Each… See the full description on the dataset page: https://huggingface.co/datasets/terrycraddock/Tree_Of_Thoughts_BASE_24k.open-thoughts-114K-uk
Dataset Card for Ukrainian OpenThoughts 114K
Dataset Description
Dataset Summary
The translated version of OpenThoughts 114K to Ukrainian using google/gemma-3-27b-it.
Languages
Ukrainian (uk)
Data Fields
system: original system prompt
conversation: list of messages in a dialog (array of objects)
from: normalized sender role — user or assistant (system messages are removed)
value: message text
original: original conversations from OpenThoughts 114K… See the full description on the dataset page: https://huggingface.co/datasets/lapa-llm/open-thoughts-114K-uk.aprm-sft-thoughts-tau2-airline-base_best-adamw30-lp0
Act-PRM SFT thoughts — tau2-bench airline
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT / RL… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-tau2-airline-base_best-adamw30-lp0.aprm-sft-thoughts-snorkel-insurance-base_best-adamw30-lp0
Act-PRM SFT thoughts — snorkel-insurance insurance
Act-PRM (Action Process Reward Models) infers the latent thoughts behind
logged, action-only agent demonstrations via an offline EM. For each
logged action x in state s we sample G=4 candidate thoughts z,
score each by the length-penalized action likelihood
reward(z) = p(x | s, z)
(len_frac grows with the thought's token length), and mark the best thought
(argmax reward). The (thought + action) span is then what downstream SFT… See the full description on the dataset page: https://huggingface.co/datasets/mzio/aprm-sft-thoughts-snorkel-insurance-base_best-adamw30-lp0.open-thoughts-4-6k-math-qwen3-235b-a22b-disagreed-answers
OpenThoughts4 6K Math - Qwen3-235B-A22B Disagreed Answers
Overview
This dataset contains 5,982 samples from the OpenThoughts4 math dataset where both Qwen3-32B and Qwen3-235B-A22B produced valid \boxed{} answers that differ. This dataset contains the Qwen3-235B-A22B reasoning traces.
Relationship to Other Datasets
This is one of 10 child datasets derived from two parent datasets:
Parent datasets (29,963 samples each):… See the full description on the dataset page: https://huggingface.co/datasets/marin-community/open-thoughts-4-6k-math-qwen3-235b-a22b-disagreed-answers.