datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Think-Bench
THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models
Official repository for "THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models".
For more details, please refer to the project page with dataset exploration and visualization tools.
[Paper] [Github] [ModelScope Dataset] [Visualization]
👀 About Think-Bench
Reasoning models have made remarkable progress in complex tasks… See the full description on the dataset page: https://huggingface.co/datasets/zhiyuan218/Think-Bench.Chinese-Qwen3-235B-Thinking-2507-Distill-100k
📌 Note: The English translation of this dataset card is provided below.
Chinese-Qwen3-235B-Thinking-2507-Distill-100k
Dataset Summary
Chinese-Qwen3-235B-Thinking-2507-Distill-100k 是一个包含约 100k 条高质量中文推理与指令数据的数据集，由 Qwen-3-235B-A22B-Thinking-2507（官方 Thinking 模式，上下文长度 32K）蒸馏生成。
该数据集覆盖了多个重要领域：
数学与工程任务（Mathematics, Applied Math, Advanced Math）
通用知识与写作（General Knowledge, Language & Writing）
技术与编程（Technology & Programming）
商业与经济（Business & Economics）… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/Chinese-Qwen3-235B-Thinking-2507-Distill-100k.Kimi-K2.6-Thinking-200x
Dataset Card (Kimi-K2.6-Thinking-200x)
Dataset Summary
Kimi-K2.6-Reasoning-207 is a high-quality distilled reasoning dataset designed for supervised fine-tuning (SFT) of small language models.
This dataset uses a curated seed question set covering Mathematics, Code, Logic, Science, Analysis, and Instruction-following domains. By calling the Kimi-K2.6 model via the Moonshot AI API as the teacher model, it generates high-quality responses featuring long-form step-by-step… See the full description on the dataset page: https://huggingface.co/datasets/uniquealexx/Kimi-K2.6-Thinking-200x.CodeX-Thinking-Gemma-4-31B-ITAll prompts were taken from Modotte/CodeX-2M-Thinking, which contains multiple traces per prompt whereas this dataset only provides one trace per prompt. Generations were with https://huggingface.co/nvidia/Gemma-4-31B-IT-NVFP4 (a mix of BF16/FP8 weights that NVIDIA configured with FP8 KV cache; benchmarks show performs similarly to BF16 for coding). No system prompt was used.
Think-Observe
Think-Observe Dataset
Welcome to the Think-Observe dataset! This dataset is designed to fine-tune Large Language Models (LLMs) to improve their accuracy and ability to reflect before providing answers. It includes a collection of questions and answers tagged with think and observe tags.
Dataset Overview
The Think-Observe dataset consists of pairs of questions and answers, with each entry tagged to signify its role in the model's training process:
think: This tag is used… See the full description on the dataset page: https://huggingface.co/datasets/ayush-thakur02/Think-Observe.pl-thinking-r1thinking-cot-1k
Thinking CoT 1K
A production-style synthetic dataset for reasoning-focused LLM/SLM training.
This dataset provides 1,000 instruction-response samples with explicit thinking_cot traces for multi-domain supervision:
Basic arithmetic reasoning (200)
Advanced math reasoning (200)
General chat + explanation behavior (400)
Coding and snippet generation tasks (200)
Why this dataset
Reasoning-focused fine-tuning often needs a clean, structured mix of:
Deterministic numeric… See the full description on the dataset page: https://huggingface.co/datasets/Sharjeelbaig/thinking-cot-1k.kayra-thinking-data
Turkish Thinking Dataset (kayra-thinking-data)
A Turkish chain-of-thought reasoning dataset combining 4 sources, deduplicated and cleaned. Designed for fine-tuning small language models (86M-1B parameters) to produce <|think|> reasoning before answers.
Dataset Overview
Metric
Value
Total Examples
4,191
Format
JSONL ({"text": "..."})
Max Length
512 tokens (Kayra tokenizer)
Avg Length
~640 tokens (cl100k), ~340 tokens (Kayra)
Language
Turkish… See the full description on the dataset page: https://huggingface.co/datasets/sixfingerdev/kayra-thinking-data.Pashto-Thinking
Pashto-Thinking
clean_multilingual_thinking元データ: https://huggingface.co/datasets/HuggingFaceH4/Multilingual-Thinking
使用したコード: https://github.com/LLMTeamAkiyama/0-data_prepare/tree/master/src/Multilingual-Thinking
データ件数: 197
平均トークン数: 872
最大トークン数: 2,339
合計トークン数: 171,812
ファイル形式: JSONL
ファイル分割数: 1
合計ファイルサイズ: 748.5 KB
加工内容：
フィルタリングによるデータクレンジング
言語フィルタリング: reasoning_languageが「English」のデータのみを抽出します。
文字数フィルタリング: 処理速度の観点から、question（質問）、thought（思考）、answer（回答）の各フィールドで、規定の文字数を超える長大なデータは事前に除外します。
繰り返し表現の除去:… See the full description on the dataset page: https://huggingface.co/datasets/LLMTeamAkiyama/clean_multilingual_thinking.Think_and_Query_value_for_R1
Introduction
This repository implements a Shapley value-based approach to quantitatively evaluate the contributions of query (q) and think (t) in generating answer (a).
Method
think_value = [loss(a|q) - loss(a|q,t) + loss(a|∅) - loss(a|t)] / 2
query_value = [loss(a|t) - loss(a|q,t) + loss(a|∅) - loss(a|q)] / 2
think_ratio = think_value/loss(a|∅)
query_ratio = query_value/loss(a|∅)
Original dataset… See the full description on the dataset page: https://huggingface.co/datasets/caihuaiguang/Think_and_Query_value_for_R1.