datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
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-student-fail-v41-clean-thinking
Nemotron-fail / DeepSeek-V4.1 clean and action-only trajectories
DeepSeek-V4.1 reward-1 trajectories for tasks on which the Nemotron student
did not obtain reward 1. This release was rebuilt from the complete reward-1
audit under v54-high-precision-canonical-reconstruction-relations.
Training paths
Path
Rows
Unique tasks
Thinking
Use
data/strict/train.jsonl.gz
12
12
Preserved and clean
Raw-thinking SFT
data/hybrid/train.jsonl.gz
58
58
Only… See the full description on the dataset page: https://huggingface.co/datasets/zhiyuanhucs/nemotron-student-fail-v41-clean-thinking.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.grug-think
grug-think
grug make dataset. dataset make model think like grug. grug think short. short think cheap. cheap think good.
big-brain model think 400 token before poke one tool. grug model think 11 word. same tool poke. same work done. many token saved. token = money. grug like money stay in pocket.
what in box
100,891 example. every example = full agent conversation: system, user, assistant, tool message. assistant turn always got <think>grug reasoning</think> first… See the full description on the dataset page: https://huggingface.co/datasets/ProCreations/grug-think.thinking-cap-tier-curricula-complete
Thinking Cap Tier Curricula — Complete Reasoning Alignment Suite (TCS v4)
[!IMPORTANT]
Dataset Release v1.2 (Sept 2026) — Clean Delimiters & Zero-Padding Architecture:
In v1.2, all 13,477 SFT samples and 3,187 SimPO preference pairs have undergone an automated token purge:
Zero <|pad|> batch residues: 100% eliminated across all files.
Zero reasoning leakage into final answers: Deliberation stays strictly inside <think>...</think>, and answers provide direct, non-repetitive… See the full description on the dataset page: https://huggingface.co/datasets/Davd-b01/thinking-cap-tier-curricula-complete.thinking-cap-tier-lima-dense
Thinking Cap Tier Curricula — LIMA Hyper-Dense Reasoning Alignment Suite (TCS v4)
[!IMPORTANT]
Dataset Release v1.2 (Sept 2026) — Clean Delimiters & Zero-Padding Architecture:
In v1.2, all 5,500 SFT and 2,000 SimPO records have undergone a complete token purge:
Zero <|pad|> batch residues: 100% eliminated across all records.
Zero reasoning leakage into final answers: Deliberation stays strictly inside <think>...</think>, and answers provide direct conclusions.
Native ChatML… See the full description on the dataset page: https://huggingface.co/datasets/Davd-b01/thinking-cap-tier-lima-dense.thinking-cap-tier-raw-traces
Thinking Cap Tier Raw Traces (TCS v4)
[!IMPORTANT]
Dataset Release v1.2 (Sept 2026) — Clean Delimiters & Zero-Padding Architecture:
All 38,158 candidate reasoning traces across all 4 tiers (candidates_low.jsonl, candidates_mid.jsonl, candidates_high.jsonl, candidates_xhigh.jsonl) are 100% sanitized:
Zero batch-padding residues (<|pad|>): Completely purged across all records.
Strict Delimiter Integrity: Generation blocks cleanly separate thought deliberation tags… See the full description on the dataset page: https://huggingface.co/datasets/Davd-b01/thinking-cap-tier-raw-traces.thinkingcap-condensed-qwen3.8-glm5.2-kimi-k3
ThinkingCap Condensed — Qwen3.8 / GLM-5.2 / Kimi-K3
Condensed ThinkingCap-style reasoning traces for SFT.
1,985 traces: each row pairs a full multi-turn teacher trace (Qwen3.8-Max,
GLM-5.2 or Kimi K3, via
r0b0tlab/qwen3.8-max-glm5.2-kimi-k3-distillation)
with a condensed TC-style version (short <think> + definitive numbered
answer) generated by
bottlecapai/ThinkingCap-Qwen3.6-27B
using the thinkingcap system prompt.
Format: JSONL (data/condensed.jsonl), 1,985 rows, UTF-8.… See the full description on the dataset page: https://huggingface.co/datasets/Davd-b01/thinkingcap-condensed-qwen3.8-glm5.2-kimi-k3.CT-RATE-Thinking
CT-RATE-Thinking: Reasoning-Augmented CT Report Dataset
🎉🎉🎉 Our paper was accepted at the 28th conference of The Medical Image Computing and Computer Assisted Intervention Society (MICCAI). See you in Daejeon, Korea, September 23–27, 2025.CT-RATE-Thinking is a reasoning-augmented dataset derived from CT-RATE, containing chain-of-thought VQA pairs and report-level thinking narratives for 3D chest CT volumes.
It was generated as part of the μ²Tokenizer project… See the full description on the dataset page: https://huggingface.co/datasets/AlpachinoNLP/CT-RATE-Thinking.HQ-Chat-2k
🧠 HQ-Chat-2K — High-Quality Conversational & Instruction-Tuning Dataset
2,000 carefully curated, high-quality conversation and instruction examples for fine-tuning Small Language Models (SLMs) and compact LLMs from ~500M to 3B parameters.
HQ-Chat-2K is a high-quality conversational and instruction-tuning dataset designed specifically for training and fine-tuning small to medium-sized Large Language Models (LLMs).
The dataset contains 2,000 curated user–assistant examples… See the full description on the dataset page: https://huggingface.co/datasets/ThinkNet/HQ-Chat-2k.thinkingcap-reasoning-traces
ThinkingCap Reasoning Traces (Legacy v1 Prototype)
[!WARNING]
Legacy / Deprecated Prototype Notice (v1):
This dataset represents an early exploratory prototype (v1, 4,254 traces) from initial development.
Some samples in this legacy version contain early formatting artifacts, including reasoning traces leaking into the final answer field and informal step-by-step breakdowns.
For modern post-training, SFT, and SimPO alignment under the TCS v4 cognitive standard, please use our… See the full description on the dataset page: https://huggingface.co/datasets/Davd-b01/thinkingcap-reasoning-traces.2026-08-02-qwen36-mixture-500k-numina-heavy-empty-think
Qwen3.6-27B SFT mixture — 500k maths-weighted, empty-think markers
499,595 tokens across 1,001 conversations, weighted toward maths, with Qwen3.6's empty
think marker on the non-maths rows. md5 c433f31eba2b5b4919fb166043caccb5.
Source
Examples
Tokens
Share
Marker
NuminaMath-CoT
611
333,351
66.9%
no
No Robots
271
82,239
16.5%
yes
TULU3
119
82,445
16.5%
yes
Total
1,001
499,595
390 marked
Derived from
qwen3.6-27b-mixture-500k-numina-heavy
by adding the… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-08-02-qwen36-mixture-500k-numina-heavy-empty-think.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.TeichAI-thinking-reasoning-x
TeichAI Thinking & Reasoning Datasets
A collection of prompts answered by large language models (LLMs) such as Google Gemini and OpenAI ChatGPT, with long-form reasoning enabled.
These datasets were originally created by TeichAI for distillation and reasoning-focused training workflows.
Schema
Each row in the dataset has the following fields:
question_hash: Truncated, base64-encoded MD5 hash of the question, useful for filtering and deduplication.
question: The… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/TeichAI-thinking-reasoning-x.grug-think-v3-10k
grug-think-v3-10k
v2 brain short. v2 brain useful. but some v2 brain wear office shirt.
"User wants hello world Python. Provide code." short English, yes. grug, no.
v3 tear off office shirt. keep brain meat.
old: User wants hello world Python. Simple code snippet, no tools needed. Provide code and brief explanation.
new: Need Python hello-world. Tiny snippet. No tool. Give code, brief explain.
complex cave different. grug no crush branch into pebble. exact path, error… See the full description on the dataset page: https://huggingface.co/datasets/ProCreations/grug-think-v3-10k.Qwen3.8-27B-Thinking-SecOPD-trainset
Qwen3.6-27B-Thinking SecOPD Trainset
Dataset summary
This public dataset contains 19,155 complete, model-specific preference records
for offline adversarial training against indirect prompt injection. The corpus
starts from the 19,157-record
Sizhe-Chen/Qwen3.6-27B-Instruct-SecPO-trainset
release. Its six non-label lineage fields are retained, while the attacked
prompts are rendered for thinking-on generation and the chosen and rejected
labels are regenerated… See the full description on the dataset page: https://huggingface.co/datasets/Sizhe-Chen/Qwen3.8-27B-Thinking-SecOPD-trainset.GLM-5.2-FP8-nemotron-codealpaca-thinking
GLM-5.2-FP8 Nemotron-CodeAlpaca Thinking Dataset
820,790 single-turn conversations generated by zai-org/GLM-5.2-FP8
with thinking enabled.
Prompt source
Rows (public)
Nemotron-Post-Training-Dataset-v2
800,944
CodeAlpaca-20k (corrected prompts, instruction + "\n\n" + input)
19,846
Total
820,790
Generation: temperature=1.0, top_p=0.95, max_tokens=24576, thinking
enabled. The CodeAlpaca prompts here include the input field.
Relationship to… See the full description on the dataset page: https://huggingface.co/datasets/JessieWei/GLM-5.2-FP8-nemotron-codealpaca-thinking.OpenThought3-Qwen3-4BOpenThought3-Qwen3-4B
OpenThought3-Qwen3-4B is a math reasoning supervised fine-tuning dataset in chat-message JSONL format.
Data Creation and Cleaning
This dataset was generated by Qwen3-4B (Non-thinking) from math-domain prompts selected from OpenThoughts3-1.2M. The generated responses were cleaned through deduplication, removal of degenerate repetition/repeater-style outputs, and template checks on the assistant… See the full description on the dataset page: https://huggingface.co/datasets/Thinking-Space/OpenThought3-Qwen3-4B.r8-thinking-fix-sft
⚠️ CRITICAL: Ollama Inference Flag Required for derived models
If you train or serve any Qwen3.5-9B-derived model from this lineage via Ollama,
you MUST pass "think": false in /api/chat requests for chat / instruction following / tool use.
The qwen3.5 RENDERER auto-injects <think> tags causing 25-46% empty-answer rates without this flag.
See dataset cudabenchmarktest/r9-research-framework/_OLLAMA_INFERENCE_WARNING.md for the full lesson learned.
R8 Thinking-Fix SFT… See the full description on the dataset page: https://huggingface.co/datasets/cudabenchmarktest/r8-thinking-fix-sft.Dataset_of_Russian_thinkingRu
RTD
Описание:Russian Thinking Dataset — это набор данных, предназначенный для обучения и тестирования моделей обработки естественного языка (NLP) на русском языке. Датасет ориентирован на задачи, связанные с генерацией текста, анализом диалогов и решением математических и логических задач.
Основная информация:
Сплит: train
Количество записей: 147.046
Цели:
Обучение моделей пониманию русского языка.
Создание диалоговых систем с естественным взаимодействием.… See the full description on the dataset page: https://huggingface.co/datasets/qwqeqw/Dataset_of_Russian_thinking.NPC-RP-Post-Thinking
AIIDE-POST-THINKING
This is the post-thinking dataset for our paper accepted as a poster presentation to AIIDE-2026
Dataset Details
This is the training dataset for chimbiwide/Gemma3-4B-post-thinking
Corresponding Links
Repository: [To be updated]
Paper: [To be updated]
Demo: [To be updated]
Uses
Suprevised-Finetuning
Dataset Creation
For more details, consult our paper.
Citation
If you… See the full description on the dataset page: https://huggingface.co/datasets/chimbiwide/NPC-RP-Post-Thinking.MBPP-Thinking-Gate-1k
MBPP Thinking-Gate SFT Dataset
This package contains two related assets:
Ready 1,000-row MBPP-style dataset (all.jsonl, train.jsonl, validation.jsonl).
It is synthetic and designed to test/train autonomous routing between <DIRECT> and <THINK>.
Official-MBPP builder (build_from_official_mbpp.py).
Run this to create the production dataset from the official Google Research MBPP source.
Why two response modes?
The training target starts with one of two routing… See the full description on the dataset page: https://huggingface.co/datasets/islam-kamel/MBPP-Thinking-Gate-1k.allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
PJMixers-Dev/allenai_WildChat-1M-prompts with responses generated with gemini-2.0-flash-thinking-exp-1219.
Generation Details
If BlockedPromptException, StopCandidateException, or InvalidArgument was returned, the sample was skipped.
If ["candidates"][0]["safety_ratings"] == "SAFETY" the sample was skipped.
If ["candidates"][0]["finish_reason"] != 1 the sample was skipped.
model =… See the full description on the dataset page: https://huggingface.co/datasets/PJMixers-Dev/allenai_WildChat-1M-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT.synthetic-self-correction-and-thinking-samples
Self Correction and Thinking
A seed library for training language models to reason with self-correction.
Teaches three reasoning behaviors -- catching your own errors, verifying correct answers, and rejecting false doubts -- across four domains, three difficulty tiers, and three reasoning modes. Also includes multi-turn user-correction conversations where the user actively corrects or challenges the assistant.
The structure at a glance
graph TB… See the full description on the dataset page: https://huggingface.co/datasets/sbussiso/synthetic-self-correction-and-thinking-samples.Strandset-Rust-Think-TR
🦀 Strandset-Rust-Think-TR (5K Cleaned & Translated)
Strandset-Rust-Think-TR, Rust programlama dili odaklı, Türkçe düşünme zinciri (Chain-of-Thought / <think>) adımları içeren 5.000 adet yüksek kaliteli talimat (instruction-tuning) örneğinden oluşan bir veri setidir.
Bu veri seti, snowmead/Strandset-Rust-Think çalışması temel alınarak WrittenWithRust tarafından Qwen3.8-27B modeli yardımıyla Türkçe dikeyine kazandırılmış ve mükerrer kayıtlarından arındırılmıştır.
⚙️… See the full description on the dataset page: https://huggingface.co/datasets/WrittenWithRust/Strandset-Rust-Think-TR.Thinking-multilingual-big-10k-sft
A dataset based off of openo1 math, 500 examples translated to 23 different languages. filtered out un-translated examples.
enjoy 👍
agentic-think-v1
agentic-think-v1
Agentic tool-calling SFT corpus with teacher-generated <think> reasoning on every assistant round.
212,396 training rows across 9 sources, built by ENERZAi for small-model (1.7B ternary) agentic SFT.
Each row is a full multi-turn conversation: system prompt, user turns, assistant turns
(each carrying its reasoning in a separate reasoning_content field), tool calls and tool results.
Per-source splits (by_source config)
Each source file is also… See the full description on the dataset page: https://huggingface.co/datasets/ENERZAiKR/agentic-think-v1.bghira_pseudo-camera-10k-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
bghira_pseudo-camera-10k-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT
bghira/pseudo-camera-10k with responses/captions generated with gemini-2.0-flash-thinking-exp-1219.
The format should be similar to that of liuhaotian/LLaVA-Instruct-150K.
Images can be found in the images.zip folder. The zip also contains .txt captions for ease of use in non-VQA tasks.
Generation Details
If BlockedPromptException, StopCandidateException, or InvalidArgument was returned, the… See the full description on the dataset page: https://huggingface.co/datasets/PJMixers-Images/bghira_pseudo-camera-10k-gemini-2.0-flash-thinking-exp-1219-CustomShareGPT.sft-repro-thinking-step630-nemotron-terminal-step1888-openthoughts-tblite-2026-08-13
Nemotron Terminal SFT reproduction evaluation artifacts
This repository contains the complete Harbor artifact tree for the 300-trial
OpenThoughts-TBLite evaluation of
laion/sft-repro-thinking-step630-nemotron-terminal-step1888.
The checkpoint was trained from the Grug stage-2 thinking checkpoint on the
Nemotron Terminal corpus for 1,888 steps.
Result
Measure
Value
Attempted / completed
300 / 300
Verifier-scoreable
259 (86.33%)
Aggregate reward, all… See the full description on the dataset page: https://huggingface.co/datasets/laion/sft-repro-thinking-step630-nemotron-terminal-step1888-openthoughts-tblite-2026-08-13.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.