datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SWE-Star
SWE-Star
Introduction
SWE-Star is a family of language models based on the Qwen2.5-Coder family and trained on the SWE-Star dataset. The dataset contains approximately 250k agentic coding trajectories distilled from Devstral-2-Small using SWE-Smith tasks.
The complete data generation, training, and evaluation pipeline is openly available in our GitHub repository, enabling anyone to reproduce our results.
Additional details are available in our blog posts.… See the full description on the dataset page: https://huggingface.co/datasets/LogicStar/SWE-Star.whittle-stop-kd
Correction - 27 August 2026
kd_mt_top32.npz is misaligned and must not be used. Its per-turn spans were
computed against a throwaway per-turn sequence and then stored against the full
conversation, so only 43 of 258 weight-8.0 positions land on the turn
terminator; the other 215 land on token 236, a partial UTF-8 byte.
kd_top32.npz is correctly aligned, but its tail weighting does not do what
the section below claims. The capture stops one position short of the
terminator, so… See the full description on the dataset page: https://huggingface.co/datasets/logic65/whittle-stop-kd.multi-zebra-logic
Dataset Card for the MultiZebraLogic dataset
This dataset includes zebra puzzles in 39 European and 5 non-European languages and in two sizes: 2x3 and 4x5. It can be used for evaluating logical reasoning ability.
The data has been generated using the code in this repo.
Dataset Details
Dataset Description
Zebra puzzles are a type of constraint satisfaction problem. They describe a number of objects, N_objects, that each have attributes… See the full description on the dataset page: https://huggingface.co/datasets/alexandrainst/multi-zebra-logic.SWE-Smith
A extended version of the original SWE-smith-py dataset with more problem descriptions!
logicnlg
LogicNLG Dataset
See the official wenhuchen/LogicNLG release on GitHub.
task110_logic2text_sentence_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task110_logic2text_sentence_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task110_logic2text_sentence_generation.GLM-5.2-Logic-Puzzles
GLM-5.2 · Logical Puzzles
6000x traces distilled from GLM-5.2 on High reasoning
Token Count: 5M~?
Distribution:
Puzzles:
•Tokenization blindless ex: counting the r's in strawberry
•Goal reasoning ex: the car wash test (theres no car wash question exactly just prompts like it so its not just benchmaxxing)
•Reading comprehension traps
•Temporal reasoning
•Many other categories not worth mentioning
Prompts… See the full description on the dataset page: https://huggingface.co/datasets/ianncity/GLM-5.2-Logic-Puzzles.task211_logic2text_classification
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task211_logic2text_classification
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks}… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task211_logic2text_classification.whittle-teacher32-complete-answers
Whittle teacher32: complete answers with per-token teacher logprobs
Research preview. Part of the Whittle compression campaign, a personal
research project. The compute for this project is self funded and donations
decide whether the next round happens: https://ko-fi.com/davida81328
What this is
Complete answers generated by Qwen3.8-27B (UD-Q5_K_XL via llama.cpp), each
ending on a real end-of-turn token because the answer is finished, with the
teacher's top-32… See the full description on the dataset page: https://huggingface.co/datasets/logic65/whittle-teacher32-complete-answers.task717_mmmlu_answer_generation_logical_fallacies
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task717_mmmlu_answer_generation_logical_fallacies
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task717_mmmlu_answer_generation_logical_fallacies.task210_logic2text_structured_text_generation
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task210_logic2text_structured_text_generation
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task210_logic2text_structured_text_generation.logic-grid-puzzles-training-pool
Logic grid puzzles training pool
Logic grid puzzles: a row of positions, a handful of attributes with one value per position, and a
list of clues that together admit exactly one arrangement. Two sets drawn for this pool by
generators run here under the seeds recorded below, and two public datasets read at the pinned
revisions named below, laid out twice. Train on either layer or on both.
pool.jsonl
Every source rewritten into one shape, 390945 rows, one JSON… See the full description on the dataset page: https://huggingface.co/datasets/Emulated-Inc/logic-grid-puzzles-training-pool.LogicHaystacks
Evaluation code:
def parse(x):
if '<answer>' in x and '</answer>' in x:
start = x.find('<answer>') + len('<answer>')
end = x.find('</answer>')
x = x[start:end]
lines = [i.lstrip('L').strip() for i in x.strip().strip('.').split(',')]
return [int(i) for i in lines if i.isnumeric()]
def jaccard(list1, list2):
intersection = len(list(set(list1).intersection(list2)))
union = (len(set(list1)) + len(set(list2))) - intersection
return… See the full description on the dataset page: https://huggingface.co/datasets/sileod/LogicHaystacks.LogiCoTThe instructions and demonstrations for building formal logical reasoning capable Generative Large Language models. CoT rationales are generated with the GPT-4 API.
For non-commercial research purposes only.
Update: Our updated paper has been accepted by the findings of EMNLP2023.
The dataset is hosted on the Huggingface Datasets. It is the only distribution channel we currently allow. You can download data examples from our Github Link
Important: To request the dataset, please
Submit an… See the full description on the dataset page: https://huggingface.co/datasets/datatune/LogiCoT.deepseek-r1-autonomous-math-logic-cot-2026
📐 Enterprise DeepSeek-R1 Autonomous Mathematical & Logic CoT SFT/DPO Dataset (2026)
High-precision multi-turn instruction tuning and preference optimization dataset with step-by-step hypothesis exploration, error discovery, and dynamic backtracking Chain-of-Thought (<thought>) reasoning trees for fine-tuning LLMs (DeepSeek-R1-Distill-Qwen, Qwen-2.5-Math, Llama-3.3, Mistral) into World-Class Olympiad Mathematicians and Formal Verification Agents.
📊 Dataset… See the full description on the dataset page: https://huggingface.co/datasets/beatsprom/deepseek-r1-autonomous-math-logic-cot-2026.BaxBench
Dataset Summary
BaxBench is a coding benchmark constructed to measure the ability of code generation models and agents to generate correct and secure code. It consists of 392 backend development tasks, which are constructed by combining 28 scenarios that describe the backend functionalities to implement and 14 backend frameworks defining the implementation tools. To assess the correctness and security of the solutions, the benchmark uses end-to-end functional tests and practical… See the full description on the dataset page: https://huggingface.co/datasets/LogicStar/BaxBench.Logics-SWE-Env-2.5K
Logics-SWE-Env-2.5K
2,553 software engineering task instances · 1,771 repositories · 4 programming languages
🤗 Related model: Logics-SWE-Qwen3.6-27B
📄 Paper: One to More, More to One
💻 GitHub: AgenticBigBang
Overview
What is this dataset?
Logics-SWE-Env-2.5K is a collection of repository-level software engineering tasks for research on coding agents and environment-based reinforcement learning. It contains 2,553 unique task instances from 1,771… See the full description on the dataset page: https://huggingface.co/datasets/Logics-MLLM/Logics-SWE-Env-2.5K.task697_mmmlu_answer_generation_formal_logic
Dataset Card for Natural Instructions (https://github.com/allenai/natural-instructions) Task: task697_mmmlu_answer_generation_formal_logic
Additional Information
Citation Information
The following paper introduces the corpus in detail. If you use the corpus in published work, please cite it:
@misc{wang2022supernaturalinstructionsgeneralizationdeclarativeinstructions,
title={Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP… See the full description on the dataset page: https://huggingface.co/datasets/Lots-of-LoRAs/task697_mmmlu_answer_generation_formal_logic.syllogistic-logic
Dataset Summary
Syllogistic-logic is a synthetic dataset designed to evaluate the logical reasoning abilities of LLMs. It focuses on the task of logical premise selection — identifying the minimal set of premises in a knowledge base that entails a given hypothesis. The dataset is built on the syllogistic fragment of first-order logic and supports systematic generalization experiments, including generalization to unseen knowledge bases and reasoning with longer or shorter inference… See the full description on the dataset page: https://huggingface.co/datasets/leobertolazzi/syllogistic-logic.logic-problems-reasoning-dataset
Dataset Card for my-distiset-a26cd729
This dataset has been created with distilabel.
Dataset Summary
This dataset contains a pipeline.yaml which can be used to reproduce the pipeline that generated it in distilabel using the distilabel CLI:
distilabel pipeline run --config "https://huggingface.co/datasets/sdiazlor/my-distiset-a26cd729/raw/main/pipeline.yaml"
or explore the configuration:
distilabel pipeline info --config… See the full description on the dataset page: https://huggingface.co/datasets/sdiazlor/logic-problems-reasoning-dataset.LogicMind-Chat-Reasoning-SFT-300K
Nemotron-Post-Training-Dataset-v2-chat Dataset Card
Overview 📌
This dataset contains 296,168 chat-style instruction/response samples generated by qwen-3-32b. Each record provides a user prompt, an explicit reasoning trace, and a final answer, plus precomputed length fields. The data is packaged as JSONL (one JSON object per line).
Highlights
Scale: 296,168 samples
Category: chat (100%)
Generator: qwen-3-32b (100%)
Structure: problem → qwen3-reasoning → qwen3-solution… See the full description on the dataset page: https://huggingface.co/datasets/Jackrong/LogicMind-Chat-Reasoning-SFT-300K.LogicIFEval
LogicIFEval
For evaluation scripts, please refer to our GitHub repository: https://github.com/mianzhang/LogicIF
The dataset contains two splits:
full: Complete benchmark dataset (3,050 instructions)
mini: Mini version for quick evaluation (749 instructions)
Each line in the JSONL files contains a single evaluation example with the following structure:
{
"task_id": "string", // Unique identifier for the problem
"test_case_id": "int", // Test case number for… See the full description on the dataset page: https://huggingface.co/datasets/billmianz/LogicIFEval.logic-50M
Task Logic: Boolean Circuit Evaluation (The "XOR" Problem)
Overview
Task Logic tests a language model's ability to evaluate boolean circuits with non-linear
gates, especially XOR. This targets the FFN/MLP layers rather than attention, testing
whether the model can perform actual computation rather than just retrieval.
Why This Task Matters
XOR/parity is the classic "hard problem" for neural networks (Minsky & Papert):
Cannot be computed by linear… See the full description on the dataset page: https://huggingface.co/datasets/ericflo/logic-50M.turkish_cyber_security_controls_dataset
Turkish Cyber Security Controls Dataset
Veri Kümesi Özeti
Bu veri kümesi; siber güvenlik kontrolleri, kontrol seçimi ve güvenli mimari tasarımı hakkında hazırlanmış 800 Türkçe kullanıcı-asistan konuşma çifti içerir. Toplam 1.600 mesajdan oluşan koleksiyon, Türkçe siber güvenlik soru-cevap ve instruction-tuning çalışmalarını desteklemek amacıyla hazırlanmıştır.
İçerik geliştirilirken başta NIST SP 800-53 Rev. 5 kontrol kataloğu olmak üzere risk temelli kontrol… See the full description on the dataset page: https://huggingface.co/datasets/logicBombExe/turkish_cyber_security_controls_dataset.designer-design-logics
DESIGNER: Design Logic Library [Project Page]
This repository contains a library of Mermaid-format Design Logics used in the paper DESIGNER: Design-Logic-Guided Multidisciplinary Data Synthesis for LLM Reasoning (ICLR 2026).
Field definitions
mermaid: Design Logic in Mermaid format, abstracted from the source question, which is a human-authored high-difficulty question.
difficulty: difficulty label of the source question
type: type label of the source question… See the full description on the dataset page: https://huggingface.co/datasets/Attention1115/designer-design-logics.ru-instruct-KAN-logic-v1
Russian Instruct KAN-Logic Dataset (v1)
Overview
ru-instruct-KAN-logic-v1 — это специализированный набор данных для instruction tuning (дообучения) языковых моделей на русском языке.
Основной фокус датасета — сложные логические рассуждения (Reasoning), математическое обоснование нейросетевых архитектур нового поколения (KAN - Kolmogorov-Arnold Networks) и теория распределенных вычислений.
Датасет содержит синтетические и курируемые пары instruction - output… See the full description on the dataset page: https://huggingface.co/datasets/K-Net-Labs/ru-instruct-KAN-logic-v1.logic-rl-24k
Logic RL 24K
Logic RL 24K is a 24,461-example English reasoning mixture prepared for reinforcement learning with verifiable rewards (RLVR). It combines procedurally generated, algorithmically verifiable tasks from NVIDIA's Nemotron RL Reasoning Gym release with logic and table-reasoning tasks selected from LLM360's Guru RL 92K collection.
Every record contains one user message, a reference answer, and task-specific metadata that can be used to route the example to the… See the full description on the dataset page: https://huggingface.co/datasets/wflying/logic-rl-24k.guru-logic-verl
GURU Logic VERL Dataset
Dataset Overview
This Hugging Face dataset contains 1,742 samples of logic reasoning problems from the GURU-RL-92k collection, specifically the logic and simulation splits after schema transformation. The data follows VERL (VerL format) specifications for reinforcement learning applications in logic reasoning tasks.
Key Features
Multi-domain Logic Reasoning: Covers ordering puzzles, zebra puzzles, graph problems, and ARC-AGI tasks… See the full description on the dataset page: https://huggingface.co/datasets/sungyub/guru-logic-verl.ru-alpaca-logic
Это переработка в alpaca-friendly формат датасетов от:
MERA-evaluation[MERA]
Из датасета взяты и переработаны только subsets (lcs, parus, rcb, rummu, ruopenbookqa, rutie, ruworldtree)
Vikhrmodels[law_mc]
Датасет переработан с учетом неободимого формата.
Всего в train["input"] - input_ids: 20088 | Наибольшая длинна: 1801 | Количество overflow при ctx(1024): 18
tensor-logic-wikipedia
Tensor Logic Wikipedia Knowledge Base
A structured knowledge base extracted from Wikipedia, designed for hybrid neural-symbolic reasoning.
Dataset Description
This dataset contains:
403,059 facts in Datalog-style format
210,188 entity embeddings (128 dimensions) learned from relationship patterns
Extracted from 37,000+ Wikipedia articles (Vital Articles + random sample)
Files
File
Description
Size
facts_only.tl
Clean facts in Relation(Subject… See the full description on the dataset page: https://huggingface.co/datasets/zekebass/tensor-logic-wikipedia.