datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
math_shepherd
Math-Shepherd Dataset
Summary
The Math-Shepherd dataset is a processed version of Math-Shepherd dataset, designed to train models using the TRL library for stepwise supervision tasks. It provides step-by-step solutions to mathematical problems, enabling models to learn and verify each step of a solution, thereby enhancing their reasoning capabilities.
Data Structure
Format: Standard
Type: Stepwise supervision
Columns:
"pompt": The problem statement.… See the full description on the dataset page: https://huggingface.co/datasets/trl-lib/math_shepherd.MathSpatialMathSpatial
Do MLLMs Really Understand Space? A Mathematical Spatial Reasoning Evaluation
Submitted to ACM Multimedia 2026 — Dataset Track
Overview •
Key Findings •
Statistics •
Getting Started •
Annotations •
Leaderboard
Overview
MathSpatial is a large-scale, open dataset ecosystem dedicated to mathematical spatial reasoning in Multimodal Large Language Models (MLLMs). It provides 10,000 problems with 26,000+ geometric diagrams, covering… See the full description on the dataset page: https://huggingface.co/datasets/shuolucs/MathSpatial.MathSpatialMathSpatial
Do MLLMs Really Understand Space? A Mathematical Spatial Reasoning Evaluation
Submitted to ACM Multimedia 2026 — Dataset Track
Overview •
Key Findings •
Statistics •
Getting Started •
Annotations •
Leaderboard
Overview
MathSpatial is a large-scale, open dataset ecosystem dedicated to mathematical spatial reasoning in Multimodal Large Language Models (MLLMs). It provides 10,000 problems with 26,000+ geometric diagrams… See the full description on the dataset page: https://huggingface.co/datasets/AnnieLKY/MathSpatial.math-sft-10B
Dataset: math-sft-10B
This dataset was uploaded from /mnt/yulan_pretrain/mount/data_final_train/math-sft-10B/no-curriculum/tmp.
Math-Shepherd
Dataset Card for Math-Shepherd
Project Page: Math-Shepherd
Paper: https://arxiv.org/pdf/2312.08935.pdf
Data Loading
from datasets import load_dataset
dataset = load_dataset("peiyi9979/Math-Shepherd")
Data Instance
Every instance consists of three data fields: "input," "label," and "task".
"input": problem + step-by-step solution, e.g.,
If Buzz bought a pizza with 78 slices at a restaurant and then decided to share it with the waiter in the ratio of 5:8, with… See the full description on the dataset page: https://huggingface.co/datasets/peiyi9979/Math-Shepherd.Maths-CollegeMaths-College
I am releasing a large Mathematics dataset in the instrution format.
This extensive dataset, comprising nearly one million instructions in JSON format, encapsulates a wide array of mathematical disciplines essential for a profound understanding of the subject.
This dataset is very useful to Researchers & Model developers.
Following Fields & sub Fields are covered:
Probability
Statistics
Liner Algebra
Algebra
Group Theory
Topology
Abstract Algebra
Graph Theory
Combinatorics… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/Maths-College.math_stage2_hard_1Math-Step-DPO-10K
Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs
🖥️Code | 🤗Data | 📄Paper
This repo contains the Math-Step-DPO-10K dataset for our paper Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs, Step-DPO is a simple, effective, and data-efficient method for boosting the mathematical reasoning ability of LLMs. Notably, Step-DPO, when applied to Qwen2-72B-Instruct, achieves scores of 70.8% and 94.0% on the test sets of MATH and GSM8K… See the full description on the dataset page: https://huggingface.co/datasets/xinlai/Math-Step-DPO-10K.math_stratos_scale_judged_and_annotated_with_difficultyMaths_competition_questionsMaths-Grade-SchoolMaths-Grade-School
I am releasing large Grade School level Mathematics datatset.
This extensive dataset, comprising nearly one million instructions in JSON format, encapsulates a diverse array of topics fundamental to building a strong mathematical foundation.
This dataset is in instruction format so that model developers, researchers etc. can easily use this dataset.
Following Fields & sub Fields are covered:
Calculus
Probability
Algebra
Liner Algebra
Trigonometry
Differential Equations… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/Maths-Grade-School.math_syntmath-story-problems
Math Story Problems Dataset
Dataset Description
This dataset contains mathematical word problems presented in multiple formats, from direct equations to complex story-based scenarios. It is designed for training and evaluating language models on mathematical reasoning tasks.
Dataset Structure
The dataset is split into three parts:
Train: 131,072 samples
Validation: 1,024 samples
Test: 3,072 samples
Features
{
"eq_qs": "string", # Equation… See the full description on the dataset page: https://huggingface.co/datasets/azminetoushikwasi/math-story-problems.MATH_SFTmath_stackexchange_qa
Math StackExchange Curated (Parquet, CC BY-SA 4.0)
This dataset is a curated collection of Math StackExchange (MSE) Q&A pairs packaged in Parquet format.Each sample contains a problem (title, question_body), its corresponding answer (answer_body), the original MSE tag string (tags), and a flag indicating whether the answer was accepted (accepted).
This dataset includes content derived from the Math StackExchange public data dump (CC BY-SA 4.0, © Stack Exchange Inc.).This derived… See the full description on the dataset page: https://huggingface.co/datasets/glopezas/math_stackexchange_qa.KlearReasoner-MathSub-30K
Dataset Summary
This dataset is a subset of the Klear-Reasoner Math RL dataset.The full dataset contains approximately 88K entries, while this release includes a 30K-entry subset.
The subset was obtained by filtering the outputs of DeepSeek-R1-0120. For each prompt, DeepSeek-R1-0120 generated 16 responses and we retained only the parts where the majority voting results matched the standard answers, in order to filter out the cases that the rule-based validator math_verify cannot… See the full description on the dataset page: https://huggingface.co/datasets/Kwai-Klear/KlearReasoner-MathSub-30K.arXiv-metadata-oai-snapshot-111math-source-forced-recovery-v2Maths-Grade-SchoolMaths-Grade-School
I am releasing large Grade School level Mathematics datatset.
This extensive dataset, comprising nearly one million instructions in JSON format, encapsulates a diverse array of topics fundamental to building a strong mathematical foundation.
This dataset is in instruction format so that model developers, researchers etc. can easily use this dataset.
Following Fields & sub Fields are covered:
Calculus
Probability
Algebra
Liner Algebra
Trigonometry
Differential Equations… See the full description on the dataset page: https://huggingface.co/datasets/pt-sk/Maths-Grade-School.Simple-MathSteps-90K
Introducing Simple-MathSteps-90K:
An open source dataset of 93,325 elementary math problems with step-by-step solutions and multiple choice answers. Designed to enhance mathematical reasoning in models ranging from 1B to 13B parameters.
Key Features
93,325 Math Problems: Generated by paraphrasing the AQuA-RAT dataset using Qwen3 4B Instruct 2507, with a focus on consistency and quality.
Detailed Step-by-Step Solutions: Clear reasoning that breaks down problems… See the full description on the dataset page: https://huggingface.co/datasets/Raymond-dev-546730/Simple-MathSteps-90K.Math-Shapes
Math-Symbols Dataset
Overview
The Math-Symbols dataset is a collection of images representing various mathematical symbols. This dataset is designed for machine learning applications, particularly in the fields of image recognition, optical character recognition (OCR), and symbol classification.
Dataset Details
Name: Math-Symbols
Type: Image dataset
Format: Images with corresponding labels
Size: 131MB (downloaded dataset files), 118MB (auto-connected Parquet… See the full description on the dataset page: https://huggingface.co/datasets/prithivMLmods/Math-Shapes.maths-vision-task-splitssat-mathsmath-sft-solutions-no-cot
Math SFT Solutions No CoT
A cleaned mathematics supervised fine-tuning dataset containing:
instruction → solution pairs
mathematical proofs
derivations
olympiad-style solutions
theorem reasoning
stepwise mathematical explanations
detailed final solutions
This dataset was built specifically for mathematical supervised fine-tuning (SFT).
Unlike many reasoning datasets, this release removes explicit chain-of-thought tags and hidden thinking traces while preserving high-quality… See the full description on the dataset page: https://huggingface.co/datasets/kaushik-harsh-99/math-sft-solutions-no-cot.MATH_SHEPHERD_DPO_FORMATmath_scalingmath-sft-solutions-no-cot-v3
Math SFT Solutions No CoT V3
Math SFT Solutions No CoT V3 is a large-scale mathematics supervised fine-tuning (SFT) dataset designed for instruction tuning and mathematical capability adaptation.
Version 3 substantially expands mathematical coverage while improving dataset quality through stronger filtering, cleaning, and supervision refinement.
Unlike reasoning-heavy datasets, this release focuses on clean instruction → response pairs without hidden chain-of-thought style… See the full description on the dataset page: https://huggingface.co/datasets/kaushik-harsh-99/math-sft-solutions-no-cot-v3.Maths
Mathematics Dataset
This dataset code generates mathematical question and answer pairs, from a range
of question types at roughly school-level difficulty. This is designed to test
the mathematical learning and algebraic reasoning skills of learning models.
Original paper: Analysing Mathematical
Reasoning Abilities of Neural Models
(Saxton, Grefenstette, Hill, Kohli).
Example questions
Question: Solve -42*r + 27*c = -1167 and 130*r + 4*c = 372 for r.
Answer: 4… See the full description on the dataset page: https://huggingface.co/datasets/slifeisenjoy/Maths.math-sft-solutions-no-cot-v4
Math SFT Solutions No CoT V4
Math SFT Solutions No CoT V4 is a large-scale mathematics supervised fine-tuning (SFT) dataset designed for instruction tuning and mathematical capability adaptation.
Version 4 expands dataset scale while improving supervision quality through stronger cleaning, deduplication, formatting refinement, and broader mathematical coverage.
Unlike reasoning-oriented datasets, this release focuses on direct instruction → response supervision and removes… See the full description on the dataset page: https://huggingface.co/datasets/kaushik-harsh-99/math-sft-solutions-no-cot-v4.math-speech-dataset