datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
MetaMathQAView the project page:
https://meta-math.github.io/
see our paper at https://arxiv.org/abs/2309.12284
Note
All MetaMathQA data are augmented from the training sets of GSM8K and MATH.
None of the augmented data is from the testing set.
You can check the original_question in meta-math/MetaMathQA, each item is from the GSM8K or MATH train set.
Model Details
MetaMath-Mistral-7B is fully fine-tuned on the MetaMathQA datasets and based on the powerful Mistral-7B model. It is… See the full description on the dataset page: https://huggingface.co/datasets/meta-math/MetaMathQA.MetaMathQA-40Karxiv.org/abs/2309.12284
View the project page:
https://meta-math.github.io/
GSM8K_zh
Dataset
GSM8K_zh is a dataset for mathematical reasoning in Chinese, question-answer pairs are translated from GSM8K (https://github.com/openai/grade-school-math/tree/master) by GPT-3.5-Turbo with few-shot prompting.
The dataset consists of 7473 training samples and 1319 testing samples. The former is for supervised fine-tuning, while the latter is for evaluation.
for training samples, question_zh and answer_zh are question and answer keys, respectively;
for testing samples, only… See the full description on the dataset page: https://huggingface.co/datasets/meta-math/GSM8K_zh.qwen3.5-metamathqaMetaMathQA-R1
oumi-ai/MetaMathQA-R1
MetaMathQA-R1 is a text dataset designed to train Conversational Language Models with DeepSeek-R1 level reasoning.
Prompts were augmented from GSM8K and MATH training sets with responses directly from DeepSeek-R1.
MetaMathQA-R1 was used to train MiniMath-R1-1.5B, which achieves 44.4% accuracy on MMLU-Pro-Math, the highest of any model with <=1.5B parameters.
Curated by: Oumi AI using Oumi inference on Parasail
Language(s) (NLP): English
License:… See the full description on the dataset page: https://huggingface.co/datasets/oumi-ai/MetaMathQA-R1.MetaMathQA_GSM8K_zh
Dataset
MetaMathQA_GSM8K_zh is a dataset for mathematical reasoning in Chinese,
question-answer pairs are translated from MetaMathQA (https://huggingface.co/datasets/meta-math/MetaMathQA) by GPT-3.5-Turbo with few-shot prompting.
The dataset consists of 231685 samples.
Citation
If you find the GSM8K_zh dataset useful for your projects/papers, please cite the following paper.
@article{yu2023metamath,
title={MetaMath: Bootstrap Your Own Mathematical Questions for Large… See the full description on the dataset page: https://huggingface.co/datasets/meta-math/MetaMathQA_GSM8K_zh.MetaMathQA
Meta Math Filtered
This is a combined and filtered (removed all the redundant rows) version of meta-math/MetaMathQA and meta-math/MetaMathQA-40K
Usage
from datasets import load_dataset
dataset = load_dataset("Sharathhebbar24/MetaMathQA", split="train")
MetaMathFewshot
A few-shot version of the MetaMath (https://huggingface.co/datasets/meta-math/MetaMathQA) dataset.
Each entry is formatted with 'question' and 'answer' keys. The 'question' key has a random number of query-answer pairs between 0 and 4 inclusive, before a final target query; the expected answer to this is stored in the content of 'answer'.
MetaMathQA-math-500DAPO-Math-17k with the MATH-500 test split converted to the same parquet schema and prompt format.
MetaMath_DPO_FewShot
Dataset Card for "MetaMath_DPO_FewShot"
GSM8K \citep{cobbe2021training} is a dataset of diverse grade school maths word problems, which has been commonly adopted as a measure of the math and reasoning skills of LLMs.
The MetaMath dataset is an extension of the training set of GSM8K using data augmentation.
It is partitioned into queries and responses, where the query is a question involving mathematical calculation or reasoning, and the response is a logical series of steps and… See the full description on the dataset page: https://huggingface.co/datasets/abacusai/MetaMath_DPO_FewShot.metamath-qwen2-math
Dataset Summary
Approximately 900k math problems, where each solution is formatted in a Chain of Thought (CoT) manner. The sources of the dataset range from metamath-qa https://huggingface.co/datasets/meta-math/MetaMathQA and https://huggingface.co/datasets/AI-MO/NuminaMath-CoT with only none-synthetic dataset only. We only use the prompts from metamath-qa and get response with Qwen2-math-72-instruct and rejection-sampling, the solution is filted based on the official evaluation… See the full description on the dataset page: https://huggingface.co/datasets/yingyingzhang/metamath-qwen2-math.GSM8K_Backwardarxiv.org/abs/2309.12284
View the project page:
https://meta-math.github.io/
metamath_gsm8k
🚀 GSM8K training set
The original answer is "\n#### Value" and now is "\n#### Value\nThe answer is: Value", and the answer is cleaned, which is consistent with the answer format with "meta-math/MetaMathQA".
💻 Dataset Usage
Run the following command to load the data:
from datasets import load_dataset
dataset = load_dataset("shuyuej/metamath_gsm8k")
dataset = dataset['train']
print(dataset)
📝 Dataset modification codes
# coding=utf-8
import re
import… See the full description on the dataset page: https://huggingface.co/datasets/shuyuej/metamath_gsm8k.metamathqa-grouped-splitcotmath_metamathMetaMathQAMetaMathQAView the project page:
https://meta-math.github.io/
see our paper at https://arxiv.org/abs/2309.12284
Note
All MetaMathQA data are augmented from the training sets of GSM8K and MATH.
None of the augmented data is from the testing set.
You can check the original_question in meta-math/MetaMathQA, each item is from the GSM8K or MATH train set.
Model Details
MetaMath-Mistral-7B is fully fine-tuned on the MetaMathQA datasets and based on the powerful Mistral-7B model. It is… See the full description on the dataset page: https://huggingface.co/datasets/agicorp/MetaMathQA.Vietnamese-395k-meta-math-MetaMathQA-gg-translatedlm-eval-results-ntnhan-Llama3-8B-MetaMath-private
Dataset Card for Evaluation run of ntnhan/Llama3-8B-MetaMath
Dataset automatically created during the evaluation run of model ntnhan/Llama3-8B-MetaMath
The dataset is composed of 62 configuration(s), each one corresponding to one of the evaluated task.
The dataset has been created from 3 run(s). Each run can be found as a specific split in each configuration, the split being named using the timestamp of the run.The "train" split is always pointing to the latest results.
An… See the full description on the dataset page: https://huggingface.co/datasets/nyu-dice-lab/lm-eval-results-ntnhan-Llama3-8B-MetaMath-private.metamathqa_binarized
Dataset Card for "metamathqa_binarized"
More Information needed
MetaMathQA-decontaminated-openai-native
MetaMathQA — decontaminated, OpenAI-native
MetaMathQA is a widely used math fine-tuning corpus. Its README states:
"None of the augmented data is from the testing set."
That is false, and this release proves it with measurements. 24,334 rows (6.16%) overlap with standard evaluation splits. If you fine-tune on the original and report MATH or GSM8K scores, those scores are inflated.
This release removes the leakage, converts to native messages, and documents every rejection.… See the full description on the dataset page: https://huggingface.co/datasets/Archangel-system/MetaMathQA-decontaminated-openai-native.MetaMathQA-10K-TR
MetaMathQA-10K-TR (Turkish Mathematical Reasoning & CoT Dataset)
MetaMathQA-10K-TR, meta-math/MetaMathQA-40K veri setinden türetilmiş, Türkçe dilinde adım adım akıl yürütme (Chain-of-Thought - CoT) ve matematiksel problem çözme yeteneği kazandırmak amacıyla hazırlanmış 10.000 satırlık yüksek kaliteli bir veri setidir.
Bu veri seti, yerel Qwen 3.8 27B modeli kullanılarak özel olarak tasarlanmış prompt mühendisliği ve sıkı biçimlendirme kuralları ile Türkçe'ye çevrilmiş ve… See the full description on the dataset page: https://huggingface.co/datasets/WrittenWithRust/MetaMathQA-10K-TR.MetaMathQA-GSM8kMetaMath-SAEworked-examples-metamath-v0
Full MetaMathQA worked-examples pack
Source: meta-math/MetaMathQA (all 395k, all types).
Train: 355,688 instances (90% of families)
Holdout: 39,155 (eval/holdout_bare.jsonl)
Docs: arms/<arm>/docs.jsonl.gz (gunzip to use)
Tokens: tokenized/<arm>/shard-00000.npy (dolma2, EOS 100257)
Arm stats
{
"fade_shuffled": {
"n_docs": 1873620,
"n_tokens": 453279629
}
}
MetaMath_DPO_FewShot
Dataset Card for "MetaMath_DPO_FewShot"
GSM8K \citep{cobbe2021training} is a dataset of diverse grade school maths word problems, which has been commonly adopted as a measure of the math and reasoning skills of LLMs.
The MetaMath dataset is an extension of the training set of GSM8K using data augmentation.
It is partitioned into queries and responses, where the query is a question involving mathematical calculation or reasoning, and the response is a logical series of steps and… See the full description on the dataset page: https://huggingface.co/datasets/agicorp/MetaMath_DPO_FewShot.a1_math_metamath_eval_1331
mlfoundations-dev/a1_math_metamath_eval_1331
Precomputed model outputs for evaluation.
Evaluation Results
Summary
Metric
AIME24
AMC23
MATH500
GPQADiamond
MMLUPro
LiveCodeBench
CodeElo
JEEBench
Accuracy
13.7
58.0
74.2
39.6
28.8
9.8
2.5
34.4
AIME24
Average Accuracy: 13.67% ± 1.29%
Number of Runs: 10
Run
Accuracy
Questions Solved
Total Questions
1
13.33%
4
30
2
16.67%
5
30
3
13.33%
4
30
4
13.33%
4
30
5
13.33%
4
30… See the full description on the dataset page: https://huggingface.co/datasets/mlfoundations-dev/a1_math_metamath_eval_1331.metamathqa_formatted
Dataset Card for "metamathqa_formatted"
More Information needed
Metamath2Py
Links
Github with source code: https://github.com/kamushekp/metamath2py
Paper: https://github.com/kamushekp/metamath2py/blob/main/out/main.pdf
Dataset Structure
The Metamath2Py Dataset consists of the following components:
1. JSONL File on Hugging Face
The dataset is provided as a JSONL file, where each line is a JSON object with the following fields:
original_name: The original name of the statement in the Metamath system.
name: The statement name in our… See the full description on the dataset page: https://huggingface.co/datasets/kamushekp/Metamath2Py.MetaMathQA
MetaMathQA Subsets
Curated subsets of meta-math/MetaMathQA for mathematical reasoning experiments.
Subsets
Subset
Samples
Description
full
395,000
All MetaMathQA samples (unchanged)
MATH
155,000
MATH_* types only (AnsAug, Rephrased, FOBAR, SV)
MATH-50K
50,000
Stratified 50K sample from MATH subset
MATH-50K Type Distribution
Type
Count
Proportion
MATH_AnsAug
24,194
48.4%
MATH_Rephrased
16,129
32.3%
MATH_FOBAR
4,839
9.7%… See the full description on the dataset page: https://huggingface.co/datasets/mtybilly/MetaMathQA.