Our dataset is gathered by using a new representation language to annotate over the AQuA-RAT dataset. AQuA-RAT has provided the questions, options, rationale, and the correct options.

• Question: A train running at the speed of 48 km / hr crosses a pole in 9 seconds . what is the length of the train ? • Rationale: Speed = ( 48 x 5 / 18 ) m / sec = ( 40 / 3 ) m / sec . length of the train = ( speed x time ) . length of the train = ( 40 / 3 x 9 ) m = 120 m . answer is c . • Options: a ) 140 , b ) 130 , c ) 120 , d ) 170 , e ) 160 • Correct Option is: C

The rationales are noisy, incomplete and sometimes incorrect. We correct these rationales and provide stepwise solutions for a portion of AQuA-RAT.

• Our Annotated Formula: multiply(divide(multiply(48, const_1000), const_3600), 9)

Dataset Card for AQuA-Rat

Homepage: https://github.com/google-deepmind/AQuA This is an unofficial curation of the AQuA-Rat dataset, uploaded here with minimal (i.e., no content-modifying) processing.

Paper: Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems (ACL Anthology)

Modifications:

Pre-tokenized splits removed since tokenization built into most LM pipelines. Fixed file suffix from .json to .jsonl. Changed options column to… See the full description on the dataset page: https://huggingface.co/datasets/mathewhe/aqua_rat.

laurentiubp/aqua-rat dataset hosted on Hugging Face and contributed by the HF Datasets community

A large-scale dataset consisting of approximately 100,000 algebraic word problems. The solution to each question is explained step-by-step using natural language. This data is used to train a program generation model that learns to generate the explanation, while generating the program that solves the question.

Dataset

This dataset was created by Việt Hưng Nguyễn

Released under Apache 2.0

Contents

Dataset

This dataset was created by Darien Schettler

Released under Apache 2.0

Contents

quzhe/aqua-rat dataset hosted on Hugging Face and contributed by the HF Datasets community

Now that I have your attention, please up-vote this dataset and read the following!!!

AQUA-RAT (Algebra Question Answering with Rationales) Dataset

This dataset contains the algebraic word problems with rationales described in our paper:

Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. (2017) Program Induction by Rationale Generation: Learning to Solve and Explain Algebraic Word Problems. In Proc. ACL. https://arxiv.org/pdf/1705.04146.pdf

The dataset consists of about 100,000 algebraic word problems with natural language rationales. Each problem is a json object consisting of four parts:

question - A natural language definition of the problem to solve options - 5 possible options (A, B, C, D and E), among which one is correct rationale - A natural language description of the solution to the problem correct - The correct option

Here is an example of a problem object:

{ "question": "A grocery sells a bag of ice for $1.25, and makes 20% profit. If it sells 500 bags of ice, how much total profit does it make?", "options": ["A)125", "B)150", "C)225", "D)250", "E)275"], "rationale": "Profit per bag = 1.25 * 0.20 = 0.25 Total profit = 500 * 0.25 = 125 Answer is A.", "correct": "A" }

Files

train.json -> untokenized training set train.tok.json -> tokenized training set dev.json -> untokenized development set dev.tok.json -> tokenized development set test.json -> untokenized test set test.tok.json -> tokenized test set

Note

This dataset has been fully crowdsourced, as described using the technique in the paper (Ling et al., 2017). The initial published results included in the paper were derived from a previous version of this dataset that cannot be released in full, and results using the published system will differ. Results using our published system will be forthcoming.

Source

https://github.com/deepmind/AQuA

https://media.giphy.com/media/YknAouVrcbkiDvWUOR/giphy.gif" alt="Alt Text"> https://media.giphy.com/media/26xBtSyoi5hUUkCEo/giphy.gif" alt="Alt Text"> https://media.giphy.com/media/4LiMmbAcvgTQs/giphy.gif" alt="Alt Text"> https://media.giphy.com/media/3o6Ztg5jGKDQSjaZ1K/giphy.gif" alt="Alt Text">

DeepMind AQUA Rat Converted Dataset

This dataset is a refined version of the original DeepMind AQUA-RAT benchmark. Originally designed as a multiple-choice question-answering dataset, AQUA-RAT has been transformed in this version to require a numerical answer in many cases.

  Overview

Conversion Approach:
Approximately two-thirds of the dataset now requires a single numerical answer.
For questions that are harder to convert to a single verifiable answer, the original… See the full description on the dataset page: https://huggingface.co/datasets/ideacode/aquarat.

MathQA (Math Problems)

Learning to solve math problems

Source

Huggingface Hub: link

About this dataset

We introduce a large-scale dataset of math word problems. Our dataset is gathered by using a new representation language to annotate over the AQuA-RAT dataset with fully-specified operational programs. AQuA-RAT has provided the questions, options, rationale, and the correct options.

Research Ideas

• A math word problem solving model can be trained on this dataset in order to better understand how to solve math word problems.
• This dataset can be used to develop new methods for automatically annotating math word problems with fully-specified operational programs.
• This dataset can be used as a benchmark for evaluating the performance of various methods for solving math word problems

Acknowledgements

License

License: CC0 1.0 Universal (CC0 1.0) - Public Domain Dedication No Copyright - You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission. See Other Information.

Columns

File: validation.csv | Column name | Description | |:----------------------|:----------------------------------------------------------| | Problem | The math word problem. (String) | | Rationale | The rationale for the math word problem. (String) | | options | The options for the math word problem. (List of strings) | | correct | The correct option for the math word problem. (String) | | annotated_formula | The annotated formula for the math word problem. (String) | | linear_formula | The linear formula for the math word problem. (String) | | category | The category for the math word problem. (String) |

File: train.csv | Column name | Description | |:----------------------|:----------------------------------------------------------| | Problem | The math word problem. (String) | | Rationale | The rationale for the math word problem. (String) | | options | The options for the math word problem. (List of strings) | | correct | The correct option for the math word problem. (String) | | annotated_formula | The annotated formula for the math word problem. (String) | | linear_formula | The linear formula for the math word problem. (String) | | category | The category for the math word problem. (String) |

File: test.csv | Column name | Description | |:----------------------|:----------------------------------------------------------| | Problem | The math word problem. (String) | | Rationale | The rationale for the math word problem. (String) | | options | The options for the math word problem. (List of strings) | | correct | The correct option for the math word problem. (String) | | annotated_formula | The annotated formula for the math word problem. (String) | | linear_formula | The linear formula for the math word problem. (String) | | category | The category for the math word problem. (String) |

Dataset from https://huggingface.co/datasets/deepmind/aqua_rat

add new column "encoded_label" with mapping:

Label Encoding Mapping: {'A': 0, 'B': 1, 'C': 2, 'D': 3, 'E': 4}

Our dataset is gathered by using a new representation language to annotate over the AQuA-RAT dataset. AQuA-RAT has provided the questions, options, rationale, and the correct options.

laurentiubp/aquarat-scored dataset hosted on Hugging Face and contributed by the HF Datasets community

本数据集是通过使用一种新的表示语言对 AQuA-RAT 数据集进行注释来收集的。 AQuA-RAT 提供了问题、选项、理由和正确选项。

LiSoViMa/AquaRat dataset hosted on Hugging Face and contributed by the HF Datasets community

MrMaxMind99/reformatted-aquarat dataset hosted on Hugging Face and contributed by the HF Datasets community

Model Name

Params

MMLU-Pro-Plus Baseline Drop MMLU-Pro Baseline Drop Added Exp MMLU Pro Plus Added MMLU-redux 2.0 Baseline Drop AQUA-RAT Baseline Drop

CohereLabs/c4ai-command-a-03-2025 111B ✅ (single inference) ✅ done ✅ (HF naive batch) ✅ done ✅ done

✅

-

-

-

google/gemma-3-12b-it 12B ✅ (HF naive batch) ✅ done ✅ (HF naive batch) ✅ done ✅ done

✅

-

-

-

meta-llama/Llama-4-Scout-17B-16E 17B ✅ (HF naive batch) ✅ done ✅ (HF naive batch) ✅ done ✅ done

✅

-

-

-

Qwen/Qwen3-4B 4B… See the full description on the dataset page: https://huggingface.co/datasets/sleeping-ai/Judgement-baseline.

EmilRyd/aquarat-sft-gt-stylized-modified dataset hosted on Hugging Face and contributed by the HF Datasets community