# gsm8k by OpenAI: AI Training Dataset
Source: https://savrn.com/datasets/gsm8k
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Dataset Card

By OpenAI, published under mit, revision 740312add88f.

### Dataset Card for GSM8K

### Table of Contents

- Dataset Description
- Dataset Summary
- Supported Tasks
- Languages
- Dataset Structure
- Data Instances
- Data Fields
- Data Splits
- Dataset Creation
- Curation Rationale
- Source Data
- Annotations
- Personal and Sensitive Information
- Considerations for Using the Data
- Social Impact of Dataset
- Discussion of Biases
- Other Known Limitations
- Additional Information
- Dataset Curators
- Licensing Information
- Citation Information

### Dataset Description

- Homepage: https://openai.com/blog/grade-school-math/
- Repository: https://github.com/openai/grade-school-math
- Paper: https://arxiv.org/abs/2110.14168
- Leaderboard: [Needs More Information]
- Point of Contact: [Needs More Information]

#### Dataset Summary

[Read the full dataset card (817 words)](https://savrn.com/datasets/gsm8k/card)

## Structure

### main 8,792 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 7,473 | 4.0 MB |
| test | 1,319 | 719.3 KB |

questionstringanswerstring

### socratic 8,792 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 7,473 | 5.2 MB |
| test | 1,319 | 943.2 KB |

questionstringanswerstring

## Details

Repository

openai/gsm8k

Publisher

OpenAI

Task category

Text generation

Tags

math-word-problems

Size category

1K<n<10K

Languages

en

Revision

740312add88f781978c0658806c59bc2815b9866

Last updated

2026-03-23

## SAVRN's Notes on gsm8k

GSM8K is tiny, 17,584 rows and 5.9 MB in 7 files, and one of the most reused math datasets we track: 1.2M monthly downloads and a dozen inbound models, from SharperSwarm and CasualSwarms to a full Sheldon RLVR and SFT family built on Qwen2.5-3B-Instruct. There are two configs, main and socratic, each 8,792 rows with 7,473 train and 1,319 test, under an MIT license.

Its real role now is reward signal for RLVR rather than supervised SFT: grading is cheap and the answers are unambiguous. MIT license, no gate, and the revision is recent, last updated 2026-03-23, so diff against any copy you pinned years ago. Contamination is the tax: any base model plausibly saw GSM8K in pretraining, so treat absolute scores as marketing and require held-out math evals for real decisions.

SAVRN Research

## Files

7 files, 5.9 MB in total.

Data4 files · 5.9 MB

Documentation1 file · 7.9 KB

Other1 file · 920 B

Repository1 file · 1.6 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| main/test-00000-of-00001.parquet | Data | 419.1 KB | ee7b8da9e381 |
| main/train-00000-of-00001.parquet | Data | 2.3 MB | ea82612ea958 |
| socratic/test-00000-of-00001.parquet | Data | 487.0 KB | 477dba702820 |
| socratic/train-00000-of-00001.parquet | Data | 2.7 MB | 54eb5fd2105a |
| README.md | Documentation | 7.9 KB | — |
| eval.yaml | Other | 920 B | — |
| .gitattributes | Repository | 1.6 KB | — |

## License and Download

License

mit

Access

No access gate

[Download from OpenAI](https://huggingface.co/datasets/openai/gsm8k)

Released by OpenAI through its official repository on Hugging Face. [Read the license](https://opensource.org/license/mit).

## Models Trained on This Dataset

- Trained on (disclosed)[SharperSwarm](https://savrn.com/models/sharperswarm)
- Trained on (disclosed)[CasualSwarms](https://savrn.com/models/casualswarms)
- Trained on (disclosed)[Al-Khwarizmi-3B-GGUF](https://savrn.com/models/al-khwarizmi-3b-gguf)
- Trained on (disclosed)[shingi-27b](https://savrn.com/models/shingi-27b)
- Trained on (disclosed)[Al-Khwarizmi-3B](https://savrn.com/models/al-khwarizmi-3b)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-RLVR-math-v1](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-rlvr-math-v1)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-SFT-touchup-v4-LoRA](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-sft-touchup-v4-lora)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-SFT-touchup-v4](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-sft-touchup-v4)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-RLAIF-grpo-v2-LoRA](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-rlaif-grpo-v2-lora)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-RLAIF-grpo-v2](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-rlaif-grpo-v2)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-RLVR-math-v1-LoRA](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-rlvr-math-v1-lora)
- Trained on (disclosed)[Qwen2.5-3B-Instruct-Sheldon-RLVR-math-v2-fullft](https://savrn.com/models/qwen2-5-3b-instruct-sheldon-rlvr-math-v2-fullft)
- Trained on (disclosed)[Meet8-0.6B](https://savrn.com/models/meet8-0-6b)
- Trained on (disclosed)[LMLM_97M_1](https://savrn.com/models/lmlm-97m-1)
- Trained on (disclosed)[bobcat-flash-1.2](https://savrn.com/models/bobcat-flash-1-2)
- Trained on (disclosed)[Qwen3.5-0.8B-Japanese-SFT-v2.5](https://savrn.com/models/qwen3-5-0-8b-japanese-sft-v2-5)

## OpenAI

[All models and datasets](https://savrn.com/model-publishers/openai)

## Explore More

- [The dataset directory](https://savrn.com/datasets)
- [The model directory](https://savrn.com/models)
- [Research papers](https://savrn.com/papers)

## Source

- Repository metadata, read 2026-10-19.
- [OpenAI website](https://github.com/openai)
- [Hugging Face record](https://huggingface.co/datasets/openai/gsm8k)
- [How the hub is built](https://savrn.com/model-hub/methodology)