# OpenMathInstruct-2 by NVIDIA: AI Training Dataset
Source: https://savrn.com/datasets/openmathinstruct-2
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Dataset Card

By NVIDIA, published under cc-by-4.0, revision 469216e3f46f.

OpenMathInstruct-2 is a math instruction tuning dataset with 14M problem-solution pairs generated using the [Llama3.1-405B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-405B-Instruct) model.

The training set problems of [GSM8K](https://github.com/openai/grade-school-math) and [MATH](https://github.com/hendrycks/math) are used for constructing the dataset in the following ways: - Solution augmentation: Generating chain-of-thought solutions for training set problems in GSM8K and MATH. - Problem-Solution augmentation: Generating new problems, followed by solutions for these new problems.

OpenMathInstruct-2 dataset contains the following fields:

- problem: Original problem from either the GSM8K or MATH training set or augmented problem from these training sets.
- generated_solution: Synthetically generated solution.
- expected_answer: For problems in the training set, it is the ground-truth answer provided in the datasets. For augmented problems, it is the majority-voting answer.
- problem_source: Whether the problem is taken directly from GSM8K or MATH or is an augmented version derived from either dataset.

[Read the full dataset card (565 words)](https://savrn.com/datasets/openmathinstruct-2/card)

## Structure

### default 21,972,791 rows

| Split | Rows | Size |
| --- | --- | --- |
| train | 13,972,791 | 15.4 GB |
| train_1M | 1,000,000 | 1.3 GB |
| train_2M | 2,000,000 | 2.8 GB |
| train_5M | 5,000,000 | 6.5 GB |

problemstringgenerated_solutionstringexpected_answerstringproblem_sourcestring

## Details

Repository

nvidia/OpenMathInstruct-2

Publisher

NVIDIA

Task category

Question answering

Tags

math, nvidia

Size category

10M<n<100M

Languages

en

Revision

469216e3f46f4dacf476b382e192485ea51a143e

Last updated

2024-11-25

## SAVRN's Notes on OpenMathInstruct-2

OpenMathInstruct-2 is synthetic math instruction data at production scale: 21,972,791 rows, 12.6 GB across 59 files, 146k monthly downloads, generated with Llama 3.1 405B Instruct, task category question answering. Per the card, the problems come from GSM8K and MATH training sets via solution augmentation and problem-solution augmentation, with expected answers taken from ground truth for originals and majority-vote answers for the augmented problems.

CC-BY-4.0 permits commercial use with attribution, and the data is English. Two flags before you commit: because it is built on GSM8K and MATH training problems, do not report scores on either benchmark from models trained here, and the majority-vote labeling on augmented problems means some answer noise is structural, not accidental. Disclosed inbound models on our hub include FinAI and cagliostro-v3.

SAVRN Research

## Files

59 files, 12.6 GB in total.

Data55 files · 12.6 GB

Documentation1 file · 6.4 KB

Other2 files · 441.0 KB

Repository1 file · 3.3 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| data/train-00000-of-00032.parquet | Data | 236.6 MB | 55bae5fa27c8 |
| data/train-00001-of-00032.parquet | Data | 236.9 MB | 4c1a2103a979 |
| data/train-00002-of-00032.parquet | Data | 236.6 MB | 29bd4017239b |
| data/train-00003-of-00032.parquet | Data | 236.8 MB | 619c696a8e43 |
| data/train-00004-of-00032.parquet | Data | 236.7 MB | e8d6d88a1cca |
| data/train-00005-of-00032.parquet | Data | 236.6 MB | 84781d52cab8 |
| data/train-00006-of-00032.parquet | Data | 236.8 MB | 85e196aac429 |
| data/train-00007-of-00032.parquet | Data | 236.8 MB | 279590a0ac58 |
| data/train-00008-of-00032.parquet | Data | 237.0 MB | bc6dbc7a97df |
| data/train-00009-of-00032.parquet | Data | 236.7 MB | fe38efc9a128 |
| data/train-00010-of-00032.parquet | Data | 237.0 MB | 574debcf9111 |
| data/train-00011-of-00032.parquet | Data | 237.0 MB | 750480cc24d2 |
| data/train-00012-of-00032.parquet | Data | 236.5 MB | 338117e09231 |
| data/train-00013-of-00032.parquet | Data | 236.7 MB | 363739df048e |
| data/train-00014-of-00032.parquet | Data | 236.7 MB | 9653648f9e65 |
| data/train-00015-of-00032.parquet | Data | 237.3 MB | 46aa107221b4 |
| data/train-00016-of-00032.parquet | Data | 236.8 MB | fc022c921698 |
| data/train-00017-of-00032.parquet | Data | 236.5 MB | 424a92182ed1 |
| data/train-00018-of-00032.parquet | Data | 236.7 MB | 57f3ce262c96 |
| data/train-00019-of-00032.parquet | Data | 236.5 MB | 9ff9798ed581 |
| data/train-00020-of-00032.parquet | Data | 236.8 MB | a9cbd3619965 |
| data/train-00021-of-00032.parquet | Data | 236.8 MB | 5f644f9d77b1 |
| data/train-00022-of-00032.parquet | Data | 236.8 MB | b546f1f094d8 |
| data/train-00023-of-00032.parquet | Data | 236.7 MB | 4ba1d104be88 |
| data/train-00024-of-00032.parquet | Data | 236.7 MB | dfbce7391b52 |
| data/train-00025-of-00032.parquet | Data | 236.9 MB | 08e7ebce1ecc |
| data/train-00026-of-00032.parquet | Data | 236.7 MB | 3e6545137ea2 |
| data/train-00027-of-00032.parquet | Data | 236.3 MB | 6d46ac4f285f |
| data/train-00028-of-00032.parquet | Data | 236.8 MB | b9378a52c74b |
| data/train-00029-of-00032.parquet | Data | 236.8 MB | 93aee9bae8fe |
| data/train-00030-of-00032.parquet | Data | 236.6 MB | 2191af67a930 |
| data/train-00031-of-00032.parquet | Data | 236.8 MB | b7fc689fb28a |
| data/train_1M-00000-of-00003.parquet | Data | 212.9 MB | 93700f39cdc8 |
| data/train_1M-00001-of-00003.parquet | Data | 213.2 MB | c208b3cda829 |
| data/train_1M-00002-of-00003.parquet | Data | 212.9 MB | 1a42cf7139b6 |
| data/train_2M-00000-of-00006.parquet | Data | 217.1 MB | 0e5070d547da |
| data/train_2M-00001-of-00006.parquet | Data | 217.0 MB | f535f5868188 |
| data/train_2M-00002-of-00006.parquet | Data | 216.9 MB | 3b9e9306db9e |
| data/train_2M-00003-of-00006.parquet | Data | 217.2 MB | 0b758d1e798f |
| data/train_2M-00004-of-00006.parquet | Data | 217.2 MB | 478feb390152 |
| data/train_2M-00005-of-00006.parquet | Data | 217.4 MB | cbec946a2d11 |
| data/train_5M-00000-of-00014.parquet | Data | 222.4 MB | b6c45a7eb504 |
| data/train_5M-00001-of-00014.parquet | Data | 222.1 MB | 73b9d40ac366 |
| data/train_5M-00002-of-00014.parquet | Data | 221.8 MB | 8313fde2b1cf |
| data/train_5M-00003-of-00014.parquet | Data | 221.9 MB | 2a5183be98b8 |
| data/train_5M-00004-of-00014.parquet | Data | 222.1 MB | 1e22f77711e0 |
| data/train_5M-00005-of-00014.parquet | Data | 222.0 MB | 69c9ea15ac79 |
| data/train_5M-00006-of-00014.parquet | Data | 222.3 MB | a293ab952250 |
| data/train_5M-00007-of-00014.parquet | Data | 222.6 MB | 5a4c2b9fe337 |
| data/train_5M-00008-of-00014.parquet | Data | 222.1 MB | 0b2d598714a1 |
| data/train_5M-00009-of-00014.parquet | Data | 222.4 MB | 7d07b4847af4 |
| data/train_5M-00010-of-00014.parquet | Data | 221.9 MB | 34168a7b268c |
| data/train_5M-00011-of-00014.parquet | Data | 222.3 MB | 7dc69d30990e |
| data/train_5M-00012-of-00014.parquet | Data | 222.4 MB | 3518f76d3d95 |
| data/train_5M-00013-of-00014.parquet | Data | 221.8 MB | 5c15c5617b51 |
| README.md | Documentation | 6.4 KB | — |
| SFT Data Diagram 1.jpg | Other | 269.0 KB | d9fa9ade48cb |
| scaling_plot.jpg | Other | 172.0 KB | dfccb18ebb30 |
| .gitattributes | Repository | 3.3 KB | — |

## License and Download

License

cc-by-4.0

Access

No access gate

[Download from NVIDIA](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2)

Released by NVIDIA through its official repository on Hugging Face. [Read the license](https://creativecommons.org/licenses/by/4.0/).

## Models Trained on This Dataset

- Trained on (disclosed)[FinAI](https://savrn.com/models/finai)
- Trained on (disclosed)[cagliostro-v3](https://savrn.com/models/cagliostro-v3-2)

## NVIDIA

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

## 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.
- [NVIDIA website](https://build.nvidia.com/models)
- [Hugging Face record](https://huggingface.co/datasets/nvidia/OpenMathInstruct-2)
- [How the hub is built](https://savrn.com/model-hub/methodology)