Datasets:
Qwen
/

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
pandas
PolyMath / README.md
psyangqi's picture
update readme
5cbffba verified
metadata
configs:
  - config_name: ar
    data_files:
      - split: top
        path: ar/top.parquet
      - split: high
        path: ar/high.parquet
      - split: medium
        path: ar/medium.parquet
      - split: low
        path: ar/low.parquet
  - config_name: bn
    data_files:
      - split: top
        path: bn/top.parquet
      - split: high
        path: bn/high.parquet
      - split: medium
        path: bn/medium.parquet
      - split: low
        path: bn/low.parquet
  - config_name: de
    data_files:
      - split: top
        path: de/top.parquet
      - split: high
        path: de/high.parquet
      - split: medium
        path: de/medium.parquet
      - split: low
        path: de/low.parquet
  - config_name: en
    data_files:
      - split: top
        path: en/top.parquet
      - split: high
        path: en/high.parquet
      - split: medium
        path: en/medium.parquet
      - split: low
        path: en/low.parquet
  - config_name: es
    data_files:
      - split: top
        path: es/top.parquet
      - split: high
        path: es/high.parquet
      - split: medium
        path: es/medium.parquet
      - split: low
        path: es/low.parquet
  - config_name: fr
    data_files:
      - split: top
        path: fr/top.parquet
      - split: high
        path: fr/high.parquet
      - split: medium
        path: fr/medium.parquet
      - split: low
        path: fr/low.parquet
  - config_name: id
    data_files:
      - split: top
        path: id/top.parquet
      - split: high
        path: id/high.parquet
      - split: medium
        path: id/medium.parquet
      - split: low
        path: id/low.parquet
  - config_name: it
    data_files:
      - split: top
        path: it/top.parquet
      - split: high
        path: it/high.parquet
      - split: medium
        path: it/medium.parquet
      - split: low
        path: it/low.parquet
  - config_name: ja
    data_files:
      - split: top
        path: ja/top.parquet
      - split: high
        path: ja/high.parquet
      - split: medium
        path: ja/medium.parquet
      - split: low
        path: ja/low.parquet
  - config_name: ko
    data_files:
      - split: top
        path: ko/top.parquet
      - split: high
        path: ko/high.parquet
      - split: medium
        path: ko/medium.parquet
      - split: low
        path: ko/low.parquet
  - config_name: ms
    data_files:
      - split: top
        path: ms/top.parquet
      - split: high
        path: ms/high.parquet
      - split: medium
        path: ms/medium.parquet
      - split: low
        path: ms/low.parquet
  - config_name: pt
    data_files:
      - split: top
        path: pt/top.parquet
      - split: high
        path: pt/high.parquet
      - split: medium
        path: pt/medium.parquet
      - split: low
        path: pt/low.parquet
  - config_name: ru
    data_files:
      - split: top
        path: ru/top.parquet
      - split: high
        path: ru/high.parquet
      - split: medium
        path: ru/medium.parquet
      - split: low
        path: ru/low.parquet
  - config_name: sw
    data_files:
      - split: top
        path: sw/top.parquet
      - split: high
        path: sw/high.parquet
      - split: medium
        path: sw/medium.parquet
      - split: low
        path: sw/low.parquet
  - config_name: te
    data_files:
      - split: top
        path: te/top.parquet
      - split: high
        path: te/high.parquet
      - split: medium
        path: te/medium.parquet
      - split: low
        path: te/low.parquet
  - config_name: th
    data_files:
      - split: top
        path: th/top.parquet
      - split: high
        path: th/high.parquet
      - split: medium
        path: th/medium.parquet
      - split: low
        path: th/low.parquet
  - config_name: vi
    data_files:
      - split: top
        path: vi/top.parquet
      - split: high
        path: vi/high.parquet
      - split: medium
        path: vi/medium.parquet
      - split: low
        path: vi/low.parquet
  - config_name: zh
    data_files:
      - split: top
        path: zh/top.parquet
      - split: high
        path: zh/high.parquet
      - split: medium
        path: zh/medium.parquet
      - split: low
        path: zh/low.parquet

PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts

PolyMath is a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs.

  • ๐Ÿ“ˆ Broad Difficulty Range: PolyMath defines and partitions mathematical difficulty across four levels using two core dimensions: Thought Depth and Knowledge Breadth, ranging from K-12 to Olympiad and advanced frontier mathematics, with 125 problems per language at each level.
logo
  • ๐ŸŒ Language Diversity: Each problem in PolyMath is available in 18 parallel language versions, encompassing over 75% of the worldโ€™s native speakers and major language families, ensuring diversity across both high-resource and low-resource languages.
logo
  • ๐Ÿง‘โ€๐Ÿซ High-Quality Annotation: Each problem translation is calibrated by language experts, avoiding direct use of LLM-generated outputs and ensuring precise term and logical clarity.
logo

๐Ÿ“Š Main Results

The leaderboard is continuously updated! See https://qwen-polymath.github.io/#leaderboard


๐Ÿ“„ Citation

If you use PolyMath in your research, please cite us:

@misc{wang2025polymath,
      title={PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts}, 
      author={Yiming Wang and Pei Zhang and Jialong Tang and Haoran Wei and Baosong Yang and Rui Wang and Chenshu Sun and Feitong Sun and Jiran Zhang and Junxuan Wu and Qiqian Cang and Yichang Zhang and Fei Huang and Junyang Lin and Fei Huang and Jingren Zhou},
      year={2025},
      eprint={2504.18428},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2504.18428}, 
}