ellamind/gsm8k-platinum-multilingual

--- viewer: true configs: - config_name: ces data_files: - split: test path: ces/test-* - config_name: dan data_files: - split: test path: dan/test-* - config_name: deu data_files: - split: test path: deu/test-* - config_name: fin data_files: - split: test path: fin/test-* - config_name: fra data_files: - split: test path: fra/test-* - config_name: ita data_files: - split: test path: ita/test-* - config_name: nld data_files: - split: test path: nld/test-* - config_name: pol data_files: - split: test path: pol/test-* - config_name: spa data_files: - split: test path: spa/test-* language: - ces - dan - deu - fin - fra - ita - nld - pol - spa license: cc-by-4.0 task_categories: - question-answering - text-generation size_categories: - 1K<n<10K dataset_info: - config_name: fra features: - name: id dtype: large_string - name: benchmark dtype: large_string - name: cleaning_status dtype: large_string - name: question dtype: large_string - name: solution dtype: large_string - name: final_answer dtype: large_string - name: flag_for_review dtype: bool - name: review_reason dtype: large_string - name: target_language dtype: large_string - name: translation_confidence dtype: large_string - name: translation_notes dtype: large_string - name: required_adaptation dtype: bool - name: adaptation_notes dtype: large_string - name: preserves_formatting dtype: bool splits: - name: test num_bytes: 1093846 num_examples: 1209 download_size: 511346 dataset_size: 1093846 - config_name: ita features: - name: id dtype: large_string - name: benchmark dtype: large_string - name: cleaning_status dtype: large_string - name: question dtype: large_string - name: solution dtype: large_string - name: final_answer dtype: large_string - name: flag_for_review dtype: bool - name: review_reason dtype: large_string - name: target_language dtype: large_string - name: translation_confidence dtype: large_string - name: translation_notes dtype: large_string - name: required_adaptation dtype: bool - name: adaptation_notes dtype: large_string - name: preserves_formatting dtype: bool splits: - name: test num_bytes: 1064384 num_examples: 1209 download_size: 505220 dataset_size: 1064384 - config_name: spa features: - name: id dtype: large_string - name: benchmark dtype: large_string - name: cleaning_status dtype: large_string - name: question dtype: large_string - name: solution dtype: large_string - name: final_answer dtype: large_string - name: flag_for_review dtype: bool - name: review_reason dtype: large_string - name: target_language dtype: large_string - name: translation_confidence dtype: large_string - name: translation_notes dtype: large_string - name: required_adaptation dtype: bool - name: adaptation_notes dtype: large_string - name: preserves_formatting dtype: bool splits: - name: test num_bytes: 1057584 num_examples: 1209 download_size: 489701 dataset_size: 1057584 --- # GSM8K Platinum Multilingual Multilingual translations of [GSM8K Platinum](https://huggingface.co/datasets/madrylab/gsm8k-platinum), a rigorously cleaned and verified version of GSM8K containing 1,209 elementary math word problems requiring multi-step arithmetic reasoning. **Source:** [madrylab/gsm8k-platinum](https://huggingface.co/datasets/madrylab/gsm8k-platinum) (test split, 1,209 questions) ## Languages | Config | Language | Examples | |--------|----------|----------| | `ces` | Czech | 100 | | `dan` | Danish | 100 | | `deu` | German | 1,209 | | `fin` | Finnish | 100 | | `fra` | French | 100 | | `ita` | Italian | 100 | | `nld` | Dutch | 100 | | `pol` | Polish | 100 | | `spa` | Spanish | 100 | More to be added later. Need a certain language? Feel free to open a discussion in the community tab. ## Translation approach Each question is translated from English using Gemini 3.1 Pro (reasoning_effort=high) with structured outputs. The translation prompt is tailored for math word problems: - **Natural-language text** (story context, reasoning explanations) is translated fluently into the target language. - **All numeric values, arithmetic expressions, and the logical structure** of each solution step are preserved exactly. - **Names** are localized to culturally natural equivalents (e.g., "Janet" → "Jana" in German). - **Currency** symbols/words are translated naturally but amounts are never changed. - **Units** are kept as-is (no conversion) if the answer depends on exact values. - The translator self-reports **confidence** and notes any adaptations made. ## Fields | Field | Description | |---|---| | **question** | Translated math word problem | | **solution** | Translated step-by-step solution (numbers preserved) | | **final_answer** | The final numeric answer (same across all languages) | | **cleaning_status** | How the example was verified: `consensus`, `verified`, or `revised` | | **flag_for_review** | Whether the translator flagged quality issues | | **review_reason** | Explanation if flagged | | **target_language** | Target language name | | **translation_confidence** | direct / confident / uncertain / problematic | | **translation_notes** | Notes on translation choices | | **required_adaptation** | Whether localization was needed | | **adaptation_notes** | Details of any adaptations | | **preserves_formatting** | Whether original formatting was preserved | ## Usage ```python from datasets import load_dataset # Load a specific language ds = load_dataset("ellamind/gsm8k-platinum-multilingual", "deu", split="test") ``` ## Citation This dataset is a translation of the original GSM8K Platinum benchmark. If you use it, please cite both the original paper and this dataset: ```bibtex @article{vendrow2025large, title={Do large language model benchmarks test reliability?}, author={Vendrow, Joshua and Vendrow, Edward and Beery, Sara and Madry, Aleksander}, journal={arXiv preprint arXiv:2502.03461}, year={2025} } @misc{gsm8k_platinum_multilingual_2026, title={{GSM8K} Platinum Multilingual: Translated Elementary Math Word Problems}, author={ellamind}, year={2026}, publisher={Hugging Face}, howpublished={\url{https://huggingface.co/datasets/ellamind/gsm8k-platinum-multilingual}} } ```