GSM8K_zh: 8,792 Chinese Math Question-Answer Pairs

GSM8K_zh contains 8,792 Chinese question-answer pairs for mathematical reasoning, translated from the English GSM8K dataset. It is split into 7,473 training samples and 1,319 testing samples for supervised fine-tuning and evaluation. The dataset was created by author ximzdawn using GPT-3.5-Turbo.

Use Cases

• Fine-tune a language model for mathematical reasoning using the 7,473 training samples with question_zh and answer_zh fields.
• Benchmark model performance on the 1,319 test samples containing question_zh to evaluate Chinese math problem-solving ability.
• Analyze the translation quality and reasoning patterns in the answer_zh field across the 8,792 samples.

Strengths

• Provides 8,792 total samples, with a clear split of 7,473 for training and 1,319 for testing.
• Focuses on a specific domain of grade school mathematical reasoning in Chinese.
• Derived from the established GSM8K benchmark, providing a translated counterpart.

Limitations

• The dataset is a machine translation of an existing English dataset, which may introduce translation artifacts or errors.
• Limited to the scope and difficulty of the original grade school math problems, which may not cover advanced topics.
• Sample size of 8,792 is moderate for training large language models without additional data.

Provenance

Source

Translated from the GSM8K (Grade School Math 8K) dataset.

Collection Method

Translated by GPT-3.5-Turbo with few-shot prompting.

Freshness

Last updated on 2026-01-12.