GSM8K Distilled Zh: Chinese Math Word Problems for Supervised Learning

7,473 training and 1,319 testing samples form this Chinese mathematical reasoning dataset. The question-answer pairs were translated from the original English GSM8K dataset using GPT-3.5-Turbo with few-shot prompting techniques. The dataset was processed using MetaMath and is hosted by PaddlePaddle.

Use Cases

• Fine-tuning language models for Chinese mathematical problem-solving based on the described question-answer pairs.
• Benchmarking model performance on grade-school level math word problems in Chinese.
• Training supervised models for step-by-step reasoning tasks using the provided training samples.
• Studying the effectiveness of distillation and translation techniques for educational datasets.

Strengths

• Contains 7,473 training samples, providing a substantial base for model training.
• Includes a separate test set of 1,319 samples for evaluation.
• Leverages a known source dataset (GSM8K) and describes a specific translation methodology using GPT-3.5-Turbo.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's specific license and terms of use are unknown.
• Row count for the full dataset is unknown, which may limit suitability assessment.

Provenance

Source

Original GSM8K dataset from OpenAI.

Collection Method

Translated from English using GPT-3.5-Turbo with few-shot prompting and processed with MetaMath.

Freshness

Last updated 2025-03-17 02:57:35; freshness should be verified.