The grade-school-math repository (sometimes called GSM8K) is a curated dataset of 8,500 high-quality grade school math word problems intended for evaluating mathematical reasoning capabilities of language models. It is structured into 7,500 training problems and 1,000 test problems. These aren’t trivial exercises — many require multi-step reasoning, combining arithmetic operations, and handling intermediate steps (e.g. “If she sold half as many in May… how many in total?”). The problems are written by human authors (not automatically generated) to ensure linguistic variety and realism. The repository maintains strict formatting (e.g. JSONL) for problem + answer pairs, and is used broadly in research to benchmark model performance under “word problem” settings. Issues are tracked (people report incorrect problems, ambiguous statements), and contributions are possible for cleaning or expanding the set.
Features
- Contains 8,500 human-written math word problems with linguistic diversity
- Split into 7,500 training and 1,000 test problems
- Problems often require multi-step reasoning, not just single operations
- Provided in a structured JSONL format (question / answer) for ease of use
- Open to issue correction submissions (errors, ambiguities tracked)
- Widely used as a benchmark for testing arithmetic and reasoning in LLMs