SKU RQA-5001 · Sold by External
GSM8K
Product specifications
| SKU | RQA-5001 |
|---|---|
| Data type | Reasoning QA |
| Volume | 8792 math word problems (7473 train / 1319 test, main config) |
| Size on disk | ~4.68 MB (main config uncompressed, 4,676,934 bytes); ~2.73 MB parquet download; ~11.76 MB total HF repo storage across both configs |
| Format | parquet (Hugging Face); JSONL in the source GitHub repo |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open dataset (HF) |
| Quality score | — |
| License | MIT License |
GSM8K (Grade School Math 8K) is a benchmark of 8,792 linguistically diverse grade-school math word problems created by human problem writers and released by OpenAI to study multi-step mathematical reasoning. Each instance has a question string and an answer string; the answer contains a chain of elementary arithmetic reasoning steps with inline calculator annotations of the form <<a op b=c>> and terminates with the final numeric solution on a line prefixed by '####'. Problems take between 2 and 8 steps to solve using only basic arithmetic (+ - x /) and require no concepts beyond early algebra. The dataset is provided on Hugging Face in two configs, 'main' and 'socratic' (the socratic config adds model-generated sub-questions before each step), each split 7,473 train / 1,319 test. It is widely used to evaluate LLM arithmetic and chain-of-thought reasoning, with scoring done by exact-match on the final number extracted after the '####' delimiter.