Glossary/GSM8KGSM8KAIGSM8K: A benchmark of 8,500 grade-school maths word problems, the standard early test of step-by-step reasoning.Related termsGuardrail agentGuardrail evaluationGrouped-query attention (GQA)GuardrailsGroup relative policy optimisation (GRPO)Guidance scaleBrowse the full glossary →