GSM8K
Grade-school mathematics word problems requiring multi-step reasoning.
Обзор
GSM8K is a set of grade-school mathematics word problems whose answers require several reasoning steps rather than a single lookup. It solves the problem that most question-answering benchmarks can be passed by recall, which says nothing about whether a model can work through a problem. It exists because multi-step arithmetic turned out to be a sharp discriminator, and it is the benchmark that chain-of-thought prompting was introduced against.
Граф технологий
Как GSM8K связана с остальным стеком.