"HumanEval is a dataset of 164 handwritten Python programming problems used to test code generation logic."