MBPP

Mostly Basic Programming Problems

974 道面向入门程序员的众包短篇 Python 编程任务,每题附自然语言描述、参考解答和三个测试用例;另有 427 题经人工核验的"sanitized"子集。通过对测试执行模型程序评分(pass@1 或解出比例)。与 HumanEval 并列为两个经典早期代码合成基准之一;现已饱和且大量受污染。

88.6%
Llama 3.1 405B Instruct
厂商自报
2021-08-16 → 2024-07-23: 59.6% → 88.6%
发布
2021-08
维护者
Google Research (Austin et al.)
状态
saturated
污染风险
high
指标
pass@1 (percent, ↑)
题量
974
领域
code
人类
无实测基线

备注. Reported numbers mix the full set, the 500-problem test split, the sanitized subset and EvalPlus MBPP+, so compare only like with like.

完整账本

系统开发者分数日期来源条件
Llama 3.1 405B InstructMeta88.6%厂商自报pass_k: 1 shots: 0
Llama 3.1 model card row 'MBPP ++ base version' (EvalPlus MBPP+ base tests, 378 problems): 88.6 for 405B, 86.0 for 70B. Not the original 500-problem test split. Date is the Llama 3.1 release.
137B LM (few-shot)Google Research59.6%论文
Paper abstract: the largest 137B model synthesizes solutions to 59.6% of MBPP problems with few-shot prompting; fine-tuning adds about 10 points.