[moyix]

At least they used HumanEval+, which adds a bunch more test cases and fixes some errors in the original benchmark!