From GitHub it looks like the code was generated with GPT-3.5. I wonder how GPT-4 does with the same prompts?