def get_code_format_reward()

in src/open_r1/rewards.py [0:0]

• 14 lines of code
• 7 McCabe index (conditional complexity)

def get_code_format_reward(language: str = "python"):
    """Format reward function specifically for code responses.

    Args:
        language: Programming language supported by E2B https://e2b.dev/docs/code-interpreting/supported-languages
    """

    def code_format_reward(completions, **kwargs):
        # if there is a language field, use it instead of the default language. This way we can have mixed language training.
        languages = kwargs["language"] if "language" in kwargs else [language] * len(completions)

        completion_contents = [completion[0]["content"] for completion in completions]
        matches = [
            re.match(
                rf"^<think>\n.*?\n</think>\n<answer>\n.*?```{sample_language}.*?```.*?\n</answer>$",
                content,
                re.DOTALL | re.MULTILINE,
            )
            for content, sample_language in zip(completion_contents, languages)
        ]
        return [1.0 if match else 0.0 for match in matches]

    return code_format_reward