def accuracy_reward()

in src/open-r1-multimodal/src/open_r1/grpo_rec.py [0:0]

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

def accuracy_reward(completions, solution, **kwargs):
    pattern = r'<answer>(.*?)</answer>'
    contents = [completion[0]["content"] for completion in completions]
    if dist.get_rank() == 0:
        print("\n******************************************************response***************************************************\n", contents[0], "\n******************************************************response***************************************************\n".replace("*", "-"))
    rewards = []
    for content, sol in zip(contents, solution):
        reward = 0.0
        content_answer_match = re.search(pattern, content, re.DOTALL)
        if content_answer_match:
            content_answer = content_answer_match.group(1).strip()
            reward = (bleu(content_answer, sol) + rouge(content_answer, sol)) / 2
        rewards.append(reward)

    return rewards