Easy extension of diverse RL algorithms for dLLMs Easy extension of extra benchmark evaluations for dLLMs Easy integration of popular and upcoming dLLM infras and HuggingFace weights DARE is a work in ...