CoRAL 混合框架：扩散模型并行规划，自回归模型执行推理
Beyond the Sequentiality Bottleneck: A Hybrid Diffusion-Autoregressive Framework for Efficient Multimodal Reasoning
扩散模型并行出多条推理路径，再让自回归模型挑最优，速度质量两头占，做推理方向可以看看这篇。
论文提出 CoRAL 框架，用 Asymmetric Planner-Executor（APE）架构解决自回归模型的序列化瓶颈。DLM Planner 并行生成 K 条多样化推理轨迹，由大型 ARM Executor 选出最优方案并合成最终答案。训练上引入 Reward-Guided Denoising for CoT（RDC），把基于结果的奖励融入扩散过程，引导目标导向推理。在视觉语言基准上，CoRAL 明显超过已有扩散方法，并与强 ARM 打出有竞争力的成绩，兼顾效率与质量。
Beyond the Sequentiality Bottleneck: A Hybrid Diffusion-Autoregressive Framework for Efficient Multimodal Reasoning
Autoregressive models (ARMs) face a "Sequentiality Bottleneck" in reasoning, while diffusion language models (DLMs) offer parallelism but lack semantic coherence. To addresses both limitations, we propose Collaborative Reasoning Asymmetric Learning (CoRAL), a hybrid framework via an Asymmetric Planner-Executor (APE) architecture: a DLM Planner generates K diverse reasoning trajectories in parallel, while a large ARM Executor selects the optimal plan and synthesizes the final answer. To train the Planner effectively, we introduce Reward-Guided Denoising for CoT (RDC), which integrates outcome-based rewards into the diffusion process to steer it toward goal-directed reasoning. Experiments on vision-language benchmarks show that CoRAL significantly outperforms prior diffusion-based methods and achieves competitive performance against strong ARMs, achieving both superior efficiency and quality.
