Parason: Revealing Subtask and Trial Parallelism in LLM Reasoning

📄 arXiv: 2608.24658v1 📥 PDF

作者: Zhengyang Zhang, Zijian Zhang, Jiaxuan Gao, Shusheng Xu, Yi Wu, Song Han, Ligeng Zhu

分类: cs.AI

发布日期: 2026-08-25

💡 一句话要点

提出Parason以解决LLM推理中的并行性问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大规模语言模型 推理加速 并行计算 试验并行性 子任务并行性 数学推理 模型训练

📋 核心要点

  1. 现有的自回归解码方法在复杂任务中执行推理时存在严重的延迟，导致响应时间过长。
  2. Parason通过揭示子任务并行性和试验并行性，采用结构化的并行轨迹和PA-GRPO进行模型训练，提升推理效率。
  3. 在数学推理基准测试中，Parason实现了约1.7倍的加速，且准确性保持在竞争水平。

📝 摘要（中文）

在大规模语言模型（LLM）的推理过程中，标准的自回归解码方式导致长推理过程的顺序执行，严重影响了复杂任务的响应时间。本文提出Parason，揭示并学习了LLM推理中的子任务并行性和试验并行性。通过分析，发现试验并行性占据了可并行推理计算的65.5%，并在困难问题中愈发显著。Parason将顺序推理轨迹转换为结构化的并行轨迹，并采用并行感知的群体相对策略优化（PA-GRPO）进行训练，最终在数学推理基准上实现了约1.7倍的加速，同时保持了竞争力的准确性。

🔬 方法详解

问题定义：本文旨在解决大规模语言模型在推理过程中因顺序执行而导致的高延迟问题。现有方法主要关注子任务并行性，忽略了试验并行性这一重要形式。

核心思路：Parason通过识别并学习子任务并行性和试验并行性，将顺序推理转化为结构化的并行轨迹，从而提高推理效率。

技术框架：Parason的整体架构包括两个主要模块：首先是推理轨迹的结构化转换，其次是基于PA-GRPO的模型训练。推理时，通过工具调用执行学习到的并行结构。

关键创新：Parason的核心创新在于同时揭示并利用子任务并行性和试验并行性，显著提高了推理的并行化程度，与传统方法相比，能够更有效地处理复杂任务。

关键设计：在训练过程中，采用了平衡准确性、延迟和并行性比例的奖励机制，确保模型在推理时能够有效利用并行结构。

🖼️ 关键图片

fig_0
img_1
img_2

📊 实验亮点

在数学推理基准AIME24和AIME25的实验中，Parason实现了约1.7倍的推理加速，相较于传统方法，显著提高了推理效率，同时保持了准确性，展示了其在复杂任务中的优势。

🎯 应用场景

Parason的研究成果在多个领域具有广泛的应用潜力，尤其是在需要快速推理的复杂任务中，如数学推理、科学计算和实时决策支持系统。其高效的并行推理能力将推动大规模语言模型在实际应用中的落地，提升用户体验和系统响应速度。

📄 摘要（原文）

Scaling test-time reasoning has substantially improved the problem-solving ability of large language models (LLMs), but standard autoregressive decoding still executes long reasoning traces sequentially, creating severe latency for difficult tasks (up to days and weeks). Parallel reasoning offers a natural remedy. However, prior systems primarily focus on Subtask Parallelism, where the model learns to decompose a high-level task into smaller chunks that can be solved independently. This approach overlooks another pervasive form of parallelism: Trial Parallelism, where multiple speculative attempts explore, verify, and aggregate competing hypotheses in parallel. In this paper, we introduce Parason, which reveals and learns both forms of parallelism in LLM reasoning. Our analysis identifies Trial Parallelism as the majority of parallelizable reasoning computation (65.5% in DeepSeek-V4's reasoning steps in HLE), and it becomes increasingly dominant on hard problems. Guided by this taxonomy, Parason converts sequential reasoning traces into structured parallel trajectories with a context-free grammar, then trains models with Parallelism-Aware Group Relative Policy Optimization (PA-GRPO), whose reward jointly balances accuracy, latency, and the two parallelism ratios. At inference time, Parason executes the learned parallel structure through tool calls, translating theoretical savings to real-world wall-clock acceleration. Experiments on mathematical reasoning benchmarks including AIME24 and AIME25 show that Parason achieves an average acceleration about 1.7$\times$ while maintaining competitive accuracy.