让大模型原生并行推理，无需外部监督。
Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning
- 通过自蒸馏训练让模型从串行转为原生并行思维
- 在8个基准上最高提升24.5%性能，推理速度加快4.6倍
- 适合追求高效智能体推理的研究者与开发者
我们提出无教师框架 Native Parallel Reasoner (NPR)，使大语言模型（LLMs）自演化出真正的并行推理能力。NPR通过三项关键创新实现：1）自蒸馏渐进训练范式，无需外部监督即可从初始格式发现过渡到严格拓扑约束；2）新型并行感知策略优化（PAPO）算法，在执行图内直接优化分支策略，支持模型通过试错学习自适应分解；3）稳健的 NPR 引擎重构 SGLang 的内存管理和流程控制，保障大规模并行强化学习训练的稳定性。在8个推理基准上，基于 Qwen3-4B 训练的 NPR 实现最高24.5%性能提升，推理速度最快达4.6倍加速。相比以往基线常退回到自回归解码，NPR实现100%真实并行执行，确立了自我演化、高效且可扩展智能体推理的新标准。
原文摘要 · Abstract (English)
We introduce Native Parallel Reasoner (NPR), a teacher-free framework that enables Large Language Models (LLMs) to self-evolve genuine parallel reasoning capabilities. NPR transforms the model from sequential emulation to native parallel cognition through three key innovations: 1) a self-distilled progressive training paradigm that transitions from ``cold-start'' format discovery to strict topological constraints without external supervision; 2) a novel Parallel-Aware Policy Optimization (PAPO) algorithm that optimizes branching policies directly within the execution graph, allowing the model to learn adaptive decomposition via trial and error; and 3) a robust NPR Engine that refactors memory management and flow control of SGLang to enable stable, large-scale parallel RL training. Across eight reasoning benchmarks, NPR trained on Qwen3-4B achieves performance gains of up to 24.5% and inference speedups up to 4.6x. Unlike prior baselines that often fall back to autoregressive decoding, NPR demonstrates 100% genuine parallel execution, establishing a new standard for self-evolving, efficient, and scalable agentic reasoning.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。