Abstract: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.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2512.07461 [cs.CL] |
| (or arXiv:2512.07461v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2512.07461 arXiv-issued DOI via DataCite |
Submission history
From: Tong Wu [view email]
[v1]
Mon, 8 Dec 2025 11:39:43 UTC (1,118 KB)
[v2]
Fri, 19 Dec 2025 02:34:49 UTC (1,117 KB)
[v3]
Thu, 14 May 2026 12:43:45 UTC (1,008 KB)