Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866913128071036928 |
|---|---|
| author | Wu, Tong Liu, Yang Bai, Jun Jia, Zixia Zhang, Shuyi Lin, Ziyong Wang, Yanting Zhu, Song-Chun Zheng, Zilong |
| author_facet | Wu, Tong Liu, Yang Bai, Jun Jia, Zixia Zhang, Shuyi Lin, Ziyong Wang, Yanting Zhu, Song-Chun Zheng, Zilong |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_07461 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning Wu, Tong Liu, Yang Bai, Jun Jia, Zixia Zhang, Shuyi Lin, Ziyong Wang, Yanting Zhu, Song-Chun Zheng, Zilong Computation and Language 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. |
| title | Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2512.07461 |