paper-with-me

Papers

Unbiased Dynamic Pruning for Efficient Group-Based Policy Optimization

2026-03-04 · Haodong Zhu, Yangyang Ren, Yanjing Li, Mingbao Lin, Linlin Yang, Xuhui Liu, Xiantong Zhen, Haiguang Liu, Baochang Zhang arxiv

Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs due to its extensive group-based sampling requirement. While recent selective data utilization methods can mitigate this overhead, they could induce estimation bias by altering the underlying sampling distribution, compromising theoretical rigor and convergence behavior. To address this limitation, we propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation through importance sampling-based correction. By incorporating mathematically derived rescaling factors, DPPO significantly accelerates GRPO training without altering the optimization objective of the full-batch baseline. Furthermore, to mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy that maximizes valid token density and hardware utilization. Extensive experiments demonstrate that DPPO consistently accelerates training across diverse models and benchmarks. For instance, on Qwen3-4B trained on MATH, DPPO achieves 2.37$\times$ training speedup and outperforms GRPO by 3.36% in average accuracy across six mathematical reasoning benchmarks.

📄 PDF Abstract BibTeX arXiv:2603.04135

Code (0)

등록된 구현이 없습니다.

Tasks

Mathematical Reasoning

Similar Papers 제목 키워드 기반

GRIP-VLM: Group-Relative Importance Pruning for Efficient Vision-Language Models

2026-05-13 · Mingzhe Huang, Weijun Wang, Xin Ding, Liang Mi 외 arxiv

In Vision-Language Models (VLMs), processing a massive number of visual tokens incurs prohibitive computational overhead. While recent training-aware pruning methods attempt to selectively discard redundant tokens, they …

Reinforcement Learning

GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification

2026-04-15 · Wangjie Gan, Miao Pan, Linbo Xi, Wenqi Zhang 외 arxiv

Large language models are typically post-trained using supervised fine-tuning (SFT) and reinforcement learning (RL), yet effectively unifying efficient knowledge injection with robust generalization remains challenging. …

Reinforcement Learning

CPPO: Accelerating the Training of Group Relative Policy Optimization-Based Reasoning Models

2025-03-28 · Zhihang Lin, Mingbao Lin, Yuan Xie, Rongrong Ji

This paper introduces Completion Pruning Policy Optimization (CPPO) to accelerate the training of reasoning models based on Group Relative Policy Optimization (GRPO). GRPO, while effective, incurs high training costs due…

GPUGSM8KMath

Hindsight-Anchored Policy Optimization: Turning Failure into Feedback in Sparse Reward Settings

2026-03-11 · Yuning Wu, Ke Wang, Devin Chen, Kai Wei arxiv

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a promising paradigm for post-training reasoning models. However, group-based methods such as Group Relative Policy Optimization (GRPO) face a critical…

Reinforcement Learning

Noise-corrected GRPO: From Noisy Rewards to Unbiased Gradients

2025-10-21 · Omar El Mansouri, Fathinah Asma Izzati, Mohamed El Amine Seddik, Salem Lahlou arxiv

Reinforcement learning from human feedback (RLHF) or verifiable rewards (RLVR), the standard paradigm for aligning LLMs or building recent SOTA reasoning models, is highly sensitive to noise from inconsistent or erroneou…

Reinforcement Learning