# Prune as You Generate: Online Rollout Pruning for Faster and Better RLVR

arXiv: https://arxiv.org/abs/2603.24840 · HTML: https://mlsanity.com/paper/2603.24840/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Placed in Reinforcement learning for reasoning, one of 17 directions in LLM reasoning.
- Builds on 51 of its 95 references that are in the map.
- First posted on arXiv 25 Mar 2026.

## Placed in

- LLM reasoning / [Reinforcement learning for reasoning](https://mlsanity.com/lr/reinforcement-learning-for-reasoning/)

## Builds on (most cited first, 30 of 51)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers](https://mlsanity.com/paper/2210.17323/) (2022)
- [Knowledge Distillation: A Survey](https://mlsanity.com/paper/2006.05525/) (2020)
- [Solving Quantitative Reasoning Problems with Language Models](https://mlsanity.com/paper/2206.14858/) (2022)
- [Understanding R1-Zero-Like Training: A Critical Perspective](https://mlsanity.com/paper/2503.20783/) (2025)
- [OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems](https://mlsanity.com/paper/2402.14008/) (2024)
- [Scalable Best-of-N Selection for Large Language Models via Self-Certainty](https://mlsanity.com/paper/2502.18581/) (2025)
- [DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs](https://mlsanity.com/paper/2406.01721/) (2024)
- [Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts](https://mlsanity.com/paper/2506.02177/) (2025)
- [Online Difficulty Filtering for Reasoning Oriented Reinforcement Learning](https://mlsanity.com/paper/2504.03380/) (2025)
- [CPPO: Accelerating the Training of Group Relative Policy Optimization-Based Reasoning Models](https://mlsanity.com/paper/2503.22342/) (2025)
- [S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models](https://mlsanity.com/paper/2505.07686/) (2025)
- [Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning](https://mlsanity.com/paper/2504.13818/) (2025)
- [Discrete-state Continuous-time Diffusion for Graph Generation](https://mlsanity.com/paper/2405.11416/) (2024)
- [SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning](https://mlsanity.com/paper/2501.13198/) (2025)
- [Mem-Gallery: Benchmarking Multimodal Long-Term Conversational Memory for MLLM Agents](https://mlsanity.com/paper/2601.03515/) (2026)
- [How Efficient is LLM-Generated Code? A Rigorous & High-Standard Benchmark](https://mlsanity.com/paper/2406.06647/) (2024)
- [History Rhymes: Accelerating LLM Reinforcement Learning with RhymeRL](https://mlsanity.com/paper/2508.18588/) (2025)
- [MoPE-CLIP: Structured Pruning for Efficient Vision-Language Models with Module-wise Pruning Error Metric](https://mlsanity.com/paper/2403.07839/) (2024)
- [QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models](https://mlsanity.com/paper/2602.20309/) (2026)
- [Ensuring User-side Fairness in Dynamic Recommender Systems](https://mlsanity.com/paper/2308.15651/) (2023)
- [Hierarchical Multi-Marginal Optimal Transport for Network Alignment](https://mlsanity.com/paper/2310.04470/) (2023)
- [Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs](https://mlsanity.com/paper/2508.14896/) (2025)
- [Logic Query of Thoughts: Guiding Large Language Models to Answer Complex Logic Queries with Knowledge Graphs](https://mlsanity.com/paper/2404.04264/) (2024)
- [Pave Your Own Path: Graph Gradual Domain Adaptation on Fused Gromov-Wasserstein Geodesics](https://mlsanity.com/paper/2505.12709/) (2025)

## Built on by (most cited first, 10 of 10)

- [TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning](https://mlsanity.com/paper/2606.11119/) (2026)
- [DuQuant++: Fine-grained Rotation Enhances Microscaling FP4 Quantization](https://mlsanity.com/paper/2604.17789/) (2026)
- [DUET: Optimize Token-Budget Allocation for Reinforcement Learning with Verifiable Rewards](https://mlsanity.com/paper/2605.08441/) (2026)
- [Predict, Don't Iterate: Efficient Adaptive-Length Infilling for Diffusion Language Models](https://mlsanity.com/paper/2609.02108/) (2026)
- [IV-CoT: Implicit Visual Chain-of-Thought for Structure-Aware Text-to-Image Generation](https://mlsanity.com/paper/2606.24849/) (2026)
- [Benchmarking Trustworthiness of SLMs: Pre-trained vs. Compressed](https://mlsanity.com/paper/2608.11981/) (2026)
- [Extending Confidence-Based Text2Cypher with Grammar and Schema Aware Filtering](https://mlsanity.com/paper/2605.10318/) (2026)
- [Efficient RLVR Scheduling via Graph-Structured Online Difficulty Estimation](https://mlsanity.com/paper/2608.17941/) (2026)
- [Improving the Diversity of LLM Outputs without a Trade-off](https://mlsanity.com/paper/2609.33038/) (2026)
- [A Step Harder, Not a Step Too Far: The Benefits and Limits of Curriculum Difficulty Progression in GRPO](https://mlsanity.com/paper/2610.13669/) (2026)

## Abstract (arXiv)

Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, methods such as GRPO and DAPO suffer from substantial computational cost, since they rely on sampling many rollouts for each prompt. Moreover, in RLVR the relative advantage is often sparse: many samples become nearly all-correct or all-incorrect, yielding low within-group reward variance and thus weak learning signals. In this paper, we introduce arrol (Accelerating RLVR via online Rollout Pruning), an online rollout pruning method that prunes rollouts during generation while explicitly steering the surviving ones more correctness-balanced to enhance learning signals. Specifically, arrol trains a lightweight quality head on-the-fly to predict the success probability of partial rollouts and uses it to make early pruning decisions. The learned quality head can further weigh candidates to improve inference accuracy during test-time scaling. To improve efficiency, we present a system design that prunes rollouts inside the inference engine and re-batches the remaining ones for log-probability computation and policy updates. Across GRPO and DAPO on Qwen-3 and LLaMA-3.2 models (1B-8B), arrol improves average accuracy by +2.30 to +2.99 while achieving up to 1.7x training speedup, and yielding up to +8.33 additional gains in average accuracy in test-time scaling. The code is available at https://github.com/Hsu1023/ARRoL.