QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training--Inference Mismatch

• URL: http://arxiv.org/abs/2604.07853v1
• Date: Thu, 09 Apr 2026 06:11:46 GMT
• Title: QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training--Inference Mismatch
• Abstract summary: We propose QaRL (Rollout Alignment Quantization-Aware RL), which aligns training-side forward with the quantized rollout to minimize mismatch.<n>On Qwen3-30B-A3B MoE for math problems, QaRL outperforms quantized-rollout training by +5.5 while improving stability and preserving low-bit throughput benefits.
• Score: 33.82215097004296
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language model (LLM) reinforcement learning (RL) pipelines are often bottlenecked by rollout generation, making end-to-end training slow. Recent work mitigates this by running rollouts with quantization to accelerate decoding, which is the most expensive stage of the RL loop. However, these setups destabilize optimization by amplifying the training-inference gap: rollouts are operated at low precision, while learning updates are computed at full precision. To address this challenge, we propose QaRL (Rollout Alignment Quantization-Aware RL), which aligns training-side forward with the quantized rollout to minimize mismatch. We further identify a failure mode in quantized rollouts: long-form responses tend to produce repetitive, garbled tokens (error tokens). To mitigate these problems, we introduce TBPO (Trust-Band Policy Optimization), a sequence-level objective with dual clipping for negative samples, aimed at keeping updates within the trust region. On Qwen3-30B-A3B MoE for math problems, QaRL outperforms quantized-rollout training by +5.5 while improving stability and preserving low-bit throughput benefits.

Related papers

• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Scheduling [54.276306194000405]
  We propose SortedRL, an online length-aware scheduling strategy for scaling reinforcement learning.<n>SortedRL reorders rollout samples based on output lengths, prioritizing short samples forming groups for early updates.<n>Experiments using LLaMA-3.1-8B and Qwen-2.5-32B on diverse tasks, including logical puzzles, show that SortedRL reduces RL training bubble ratios by over 50%.
  arXiv  Detail & Related papers  (2026-03-24T16:48:31Z)
• IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL [75.28528082188787]
  We study the compute-optimal allocation of sampling compute for on-policy reinforcement learning methods in large language models (LLMs)<n>We find that the compute-optimal number of parallel rollouts per problem increases predictably with compute budget and then saturates.<n>Our results recast RL scaling laws as prescriptive allocation rules and provide practical guidance for compute-efficient LLM RL post-training.
  arXiv  Detail & Related papers  (2026-03-12T16:49:21Z)
• QuRL: Efficient Reinforcement Learning with Quantized Rollout [23.326106976928898]
  Reinforcement learning with verifiable rewards (RLVR) has become a trending paradigm for training reasoning large language models (LLMs)<n>Due to the autoregressive decoding nature of LLMs, the rollout process becomes the efficiency bottleneck of RL training, consisting of up to 70% of the total training time.<n>We propose Quantized Reinforcement Learning (QuRL) that uses a quantized actor for accelerating the rollout.
  arXiv  Detail & Related papers  (2026-02-15T01:48:10Z)
• Just-In-Time Reinforcement Learning: Continual Learning in LLM Agents Without Gradient Updates [53.3717573880076]
  We introduce Just-In-Time Reinforcement Learning (JitRL), a training-free framework that enables test-time policy optimization without any gradient updates.<n>JitRL maintains a dynamic, non-parametric memory of experiences and retrieves relevant trajectories to estimate action advantages on-the-fly.<n>Experiments on WebArena and Jericho demonstrate that JitRL establishes a new state-of-the-art among training-free methods.
  arXiv  Detail & Related papers  (2026-01-26T14:16:51Z)
• Beat the long tail: Distribution-Aware Speculative Decoding for RL Training [75.75462952580796]
  We propose a Distribution Aware Speculative decoding framework that accelerates RL rollouts without altering model outputs.<n>Experiments on math and code reasoning tasks show that DAS reduces rollout time up to 50% while preserving identical training curves.
  arXiv  Detail & Related papers  (2025-11-17T19:02:12Z)
• CoPRIS: Efficient and Stable Reinforcement Learning via Concurrency-Controlled Partial Rollout with Importance Sampling [11.252930904797]
  We propose Concurrency- Controlled Partial Rollout with Importance Sampling (CoPRIS)<n>CoPRIS mitigates long-tail inefficiencies by maintaining a fixed number of concurrent rollouts, early-terminating once sufficient samples are collected, and reusing unfinished trajectories in subsequent rollouts.<n>Experiments show that CoPRIS achieves up to 1.94x faster training while maintaining comparable or superior performance to synchronous RL systems.
  arXiv  Detail & Related papers  (2025-11-05T11:39:32Z)
• BroRL: Scaling Reinforcement Learning via Broadened Exploration [88.69554867685243]
  Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a key ingredient for unlocking complex reasoning capabilities in large language models.<n>Recent work ProRL has shown promise in scaling RL by increasing the number of training steps.<n>We investigate a complementary paradigm for scaling RL, BroR-Lincreasing the number of rollouts per example to hundreds.
  arXiv  Detail & Related papers  (2025-10-01T17:59:02Z)
• SPEQ: Offline Stabilization Phases for Efficient Q-Learning in High Update-To-Data Ratio Reinforcement Learning [51.10866035483686]
  High update-to-data (UTD) ratio algorithms in reinforcement learning (RL) improve sample efficiency but incur high computational costs, limiting real-world scalability.<n>We propose Offline Stabilization Phases for Efficient Q-Learning (SPEQ), an RL algorithm that combines low-UTD online training with periodic offline stabilization phases.<n>During these phases, Q-functions are fine-tuned with high UTD ratios on a fixed replay buffer, reducing redundant updates on suboptimal data.
  arXiv  Detail & Related papers  (2025-01-15T09:04:19Z)
• Continuous Control with Coarse-to-fine Reinforcement Learning [15.585706638252441]
  We present a framework that trains RL agents to zoom-into a continuous action space in a coarse-to-fine manner. We introduce a concrete, value-based algorithm within the framework called Coarse-to-fine Q-Network (CQN) CQN robustly learns to solve real-world manipulation tasks within a few minutes of online training.
  arXiv  Detail & Related papers  (2024-07-10T16:04:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.