IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL

• URL: http://arxiv.org/abs/2603.12151v1
• Date: Thu, 12 Mar 2026 16:49:21 GMT
• Title: IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL
• Abstract summary: 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.
• Score: 75.28528082188787
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: While scaling laws guide compute allocation for LLM pre-training, analogous prescriptions for reinforcement learning (RL) post-training of large language models (LLMs) remain poorly understood. We study the compute-optimal allocation of sampling compute for on-policy RL methods in LLMs, framing scaling as a compute-constrained optimization over three resources: parallel rollouts per problem, number of problems per batch, and number of update steps. We find that the compute-optimal number of parallel rollouts per problem increases predictably with compute budget and then saturates. This trend holds across both easy and hard problems, though driven by different mechanisms: solution sharpening on easy problems and coverage expansion on hard problems. We further show that increasing the number of parallel rollouts mitigates interference across problems, while the number of problems per batch primarily affects training stability and can be chosen within a broad range. Validated across base models and data distributions, our results recast RL scaling laws as prescriptive allocation rules and provide practical guidance for compute-efficient LLM RL post-training.

Related papers

• 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)
• CoScale-RL: Efficient Post-Training by Co-Scaling Data and Computation [8.290384911182615]
  Training Large Reasoning Model (LRM) is usually unstable and unpredictable.<n>We propose CoScale-RL, a novel scaling strategy with better data and computational efficiency.
  arXiv  Detail & Related papers  (2026-01-21T06:17:52Z)
• CurES: From Gradient Analysis to Efficient Curriculum Learning for Reasoning LLMs [53.749193998004166]
  Curriculum learning plays a crucial role in enhancing the training efficiency of large language models.<n>We propose CurES, an efficient training method that accelerates convergence and employs Bayesian posterior estimation to minimize computational overhead.
  arXiv  Detail & Related papers  (2025-10-01T15:41:27Z)
• How to Train Your LLM Web Agent: A Statistical Diagnosis [96.86317871461834]
  We present the first statistically grounded study on compute allocation for LLM web-agent post-training.<n>Our approach uses a two-stage pipeline, training a Llama 3.1 8B student to imitate a Llama 3.3 70B teacher via supervised fine-tuning (SFT) and on-policy reinforcement learning.<n>Our results show that combining SFT with on-policy RL consistently outperforms either approach alone on both WorkArena and MiniWob++.
  arXiv  Detail & Related papers  (2025-07-05T17:12:33Z)
• Improving Data Efficiency for LLM Reinforcement Fine-tuning Through Difficulty-targeted Online Data Selection and Rollout Replay [69.67914133280296]
  Reinforcement learning (RL) has become an effective approach for fine-tuning large language models (LLMs)<n>We propose two techniques to improve data efficiency in LLM RL fine-tuning: difficulty-targeted online data selection and rollout replay.<n>Our method reduces RL fine-tuning time by 23% to 62% while reaching the same level of performance as the original GRPO algorithm.
  arXiv  Detail & Related papers  (2025-06-05T17:55:43Z)
• Enhancing Efficiency and Exploration in Reinforcement Learning for LLMs [16.64613090265212]
  Reasoning large language models (LLMs) excel in complex tasks.<n>Existing approaches allocate an equal number of rollouts to all questions during reinforcement learning (RL)<n>We propose a mechanism for dynamically allocating rollout budgets based on the difficulty of the problems.
  arXiv  Detail & Related papers  (2025-05-24T07:28:29Z)
• Cost-Optimal Grouped-Query Attention for Long-Context Modeling [45.981681856747365]
  Grouped-Query Attention (GQA) is a widely adopted strategy for reducing the computational cost of attention layers in large language models.<n>We analyze the relationship among context length, model size, GQA configuration, and model loss.<n>We propose a recipe for deriving cost-optimal GQA configurations.
  arXiv  Detail & Related papers  (2025-03-12T17:50:42Z)
• Multi-Bin Batching for Increasing LLM Inference Throughput [19.652542432683234]
  Large language models (LL) grow in popularity improving the efficiency of their systems.<n> requests is a critical step in scheduling jobs on servers.<n> requests often have varying generation lengths, causing resource underutilization.<n>We formalize this problem from a queueing-theoretic perspective, and aim to design a throughput control policy.
  arXiv  Detail & Related papers  (2024-12-03T03:16:12Z)
• Online Learning of Weakly Coupled MDP Policies for Load Balancing and Auto Scaling [42.6574685545681]
  This paper introduces a novel model and algorithms for tuning load balancers coupled with auto scalers, considering bursty traffic arriving at finite queues. We begin by presenting the problem as a weakly coupled Markov Decision Processes (MDP), solvable via a linear program (LP) We extend it to tackle the problem of online parameter learning and policy optimization using a two-timescale algorithm based on the LP Lagrangian.
  arXiv  Detail & Related papers  (2024-06-20T09:34:24Z)
• LaGR-SEQ: Language-Guided Reinforcement Learning with Sample-Efficient Querying [71.86163159193327]
  Large language models (LLMs) have recently demonstrated their impressive ability to provide context-aware responses via text. This ability could potentially be used to predict plausible solutions in sequential decision making tasks pertaining to pattern completion. We introduce LaGR, which uses this predictive ability of LLMs to propose solutions to tasks that have been partially completed by a primary reinforcement learning (RL) agent.
  arXiv  Detail & Related papers  (2023-08-21T02:07:35Z)
• Deep Policy Iteration with Integer Programming for Inventory Management [8.27175065641495]
  We present a framework for optimizing long-term discounted reward problems with large accessible action space and state dependent constraints.<n>Our proposed Programmable Actor Reinforcement Learning (PARL) uses a deep-policy method that leverages neural networks (NNs) to approximate the value function.<n>We benchmark the proposed algorithm against state-of-the-art RL algorithms and commonly used replenishments and find it considerably outperforms existing methods by as much as 14.7% on average.
  arXiv  Detail & Related papers  (2021-12-04T01:40:34Z)

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.