IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL

TL;DR

This paper investigates how to optimally allocate compute resources for reinforcement learning post-training of large language models, providing practical guidelines to improve efficiency based on empirical findings.

Contribution

It introduces a compute-constrained optimization framework for sampling in LLM RL, revealing predictable scaling behaviors and offering practical resource allocation strategies.

Findings

Optimal parallel rollouts increase with compute and saturate

Increasing rollouts reduces interference across problems

Number of problems per batch affects training stability

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.