AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

IsoCompute Playbook: Optimally Scaling Sampling Compute for LLM RL

Zhoujun Cheng, Yutao Xie, Yuxiao Qu, Amrith Setlur, Shibo Hao, Varad Pimpalkhute, Tongtong Liang, Feng Yao, Zhengzhong Liu, Eric P. Xing, Virginia Smith, Ruslan Salakhutdinov and 3 more

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
increase parallel rollouts per problem n as sampling compute growsGRPO on Qwen2.5-7B-Instruct · Easy and Hard Guru-Math splitsvalidation avg@4 rewardImproved
raise n beyond the saturation point (n=2,048 vs n=512) (ablation)GRPO on Qwen2.5-7B-Instruct · Easy Guru-Math splitvalidation avg@4 rewardNo clear effect
use very large n=512 on the hard set (ablation)GRPO on Qwen2.5-7B-Instruct · Hard Guru-Math splitfinal validation rewardHurt
use larger n on the easy set (ablation)GRPO on Qwen2.5-7B-Instruct · Easy splitworst@4 (sharpening)Improved
use larger n on the hard set (ablation)GRPO on Qwen2.5-7B-Instruct · Hard splitbest@4 (coverage)Improved
use larger n on the extremely hard set (pass@128 = 0) (ablation)GRPO on Qwen2.5-7B-Instruct · Extremely Hard setworst@4 and average rewardHurt
use square-root learning-rate scaling with batch size instead of constant or linear (ablation)GRPO on Qwen2.5-7B-Instruct · Easy set, B=8,192convergence stabilityImproved
add KL and entropy regularization on the easy set (ablation)GRPO on Qwen2.5-7B-Instruct · Easy splittraining stabilityImproved
add KL and entropy regularization on the hard set (ablation)GRPO on Qwen2.5-7B-Instruct · Hard splittraining stabilityHurt
add zero-variance filtering to KL and entropy terms (ablation)GRPO on Qwen2.5-7B-Instruct · Hard splittraining stabilityNo clear effect
decouple baseline estimation: 256 rollouts for advantages but update with 64 (ablation)GRPO on Qwen2.5-7B-Instruct · Bp=128best validation rewardNo clear effect
vary problems per batch Bp at fixed n (ablation)GRPO on Qwen2.5-7B-Instruct · Easy setvalidation rewardNo clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy