← 研踪 PaperTrailarXiv:2509.25808 · v2
Improving Sampling Efficiency in RLVR through Adaptive Rollout and Response Reuse
首次提交 2025-09-30 13:29
版本更新 2026-11-01 18:47
雷达观测 2026-11-04 23:43
摘要
Group relative policy optimization (GRPO) trains large language models with verifiable rewards by normalizing rewards within response groups. When every response in a group receives the same reward, the advantages vanish and the group provides no policy gradient signal. We propose Adaptive Rollout and Response Reuse Policy Optimization (AR3PO) to obtain informative groups with fewer generations. Adaptive rollout samples responses to each prompt in stages and stops sampling once the model generates a correct response. If all sampled responses are incorrect, response reuse replaces one of them with a past correct response to the same prompt, when available, and recomputes the group advantages. Policy-recomputed reuse trains on the updated groups using probabilities recomputed under the rollout policy; advantage-only reuse uses historical responses only to compute advantages. Across four mathematical benchmarks with 7B and 8B models, AR3PO consistently improves over GRPO and achieves performance comparable to DAPO while requiring up to $4.2\times$ fewer generated responses. Experiments with a 32B model also show comparable performance to DAPO at a similar training stage with substantially fewer generated responses.
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-01 18:47 · 最新观测版本
Improving Sampling Efficiency in RLVR through Adaptive Rollout and Response Reuse
Group relative policy optimization (GRPO) trains large language models with verifiable rewards by normalizing rewards within response groups. When every response in a group receives the same reward, the advantages vanish and the group provides no policy gradient signal. We propose Adaptive Rollout and Response Reuse Policy Optimization (AR3PO) to obtain informative groups with fewer generations. Adaptive rollout samples responses to each prompt in stages and stops sampling once the model generates a correct response. If all sampled responses are incorrect, response reuse replaces one of them with a past correct response to the same prompt, when available, and recomputes the group advantages. Policy-recomputed reuse trains on the updated groups using probabilities recomputed under the rollout policy; advantage-only reuse uses historical responses only to compute advantages. Across four mathematical benchmarks with 7B and 8B models, AR3PO consistently improves over GRPO and achieves performance comparable to DAPO while requiring up to $4.2\times$ fewer generated responses. Experiments with a 32B model also show comparable performance to DAPO at a similar training stage with substantially fewer generated responses.
查看此版本解读 →查看此版本原文 ↗