跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.00782 · v2

One Offset, Two Directions: Baseline Adjustment via Subgroup Offsets

首次提交 2026-10-31 17:54
版本更新 2026-11-13 17:17
雷达观测 2026-11-16 11:42
arXiv ↗PDF ↗

摘要

Group Relative Policy Optimization (GRPO) estimates advantages against the within-prompt mean reward, a baseline shared across the group. In practice, rollouts within a prompt are not always homogeneous, differing in their solution strategies and reasoning trajectories, along which they may group into subgroups. We further find that these heterogeneous subgroups may exhibit distinct reward distributions. The global baseline, however, collapses them onto a single credit axis, where the credit reflects neither a rollout's standing within its subgroup nor the subgroup's standing relative to the rest. We therefore study how to adjust the baseline when the rollouts of a prompt fall into heterogeneous subgroups. We define the \emph{subgroup offset} as the deviation of a subgroup mean from the global mean, and shift the baseline along it by a single signed coefficient λλ. The sign of λλ yields two opposing directions: \emph{Alignment} (λ>0λ{>}0) adds the offset to the baseline, drawing credit toward a rollout's relative position within its subgroup; \emph{Contrast} (λ<0λ{<}0) subtracts it, drawing a rollout's credit toward the relative quality of the subgroup it belongs to. The adjustment acts on the baseline alone, leaving the reward signal and the optimization objective unchanged. In this work, we induce subgroups from strategies, grouping rollouts that share a strategy into the same subgroup. We evaluate both directions on Qwen3-8B-Base and Olmo3-7B-Base across four competition-level math benchmarks, against a λ=0λ{=}0 anchor that shares the same strategy-conditioned sampling and differs only in the baseline. Against this anchor each direction yields modest but consistent gains in multi-sample coverage, with Contrast the stronger of the two.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v2,暂无更早版本可对照。

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 · 2026-11-13 17:17 · 最新观测版本

One Offset, Two Directions: Baseline Adjustment via Subgroup Offsets

Group Relative Policy Optimization (GRPO) estimates advantages against the within-prompt mean reward, a baseline shared across the group. In practice, rollouts within a prompt are not always homogeneous, differing in their solution strategies and reasoning trajectories, along which they may group into subgroups. We further find that these heterogeneous subgroups may exhibit distinct reward distributions. The global baseline, however, collapses them onto a single credit axis, where the credit reflects neither a rollout's standing within its subgroup nor the subgroup's standing relative to the rest. We therefore study how to adjust the baseline when the rollouts of a prompt fall into heterogeneous subgroups. We define the \emph{subgroup offset} as the deviation of a subgroup mean from the global mean, and shift the baseline along it by a single signed coefficient λλ. The sign of λλ yields two opposing directions: \emph{Alignment} (λ>0λ{>}0) adds the offset to the baseline, drawing credit toward a rollout's relative position within its subgroup; \emph{Contrast} (λ<0λ{<}0) subtracts it, drawing a rollout's credit toward the relative quality of the subgroup it belongs to. The adjustment acts on the baseline alone, leaving the reward signal and the optimization objective unchanged. In this work, we induce subgroups from strategies, grouping rollouts that share a strategy into the same subgroup. We evaluate both directions on Qwen3-8B-Base and Olmo3-7B-Base across four competition-level math benchmarks, against a λ=0λ{=}0 anchor that shares the same strategy-conditioned sampling and differs only in the baseline. Against this anchor each direction yields modest but consistent gains in multi-sample coverage, with Contrast the stronger of the two.

查看此版本解读 →查看此版本原文 ↗