← 研踪 PaperTrailarXiv:2611.15717 · v1
Outside the Purview of the Reward: Capability Drift in Multilingual Post-Training
首次提交 2026-11-12 00:55
版本更新 2026-11-12 00:55
雷达观测 2026-11-13 14:16
摘要
Post-training can improve rewarded behavior while weakening capabilities users still need. We study two GRPO branches from a shared supervised checkpoint in a multilingual mixture-of-experts model. The released branch gains 36.07 points on MATH but loses 19.32 F1 points on XQuAD-Indic, while XQuAD-English improves under the same scoring rule. An independently trained sibling adds Indic instruction following and language-aware shaping. It recovers the regressing Indic task scores and preserves MATH, but does not restore script compliance and trades off competition mathematics and code. In a later round, judge training is followed by a sharp extraction decline from the Indic branch; reward-free distillation from its earlier checkpoint restores most of the lost F1. A MATH score decomposition associates most of the measured gain with increased boxed-answer emission; accuracy among boxed answers changes little. These results show why benchmark scores should be read alongside the response forms they reward.
作者备注：Accepted at the Foundations of LLM Post-Training in Changing Environments, NeurIPS 2026
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。
v1 · 2026-11-12 00:55 · 最新观测版本
Outside the Purview of the Reward: Capability Drift in Multilingual Post-Training
Post-training can improve rewarded behavior while weakening capabilities users still need. We study two GRPO branches from a shared supervised checkpoint in a multilingual mixture-of-experts model. The released branch gains 36.07 points on MATH but loses 19.32 F1 points on XQuAD-Indic, while XQuAD-English improves under the same scoring rule. An independently trained sibling adds Indic instruction following and language-aware shaping. It recovers the regressing Indic task scores and preserves MATH, but does not restore script compliance and trades off competition mathematics and code. In a later round, judge training is followed by a sharp extraction decline from the Indic branch; reward-free distillation from its earlier checkpoint restores most of the lost F1. A MATH score decomposition associates most of the measured gain with increased boxed-answer emission; accuracy among boxed answers changes little. These results show why benchmark scores should be read alongside the response forms they reward.
查看此版本解读 →查看此版本原文 ↗