arXiv cs.AI· Sutanshu Saksena Raj, Daksh Maan, Avinash Benki, Parijat Chatterjee, Pasunuti Prasanjith, Purusharth Malik·· 3 小时前AI 评分34
强化学习后印度语系任务的提升与退化
Indic-Task Gains and Regressions after Reinforcement Learning
AI 导读
一项研究在覆盖英语和八种印度语言的 30B-A3B MoE 监督微调模型上，用 GRPO 在四个环境中进行强化学习，并跨 34 个基准对比前后表现。MATH 提升 36.07 分，多数印度语基准也改善，但 XQuAD-Indic F1 下降 19.32 分（英语答案占比从 8.7% 升至 91.7%），IndicQA 下降 8.64。
正文 · 原文
Abstract:We run reinforcement learning (RL) with GRPO across four environments, starting from a supervised fine-tuned 30B-A3B mixture-of-experts model covering English and eight Indic languages. We compare the model before and after RL across 34 benchmarks in one evaluation campaign. MATH improves by 36.07 points, and most Indic benchmarks also improve. Yet XQuAD-Indic F1 falls 19.32 points as English answers rise from 8.7% to 91.7%; IndicQA falls 8.64, while English XQuAD gains 11.71 F1. A second RL run from the same supervised initialization adds an Indic environment together with language and repetition rewards. Compared with the released model, it improves both passage-answering and instruction-following benchmarks, along with preserving MATH performance. But it loses ground on reasoning and coding benchmarks. Several training choices changed together, so we cannot isolate the effect of any one of the reward terms. Script compliance remains unresolved. Adding a language reward did not ensure use of the requested script. Passage answering and script compliance therefore need direct evaluation, even when other benchmark scores improve.
| Comments: | Accepted at the GlobalSouthAI Workshop, NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.16228 [cs.CL] |
| (or arXiv:2611.16228v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.16228 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sutanshu Saksena Raj [view email]
[v1]
Thu, 12 Nov 2026 03:21:07 UTC (22 KB)
来源：arXiv cs.AI · arxiv.org