针对孟加拉语数学题，构建了能自动分级难度的推理模型和数据集。
GanitLLM: Difficulty-Aware Bengali Mathematical Reasoning through Curriculum-GRPO
- 基于难度感知采样与可验证奖励设计新训练流程。
- 在两个孟加拉语数学数据集上准确率提升8~6个百分点。
- 适合关注低资源语言推理与教育AI的研究者。
我们提出名为GanitLLM（源自孟加拉语中‘数学’一词）的孟加拉语数学推理模型，同时构建了一个新的难度感知孟加拉语数学语料库及基于课程的GRPO训练流程。孟加拉语是全球使用最广泛的语言之一，但现有大模型或仅用英语推理后翻译，或在多步孟加拉语数学题上表现失败，部分原因是强化学习方法针对高资源语言优化，在低资源场景下因奖励稀疏而崩溃。为此，我们构建了经过严格过滤与去污染的孟加拉语数学数据集，通过强评估模型的pass@k自动标注难度。基于该数据集，提出Curriculum-GRPO：结合多阶段训练（SFT + GRPO）、难度感知采样与格式、数值正确性及孟加拉语推理的可验证奖励。在Bn-MGSM和Bn-MSVAMP数据集上，GanitLLM-4B相比其Qwen3-4B基线分别提升8和6个准确率点，孟加拉语推理词占比从14%增至超过88%，平均解题长度由943词降至193词。项目页面见 https://dipta007.github.io/GanitLLM
原文摘要 · Abstract (English)
We present a Bengali mathematical reasoning model called GanitLLM (named after the Bangla word for mathematics, Ganit), together with a new difficulty-aware Bengali math corpus and a curriculum-based GRPO pipeline. Bengali is one of the world's most widely spoken languages, yet existing LLMs either reason in English and then translate, or simply fail on multi-step Bengali math, in part because reinforcement learning recipes are tuned for high-resource languages and collapse under reward sparsity in low-resource settings. To address this, we construct Ganit, a rigorously filtered and decontaminated Bengali math dataset with automatic difficulty tags derived from the pass@k of a strong evaluator model. Building on this dataset, we propose Curriculum-GRPO, which combines multi-stage training (SFT + GRPO) with difficulty-aware sampling and verifiable rewards for format, numerical correctness, and Bengali reasoning. On Bn-MGSM and Bn-MSVAMP, GanitLLM-4B improves over its Qwen3-4B base by +8 and +6 accuracy points, respectively, while increasing the percentage of Bengali reasoning tokens from 14% to over 88% and reducing average solution length from 943 to 193 words. Project page is available at https://dipta007.github.io/GanitLLM
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。