本記事は、当社が開発・公開した数理推論データセット APTOinc/llm-math-reasoning-dataset の設計背景、仕様、検証結果、そして今後の展望をまとめたものです。 大規模言語モデル（LLM）は、近年その性能を飛躍的に向上させていますが、複数ステップの計算や厳密 ...

We present rStar-Math to demonstrate that small language models (SLMs) can rival or even surpass the math reasoning capability of OpenAI o1, without distillation from superior models. rStar-Math ...

Recent advances in Vision Language Models (VLMs) have shown significant progress in mathematical reasoning, yet they still face a critical bottleneck with problems that require visual assistance, such ...

TL;DR: We enhance the mathematical reasoning ability of LLMs solely through Verifiable Reward filtering and the self-improvement training paradigm of DPO. The final model, Qwen2.5-7B-DPO-VP, ...

Math Riddles are so challenging, but that makes them worthwhile to solve. Math riddles are logical problems that require strong analytical abilities, high IQ, knowledge of math concepts, and good ...

Microsoft researchers have developed ‘rStar-Math’, a method that enables small language models (SLMs) to solve challenging math problems with remarkable accuracy, matching or even surpassing larger ...

Mathematical reasoning has long presented a formidable challenge for AI, demanding not only an understanding of abstract concepts but also the ability to perform multi-step logical deductions with ...

Solving math riddles is fun, especially when you get them right. But did you know that solving them can also improve your brain power? In fact, research shows that practicing math riddles helps ...

Despite their impressive performance, Apple’s research reveals that large language models still struggle with true mathematical reasoning, relying on pattern-matching instead of formal logic - a ...