ChemDFM-R：一种融合原子化化学知识的化学推理大语言模型

摘要

尽管大型语言模型（LLMs）已取得显著进展，但其在化学等科学领域的应用仍受限于浅层的领域理解与有限的推理能力。本研究聚焦于化学这一特定领域，开发了化学推理大型语言模型ChemDFM-R。我们首先构建了一个包含原子化知识点的综合数据集，以增强模型对化学基本原理与逻辑结构的理解。随后，提出了一种混合源蒸馏策略，将专家精心整理的知识与通用领域的推理技能相结合，并通过领域特定的强化学习进一步提升化学推理能力。在多样化的化学基准测试中，ChemDFM-R展现了最先进的性能，同时提供了可解释、基于推理的输出。进一步的案例分析表明，显式的推理链显著提升了模型在现实人机协作场景中的可靠性、透明度及实际应用价值。

English

While large language models (LLMs) have achieved impressive progress, their application in scientific domains such as chemistry remains hindered by shallow domain understanding and limited reasoning capabilities. In this work, we focus on the specific field of chemistry and develop a Chemical Reasoner LLM, ChemDFM-R. We first construct a comprehensive dataset of atomized knowledge points to enhance the model's understanding of the fundamental principles and logical structure of chemistry. Then, we propose a mix-sourced distillation strategy that integrates expert-curated knowledge with general-domain reasoning skills, followed by domain-specific reinforcement learning to enhance chemical reasoning. Experiments on diverse chemical benchmarks demonstrate that ChemDFM-R achieves state-of-the-art performance while providing interpretable, rationale-driven outputs. Further case studies illustrate how explicit reasoning chains significantly improve the reliability, transparency, and practical utility of the model in real-world human-AI collaboration scenarios.

ChemDFM-R：一种融合原子化化学知识的化学推理大语言模型

ChemDFM-R: An Chemical Reasoner LLM Enhanced with Atomized Chemical Knowledge

摘要

Support