A Dual-Mode Large Language Model Assistant for On-Surface Reaction via Fine-Tuning and Retrieval-Augmented Generation

Key Points

• The aim is to develop a language model assistant that facilitates better understanding of surface reactions through data retrieval and fine-tuning.
• Creation of a dual-mode large language model combining fine-tuning and retrieval-augmented generation techniques.
• Analysis of fragmented knowledge in existing chemical databases and literature.
• Systematic evaluation of the model's effectiveness in enabling data-driven discovery.
• Significant improvement in retrieving relevant information from unstructured literature.
• Enhanced capability to analyze surface reactions for various applications in catalysis and energy conversion.
• Illustration of the model's impact on accelerating molecular-scale fabrication processes.