OpenDFM/ChemDFM-R-14B

ChemDFM-R-14B by OpenDFM is a 14 billion parameter chemical reasoning large language model specifically designed to enhance domain understanding and reasoning capabilities in chemistry. It utilizes a comprehensive dataset of atomized chemical knowledge, ChemFG, for functional group annotation and chemical reaction analysis. The model employs a mix-sourced distillation method and domain-specific reinforcement learning to achieve cutting-edge performance on diverse chemical benchmarks, providing interpretable and rationale-driven outputs for complex chemical problems.