TAG-DLM Unifies Textual Reasoning and Graph Message Passing

TAG-DLM integrates text and topology by injecting graph structure into a masked diffusion language model via a topology attention mask. This allows the model to perform joint textual reasoning and graph message passing within a single generative architecture.