[ICLR'26] A scalable, effective, and general way to augment LLMs with billion-scale knowledge graphs using very little GPU memory cost. - View it on GitHub