Generative Representational Instruction Tuning: A Unified Model for Text Embedding and Generation
A single large language model can be trained to handle both text embedding and generation tasks by distinguishing between them through instructions, outperforming specialized models on both tasks.
