GReaTer: 推理上的梯度使小型语言模型成为强大的提示优化器

GReaTer: 梯度 over 推理 Makes Smaller Language Models Strong Prompt Optimizers论文概述本文是一篇关于提示工程的研究论文，由 Sarkar Snigdha Sarathi Das 等6位研究者共同完成。 GReaTer introduces a novel prompt 优化technique that directly incorporates gradient information over task-specific 推理, enabling open-source lightweight language models to self-optimize prompts without dependence on costly closed-source 大型语言模型 (LLM)s. Unlike text ...

阅读全文