improves专题

Self-Consistency Improves Chain of Thought Reasoning in Language Models阅读笔记

论文链接:https://arxiv.org/pdf/2203.11171.pdf 又到了读论文的时间,内心有点疲惫。这几天还是在看CoT的文章,今天这篇是讲如何利用self-consistency(自我一致性)来改进大语言模型的思维链推理过程。什么是self-consistency呢,读完论文感觉可以这么解释,就是有个渣男/大语言模型,你问了他五次昨天晚上九点跟谁在一起/文章里问大语言模型一个

Repetition Improves Language Model Embeddings

论文结论: echo embeddings将句子重复拼接送入到decoder-only模型中,将第二遍出现的句子特征pooling作为sentence embedding效果很好,优于传统方法 echo embeddings与传统embedding方法区别,如图所示: Classical embeddings: Feed sentence x to the language model

MVP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction

MVP: Multi-view Prompting Improves Aspect Sentiment Tuple Prediction 论文地址: https://arxiv.org/pdf/2305.12627.pdf 论文代码: https://github.com/ZubinGou/multi-view-prompting 个人阅读笔记, 如有错误欢迎指正交流 1. 介绍 Mul