深入理解Label Words are Anchors: An Information Flow Perspective for Understanding In-Context Learningabs
只测试了GPT模型
探究了ICL(in context learning)如何学习上下文的机制
提出”Information Flow with Labels as Anchors”假说...
llama2.c体验
Have you ever wanted to inference a baby Llama 2 model in pure C? No? Well, now you can!
为什么要体验llama2.c? 因为我现在在做一些大模型相关的东西, 但是设备资源又不是十分充足, 想要先通过一个很小很小的llama来验证跑通再迁移到实验环境。偶然发现llama2.c的环境...