What Is an LLM Really Doing During Inference? It’s More Than “Predicting the Next Token”
中文摘要
本文探讨了大模型推理的本质，指出其过程并非简单的“预测下一个Token”，而是更复杂的机制。
English Summary
This article explores LLM inference, explaining that the process involves more complex mechanisms than simply predicting the next token.
原文节选
A large language model does not “write an answer” the way a human writes a paragraph. During inference, it repeatedly predicts the next… Continue reading on Medium »