大模型 / 大语言模型Large Language Model, LLM
Large Language Model, LLM
解决什么问题What problem it solves
读过海量文本、学会「接着往下写」的模型，能对话、能写作、能写代码。After training, the learned parameters that let a system generalise to inputs it has never seen.
这是什么What it is
参数量极大、在海量文本上训练出来的模型，通常几十亿到数千亿参数。
它的核心能力只有一句话：给定前面的文字，预测下一个最可能的词。
The trained artefact: parameters plus the architecture that produced them. "Loading a model" means loading these, not running a program.
为什么需要知道Why it matters
这是你日常接触最多的东西。选模型、选档位、算成本，都围绕它。Model and product are different things. Many misunderstandings about AI products come from treating a model as if it were an application.
别名Chinese aliases
大语言模型 · LLM · 语言模型