[_akhe]

It's a next token prediction library. "Markov chain" basically means finite state machine and is a looser concept. If you want to call the token prediction methodology Markovian you can - sounds cool! The implementation another person linked here ranking words would also qualify any LLM as using Markovian dynamics, but what is the point of calling it something so abstract?

More accurately, it's literally a language model:

  {
    I: { 
      want: { 
        to: { 
          be: { ... }, 
          know: { ... }
        }
      },
      will: { ... }
    },
    ...
  }

Every word of every sentence is modeled and ranked, and there are methods to perform operations on it. If you added a lot more words and phrases to the model, it would be a "large" language model. It also supports non-words though, so it's more accurately a "next token prediction library" that can be used to create language models.