Using chat models

Chat models are increasingly adopted for their ability to handle conversational tasks,, and as we’ll see later, a lot of features (such as multimodal inputs) are supported on LangChain only for chat models. There are two key differences between a chat model and an LLM interface:

• A chat model takes either a prompt or a list of BaseMessage as input
• A chat model provides BaseMessage as output

First, what is BaseMessage? It’s a data structure to represent messages with three fields – content, type (who is the author of the message), and additional_kwargs as an optional dict. Here’s an example:

from langchain_core.messages import BaseMessage
message = BaseMessage(
    content="Hi, how are you?",
    type="human", additional_kwargs={"chapter": 2})

In practice, we typically deal with HumanMessage or AIMessage (the names of these classes speak for themselves...