langchain/docs/modules/models/getting_started.ipynb

Getting Started

One of the core value props of LangChain is that it provides a standard interface to models. This allows you to swap easily between models. At a high level, there are two main types of models:

• Language Models: good for text generation
• Text Embedding Models: good for turning text into a numerical representation

Language Models

There are two different sub-types of Language Models:

• LLMs: these wrap APIs which take text in and return text
• ChatModels: these wrap models which take chat messages in and return a chat message

This is a subtle difference, but a value prop of LangChain is that we provide a unified interface accross these. This is nice because although the underlying APIs are actually quite different, you often want to use them interchangeably.

To see this, let's look at OpenAI (a wrapper around OpenAI's LLM) vs ChatOpenAI (a wrapper around OpenAI's ChatModel).

In [1]:

from langchain.llms import OpenAI
from langchain.chat_models import ChatOpenAI

In [2]:

llm = OpenAI()

In [3]:

chat_model = ChatOpenAI()

text -> text interface

In [4]:

llm.predict("say hi!")

Out [4]:

'\n\nHi there!'

In [5]:

chat_model.predict("say hi!")

Out [5]:

'Hello there!'

messages -> message interface

In [6]:

from langchain.schema import HumanMessage

In [8]:

llm.predict_messages([HumanMessage(content="say hi!")])

Out [8]:

AIMessage(content='\n\nHello! Nice to meet you!', additional_kwargs={}, example=False)

In [9]:

chat_model.predict_messages([HumanMessage(content="say hi!")])

Out [9]:

AIMessage(content='Hello! How can I assist you today?', additional_kwargs={}, example=False)

In [ ]: