Runnables and LCEL

Problem-solving typically involves a series of steps. When using LLMs, some applications have relatively complex requirements and a sequence of calls for performing an operation. Here’s when runnables come into play.

What is a runnable?

In LangChain, runnables are a powerful abstraction representing any callable unit of work. They're used to encapsulate and manage different kinds of tasks, including LLM calls, database queries, or calls to external APIs. This allows us to chain together diverse operations in a consistent and manageable way, making it easy to construct complex workflows within LangChain applications.

We have been using the ChatGroq ChatModel with the invoke() method to send our queries to the model. The chat model class is a type of a runnable component. Runnable components use a runnable interface that enables these components to work in a consistent manner. Methods like invoke(), batch() and stream() are common for all runnable components.

• invoke() processes a single input and returns a single output, ideal for individual requests.

• batch() allows us to process a list of inputs simultaneously, returning a list of corresponding outputs, which is much more efficient for handling multiple requests at once.

• stream() processes a single input but returns an output as a stream of chunks, which is useful for displaying real-time progress or handling very large outputs.

So far, we have used the Prompt , ChatModel, and OutputParser earlier in the course. These are all types of runnables. If you were paying attention, you might have noticed that we were using the output of one component as the input for another.

1. The PromptValue output from the PromptTemplate was passed as input to the ChatModel.

2. The string output from the ChatModel was passed as input to the OutputParser.

Essentially, runnable components take an input and transform it into an output. Now, imagine a more complex workflow. It would become difficult to create and manage this series ...