View Source LangChain.Chains.RoutingChain (LangChain v0.3.0-rc.0)

Run a router based on a user's initial prompt to determine what category best matches from the given options. If there is no good match, the value "DEFAULT" is returned.

Here's an example:

routes = [
  PromptRoute.new!(%{
    name: "marketing_email",
    description: "Create a marketing focused email",
    chain: marketing_email_chain
  }),
  PromptRoute.new!(%{
    name: "blog_post",
    description: "Create a blog post that will be linked from the company's landing page",
    chain: blog_post_chain
  }),
]

selected_route =
  RoutingChain.new(%{
    llm: ChatOpenAI.new(%{model: "gpt-3.5-turbo", stream: false}),
    input_text: "Let's create a marketing blog post about our new product 'Fuzzy Furries'",
    routes: routes,
    default_route: PromptRoute.new!(%{name: "DEFAULT", chain: fallback_chain})
  })
  |> RoutingChain.evaluate()

# The PromptRoute for the `blog_post` should be returned as the `selected_route`.

The llm is the model used to make the determination of which route is the best match. A smaller, faster LLM may be a great choice for the routing decision, then a more complex LLM may be used for a selected route.

The default_route is required and is used as a fallback if the user's prompt doesn't match any of the specified routes. It may also be used in some fallback error situations as well.

Summary

Functions

Runs the RoutingChain and evaluates the result to return the selected chain.

Start a new RoutingChain.

Start a new RoutingChain and return it or raise an error if invalid.

Run a simple RoutingChain to analyze the input_text and determine which of the given routes is the best match.

Types

@type t() :: %LangChain.Chains.RoutingChain{
  default_route: term(),
  input_text: term(),
  llm: term(),
  routes: term(),
  verbose: term()
}

Functions

Link to this function

evaluate(chain, opts \\ [])

View Source
@spec evaluate(t(), Keyword.t()) :: LangChain.Routing.PromptRoute.t()

Runs the RoutingChain and evaluates the result to return the selected chain.

@spec new(attrs :: map()) :: {:ok, t()} | {:error, Ecto.Changeset.t()}

Start a new RoutingChain.

@spec new!(attrs :: map()) :: t() | no_return()

Start a new RoutingChain and return it or raise an error if invalid.

@spec run(t(), Keyword.t()) ::
  {:ok, LangChain.Chains.LLMChain.t(),
   LangChain.Message.t() | [LangChain.Message.t()]}
  | {:error, String.t()}

Run a simple RoutingChain to analyze the input_text and determine which of the given routes is the best match.

A simpler, faster LLM may be a great fit for running the analysis. If it fails to find a good match, the default_route is used. The default_route's name is supplied to the LLM as well. The name "DEFAULT" is suggested for this route.