View Source LangChain.Utils.ApiOverride (LangChain v0.3.0-rc.0)

Tools for overriding API results. Used for testing.

Works by setting and checking for special use of the Process dictionary.

Test Example

import LangChain.Utils.ApiOverride

model = ChatOpenAI.new!(%{temperature: 1, stream: true})

# Define the fake response to return
fake_messages = [
  [MessageDelta.new!(%{role: :assistant, content: nil, status: :incomplete})],
  [MessageDelta.new!(%{content: "Sock", status: :incomplete})]
]

# Made NOT LIVE here. Will not make the external call to the LLM
set_api_override({:ok, fake_messages})

# We can construct an LLMChain from a PromptTemplate and an LLM.
{:ok, updated_chain, _response} =
  %{llm: model, verbose: false}
  |> LLMChain.new!()
  |> LLMChain.add_message(
    Message.new_user!("What is a good name for a company that makes colorful socks?")
  )
  |> LLMChain.run()

assert %Message{role: :assistant, content: "Sock"} = updated_chain.last_message

Summary

Functions

get_api_override()

Get the API override to return. Returned as {:ok, config_tuple}. If not set, it returns :not_set.

override_api_return?()

Return if an override for the API response is set. Used for testing.

set_api_override(config_tuple)

Set the data and callback to use as a fake API response. An :ok tuple indicates a successful API call. The fake_response_data is the data to treat as returned. The callback_name is the callback handler name to execute.