Dspy.LM behaviour (dspy v0.1.0)

Behaviour for language model clients.

Defines the interface for interacting with different language model providers like OpenAI, Anthropic, local models, etc.

Summary

Types

message()

request()

response()

t()

Callbacks

generate(lm, request)

Generate a completion from the language model.

supports?(lm, feature)

Check if the language model supports a specific feature.

Functions

assistant_message(content)

Create an assistant message.

generate(request)

Generate a completion using the configured language model.

generate(lm, request)

Generate a completion using a specific language model.

generate_text(prompt, opts \\ [])

Generate text from a simple prompt string.

message(role, content)

Create a chat message.

supports?(lm, feature)

Check if a language model supports a feature.

system_message(content)

Create a system message.

user_message(content)

Create a user message.

Types

message()

@type message() :: %{role: String.t(), content: String.t()}

request()

@type request() :: %{
  messages: [message()],
  max_tokens: pos_integer() | nil,
  temperature: float() | nil,
  stop: [String.t()] | nil,
  tools: [map()] | nil
}

response()

@type response() :: %{
  choices: [%{message: message(), finish_reason: String.t() | nil}],
  usage:
    %{
      prompt_tokens: pos_integer(),
      completion_tokens: pos_integer(),
      total_tokens: pos_integer()
    }
    | nil
}

t()

@type t() :: struct()

Callbacks

generate(lm, request)

@callback generate(lm :: t(), request :: request()) :: {:ok, response()} | {:error, any()}

Generate a completion from the language model.

supports?(lm, feature)

(optional)

@callback supports?(lm :: t(), feature :: atom()) :: boolean()

Check if the language model supports a specific feature.

Functions

assistant_message(content)

Create an assistant message.

generate(request)

Generate a completion using the configured language model.

generate(lm, request)

Generate a completion using a specific language model.

generate_text(prompt, opts \\ [])

Generate text from a simple prompt string.

message(role, content)

Create a chat message.

supports?(lm, feature)

Check if a language model supports a feature.

system_message(content)

Create a system message.

user_message(content)

Create a user message.