Create a prediction module — module • dsprrr

Create a standard structured-prediction module. This is the primary constructor in the dsprrr journey: signature() -> module() -> run() -> evaluate() -> compile().

Agentic, reasoning, and code-executing programs use explicit constructors such as react(), chain_of_thought(), multi_chain_comparison(), program_of_thought(), code_act(), rlm_module(), and flex(). Keeping those choices in the function name prevents configuration from silently changing the kind of program being built.

Usage

module(
  signature,
  chat = NULL,
  template = "",
  demos = list(),
  config = list(),
  ...
)

Arguments

signature: A Signature object defining the module interface.
chat: Optional ellmer Chat object. When supplied, run() uses it unless an explicit .llm is provided.
template: Optional glue template for prompt generation.
demos: Optional list of demonstration examples.
config: Optional prediction configuration. Model parameters such as temperature belong here, for example config = list(temperature = 0.2).
...: Must be empty. Use a dedicated constructor for advanced module behavior.

Value

A PredictModule executed with run().

Examples

classifier <- signature("text -> sentiment") |>
  module(template = "Analyze: {text}")

configured <- signature("question -> answer") |>
  module(config = list(temperature = 0.2))

if (FALSE) { # \dontrun{
llm <- ellmer::chat_openai()
result <- classifier |>
  run(text = "Great package!", .llm = llm)
} # }