Suggest tidymodels parameters for a module — module

Construct a dials::parameters() set from the information available on a module. Numeric parameters (e.g., temperature, top_p) derive their ranges from observed optimisation trials (if present) or fall back to sensible defaults. Qualitative parameters (e.g., prompt_style) are converted to value sets.

For reasoning models (OpenAI o1/o3/o4-mini, GPT-5 series), temperature and top_p are automatically excluded since these models don't support them. Instead, reasoning_effort is included as a tunable parameter.

Usage

module_parameters(
  module,
  model = NULL,
  include = NULL,
  exclude = c("id", "instructions", "instructions_suffix")
)

Arguments

module: A DSPrrr module (created with module()).
model: Optional model name string. When provided, parameters are filtered based on model capabilities (e.g., reasoning models exclude temperature/top_p and include reasoning_effort).
include: Optional character vector restricting which parameters are returned. Defaults to all parameters discovered in the module configuration and optimisation trials.
exclude: Character vector of parameter names to ignore. Defaults to internal bookkeeping fields such as id and instructions.

Value

A dials::parameters object describing the candidate tunables. Returns an empty parameter set when no tunables are discovered.

Examples

if (FALSE) { # \dontrun{
sig <- signature("text -> sentiment")
mod <- module(sig, type = "predict", config = list(temperature = 0.2))
optimize_grid(mod,
  devset = tibble::tibble(text = "sample", target = "positive"),
  parameters = list(temperature = c(0.1, 0.5))
)
module_parameters(mod)

# For reasoning models, temperature is excluded
module_parameters(mod, model = "o3")
} # }