Skip to contents

GEPA Teleprompter

Source: R/teleprompter-gepa.R
GEPA.Rd

Genetic/evolutionary prompt optimizer that evolves instruction variants using reflection on failed examples.

Usage

GEPA(
  metric = NULL,
  metric_threshold = NULL,
  max_errors = 5L,
  metrics = NULL,
  population_size = 20L,
  generations = 10L,
  mutation_rate = 0.1,
  crossover_rate = 0.7,
  selection = "pareto",
  seed = NULL,
  log_dir = NULL,
  verbose = TRUE,
  track_stats = TRUE
)

Arguments

metric

A single metric function (fallback when metrics is NULL).

metric_threshold

Minimum score for an example to be considered successful.

max_errors

Maximum number of errors allowed during optimization.

metrics

Named list of metric functions for evaluation.

population_size

Size of the population. Default is 20.

generations

Number of generations to run. Default is 10.

mutation_rate

Probability of mutation. Default is 0.1.

crossover_rate

Probability of crossover. Default is 0.7.

selection

Selection strategy: "pareto" or "current_best".

seed

Random seed for reproducibility.

log_dir

Optional directory for trial logging.

verbose

Whether to print progress messages.

track_stats

Whether to record generation statistics.