Class: DSPy::Teleprompt::GEPA

Inherits:
Teleprompter show all
Extended by:
T::Sig
Defined in:
lib/dspy/teleprompt/gepa.rb

Overview

GEPA: Genetic-Pareto Reflective Prompt Evolution optimizer Uses natural language reflection to evolve prompts through genetic algorithms and Pareto frontier selection for maintaining diverse high-performing candidates

Defined Under Namespace

Classes: CrossoverEngine, CrossoverType, ExecutionTrace, FitnessEvaluator, FitnessScore, GEPAConfig, GeneticEngine, InstructionProposer, MutationEngine, MutationType, ParetoSelector, ReflectionEngine, ReflectionResult, TraceCollector

Instance Attribute Summary collapse

Attributes inherited from Teleprompter

#evaluator, #metric

Instance Method Summary collapse

Methods inherited from Teleprompter

#create_evaluator, #ensure_typed_examples, #evaluate_program, #save_results, #validate_inputs

Constructor Details

#initialize(metric: nil, config: nil) ⇒ GEPA

Returns a new instance of GEPA.



2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
 
# File 'lib/dspy/teleprompt/gepa.rb', line 2476

def initialize(metric: nil, config: nil)
  @config = config || GEPAConfig.new

  # Validate that reflection_lm is configured
  unless @config.reflection_lm
    raise ArgumentError, "reflection_lm must be configured for GEPA optimization. Set config.reflection_lm to a DSPy::LM instance."
  end

  super(metric: metric, config: @config)
end

Instance Attribute Details

#config ⇒ Object (readonly)

Returns the value of attribute config.



2468
2469
2470
 
# File 'lib/dspy/teleprompt/gepa.rb', line 2468

def config
  @config
end

Instance Method Details

#compile(program, trainset:, valset: nil) ⇒ Object 



2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
 
# File 'lib/dspy/teleprompt/gepa.rb', line 2496

def compile(program, trainset:, valset: nil)
  validate_inputs(program, trainset, valset)

  instrument_step('gepa_compile', {
    trainset_size: trainset.size,
    valset_size: valset&.size || 0,
    num_generations: @config.num_generations,
    population_size: @config.population_size
  }) do
    # Always perform full GEPA genetic algorithm optimization
    perform_gepa_optimization(program, trainset, valset)
  end
end