Module: DSPy::Teleprompt::Utils

Extended by:
T::Sig
Defined in:
lib/dspy/teleprompt/utils.rb

Overview

Bootstrap utilities for MIPROv2 optimization Handles few-shot example generation and candidate program evaluation

Defined Under Namespace

Classes: BootstrapConfig, BootstrapResult

Class Method Summary collapse

Class Method Details

.create_n_fewshot_demo_sets(program, trainset, config: BootstrapConfig.new, metric: nil) ⇒ Object 



105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
 
# File 'lib/dspy/teleprompt/utils.rb', line 105

def self.create_n_fewshot_demo_sets(program, trainset, config: BootstrapConfig.new, metric: nil)
  DSPy::Context.with_span(
    operation: 'optimization.bootstrap_start',
    'dspy.module' => 'Bootstrap',
    'bootstrap.trainset_size' => trainset.size,
    'bootstrap.max_examples' => config.max_bootstrapped_examples,
    'bootstrap.num_candidate_sets' => config.num_candidate_sets
  ) do
    # Convert to typed examples if needed
    typed_examples = ensure_typed_examples(trainset)
    
    # Generate successful examples through bootstrap
    successful_examples, failed_examples = generate_successful_examples(
      program, 
      typed_examples, 
      config,
      metric
    )

    # Create candidate sets from successful examples
    candidate_sets = create_candidate_sets(successful_examples, config)

    # Gather statistics
    statistics = {
      total_trainset: trainset.size,
      successful_count: successful_examples.size,
      failed_count: failed_examples.size,
      success_rate: successful_examples.size.to_f / (successful_examples.size + failed_examples.size),
      candidate_sets_created: candidate_sets.size,
      average_set_size: candidate_sets.empty? ? 0 : candidate_sets.map(&:size).sum.to_f / candidate_sets.size
    }

    emit_bootstrap_complete_event(statistics)

    BootstrapResult.new(
      candidate_sets: candidate_sets,
      successful_examples: successful_examples,
      failed_examples: failed_examples,
      statistics: statistics
    )
  end
end

.eval_candidate_program(program, examples, config: BootstrapConfig.new, metric: nil) ⇒ Object 



157
158
159
160
161
162
163
164
 
# File 'lib/dspy/teleprompt/utils.rb', line 157

def self.eval_candidate_program(program, examples, config: BootstrapConfig.new, metric: nil)
  # Use minibatch evaluation for large datasets
  if examples.size > config.minibatch_size
    eval_candidate_program_minibatch(program, examples, config, metric)
  else
    eval_candidate_program_full(program, examples, config, metric)
  end
end

.eval_candidate_program_full(program, examples, config, metric) ⇒ Object 



200
201
202
203
204
205
206
207
208
209
210
211
 
# File 'lib/dspy/teleprompt/utils.rb', line 200

def self.eval_candidate_program_full(program, examples, config, metric)
  # Create evaluator with proper configuration
  evaluator = DSPy::Evaluate.new(
    program,
    metric: metric || default_metric_for_examples(examples),
    num_threads: config.num_threads,
    max_errors: config.max_errors
  )

  # Run evaluation
  evaluator.evaluate(examples, display_progress: false)
end

.eval_candidate_program_minibatch(program, examples, config, metric) ⇒ Object 



175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
 
# File 'lib/dspy/teleprompt/utils.rb', line 175

def self.eval_candidate_program_minibatch(program, examples, config, metric)
  DSPy::Context.with_span(
    operation: 'optimization.minibatch_evaluation',
    'dspy.module' => 'Bootstrap',
    'minibatch.total_examples' => examples.size,
    'minibatch.size' => config.minibatch_size,
    'minibatch.num_batches' => (examples.size.to_f / config.minibatch_size).ceil
  ) do
    # Randomly sample a minibatch for evaluation
    sample_size = [config.minibatch_size, examples.size].min
    sampled_examples = examples.sample(sample_size)
    
    eval_candidate_program_full(program, sampled_examples, config, metric)
  end
end