Class: DSPy::Propose::GroundedProposer

Inherits:
Object
  • Object
show all
Extended by:
T::Sig
Defined in:
lib/dspy/propose/grounded_proposer.rb

Overview

Grounded Proposer for generating better instructions based on training data Analyzes task patterns and creates contextually appropriate instructions

Defined Under Namespace

Classes: Config, ProposalResult

Constant Summary collapse

MAX_HISTORY_INSTRUCTIONS = 
5
TIPS =

Python-compatible TIPS dictionary for instruction generation

{
  "none" => "",
  "creative" => "Don't be afraid to be creative when creating the new instruction!",
  "simple" => "Keep the instruction clear and concise.",
  "description" => "Make sure your instruction is very informative and descriptive.",
  "high_stakes" => "The instruction should include a high stakes scenario in which the LM must solve the task!",
  "persona" => 'Include a persona that is relevant to the task in the instruction (ie. "You are a ...")'
}.freeze

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(config: nil, program: nil, trainset: nil) ⇒ GroundedProposer

Returns a new instance of GroundedProposer.



146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
 
# File 'lib/dspy/propose/grounded_proposer.rb', line 146

def initialize(config: nil, program: nil, trainset: nil)
  @config = config || Config.new
  @program = program
  @trainset = trainset
  @dataset_summary = nil
  @program_code_string = nil

  # Generate dataset summary if data-aware mode enabled (Python: use_dataset_summary)
  if @config.use_dataset_summary && trainset && !trainset.empty?
    begin
      require_relative 'dataset_summary_generator'
      @dataset_summary = DatasetSummaryGenerator.create_dataset_summary(
        trainset,
        @config.view_data_batch_size,
        DSPy.current_lm,
        verbose: @config.verbose
      )
    rescue => e
      DSPy.logger.warn("Failed to generate dataset summary: #{e.message}")
      @dataset_summary = nil
    end
  end

  # Extract program source code if program-aware mode enabled
  if @config.program_aware && program
    @program_code_string = extract_program_source(program)
  end
end

Instance Attribute Details

#config ⇒ Object (readonly)

Returns the value of attribute config.



137
138
139
 
# File 'lib/dspy/propose/grounded_proposer.rb', line 137

def config
  @config
end

Instance Method Details

#propose_instructions(signature_class, examples, few_shot_examples: nil, current_instruction: nil, trial_logs: nil) ⇒ Object 



209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
 
# File 'lib/dspy/propose/grounded_proposer.rb', line 209

def propose_instructions(signature_class, examples, few_shot_examples: nil, current_instruction: nil, trial_logs: nil)
  DSPy::Context.with_span(
    operation: 'optimization.instruction_proposal',
    'dspy.module' => 'GroundedProposer',
    'proposal.signature' => signature_class.name,
    'proposal.num_examples' => examples.size,
    'proposal.has_few_shot' => !few_shot_examples.nil?,
    'proposal.has_current_instruction' => !current_instruction.nil?
  ) do
    # Analyze the task and training data
    analysis = analyze_task(signature_class, examples, few_shot_examples)
    
    # Generate instruction candidates
    candidates = generate_instruction_candidates(
      signature_class,
      analysis,
      current_instruction,
      few_shot_examples: few_shot_examples,
      trial_logs: trial_logs
    )

    # Filter and rank candidates
    filtered_candidates = filter_and_rank_candidates(candidates, analysis)

    metadata = {
      generation_timestamp: Time.now.iso8601,
      model_used: DSPy.current_lm.model,
      num_examples_analyzed: [examples.size, @config.view_data_batch_size].min,
      original_instruction: current_instruction
    }

    result = ProposalResult.new(
      candidate_instructions: filtered_candidates,
      analysis: analysis,
      metadata: metadata
    )

    emit_proposal_complete_event(result)
    result
  end
end

#propose_instructions_for_program(trainset:, program:, demo_candidates:, trial_logs: nil, num_instruction_candidates: nil) ⇒ Object 



260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
 
# File 'lib/dspy/propose/grounded_proposer.rb', line 260

def propose_instructions_for_program(trainset:, program:, demo_candidates:, trial_logs: nil, num_instruction_candidates: nil)
  num_candidates = num_instruction_candidates || @config.num_instruction_candidates

  current_instruction = if program.respond_to?(:prompt) && program.prompt.respond_to?(:instruction)
    program.prompt.instruction
  else
    nil
  end

  few_shot_examples = demo_candidates[0]&.flatten&.take(@config.num_demos_in_context) || []

  signature_class = if program.respond_to?(:signature_class)
    program.signature_class
  else
    raise ArgumentError, "Program must expose signature_class for instruction proposal"
  end

  base_result = propose_instructions(
    signature_class,
    trainset,
    few_shot_examples: few_shot_examples,
    current_instruction: current_instruction,
    trial_logs: trial_logs
  )

  predictor_instructions = { 0 => base_result.candidate_instructions.take(num_candidates) }

  ProposalResult.new(
    candidate_instructions: base_result.candidate_instructions,
    analysis: base_result.analysis,
    metadata: base_result.metadata,
    predictor_instructions: predictor_instructions
  )
end