Class: LangsmithrbRails::Evaluation::LLMEvaluator

Inherits:
Evaluator
  • Object
show all
Defined in:
lib/langsmithrb_rails/evaluation/llm_evaluator.rb

Overview

Evaluator that uses an LLM to evaluate responses

Instance Attribute Summary

Attributes inherited from Evaluator

#client, #project_name, #tags

Instance Method Summary collapse

Methods inherited from Evaluator

#evaluate_dataset, #evaluate_run, #evaluate_runs

Constructor Details

#initialize(llm:, criteria: nil, client: nil, project_name: nil, tags: []) ⇒ LLMEvaluator

Initialize a new LLM evaluator



15
16
17
18
19
 
# File 'lib/langsmithrb_rails/evaluation/llm_evaluator.rb', line 15

def initialize(llm:, criteria: nil, client: nil, project_name: nil, tags: [])
  super(client: client, project_name: project_name, tags: tags)
  @llm = llm
  @criteria = criteria || "Evaluate the response for accuracy, relevance, and completeness."
end

Instance Method Details

#evaluate(prediction, reference = nil, input = nil) ⇒ Hash

Evaluate a prediction against a reference



26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
 
# File 'lib/langsmithrb_rails/evaluation/llm_evaluator.rb', line 26

def evaluate(prediction, reference = nil, input = nil)
  # Extract strings
  prediction_str = extract_string(prediction)
  reference_str = extract_string(reference)
  input_str = input.is_a?(Hash) ? input.to_json : input.to_s if input
  
  # Create evaluation prompt
  prompt = create_evaluation_prompt(prediction_str, reference_str, input_str)
  
  # Get evaluation from LLM
  begin
    evaluation = get_llm_evaluation(prompt)
    
    # Parse the evaluation
    score, feedback = parse_evaluation(evaluation)
    
    {
      score: score,
      metadata: {
        feedback: feedback,
        criteria: @criteria,
        evaluation: evaluation
      }
    }
  rescue => e
    {
      score: nil,
      metadata: {
        error: "Evaluation failed: #{e.message}"
      }
    }
  end
end