Class: Raif::Evals::LlmJudges::Binary

Inherits:
Raif::Evals::LlmJudge show all
Defined in:
lib/raif/evals/llm_judges/binary.rb

Constant Summary

Constants included from Concerns::LlmResponseParsing

Concerns::LlmResponseParsing::ASCII_CONTROL_CHARS

Instance Attribute Summary

Attributes inherited from Task

#files, #images

Instance Method Summary collapse

Methods inherited from Raif::Evals::LlmJudge

#default_llm_model_key, #judgment_confidence, #judgment_reasoning, #low_confidence?

Methods inherited from Task

json_response_schema, prompt, #re_run, run, #run, #status, system_prompt

Methods included from Concerns::LlmResponseParsing

#parse_html_response, #parse_json_response, #parsed_response

Methods included from Concerns::HasAvailableModelTools

#available_model_tools_map

Methods included from Concerns::HasRequestedLanguage

#requested_language_name, #system_prompt_language_preference

Methods included from Concerns::HasLlm

#default_llm_model_key, #llm

Methods inherited from ApplicationRecord

table_name_prefix

Instance Method Details

#build_prompt ⇒ Object 



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/raif/evals/llm_judges/binary.rb', line 33

def build_prompt
  prompt = <<~PROMPT
    Evaluation criteria: #{criteria}

    #{strict_mode ? "Apply the criteria strictly without any leniency." : "Apply reasonable judgment while adhering to the criteria."}
  PROMPT

  if examples.present?
    prompt += "\nHere are examples of how to evaluate:"
    examples.each do |example|
      prompt += format_example(example)
    end
  end

  prompt += additional_context_prompt if additional_context.present?

  prompt += <<~PROMPT.rstrip

    Now evaluate this content:
    #{content_to_judge}

    Does this content meet the evaluation criteria?
  PROMPT

  prompt
end

#build_system_prompt ⇒ Object 



17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
 
# File 'lib/raif/evals/llm_judges/binary.rb', line 17

def build_system_prompt
  <<~PROMPT.strip
    You are an expert evaluator assessing whether content meets specific criteria.
    Your task is to make binary pass/fail judgments with clear reasoning.

    First, provide detailed reasoning/explanation of your evaluation. Then, provide a precise pass/fail judgment.

    Respond with JSON matching this schema:
    {
      "passes": boolean,
      "reasoning": "detailed explanation",
      "confidence": 0.0-1.0
    }
  PROMPT
end

#passes? ⇒ Boolean

Judgment accessor methods

Returns:

  • (Boolean) 


61
62
63
 
# File 'lib/raif/evals/llm_judges/binary.rb', line 61

def passes?
  parsed_response["passes"] if completed?
end