Class: TestRunner

Inherits:
Object
  • Object
show all
Defined in:
lib/kaba/test_runner.rb

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(path, schema:, type_name:, prompt:, validate:) ⇒ TestRunner 



4
5
6
7
8
9
10
11
12
13
14
15
16
 
# File 'lib/kaba/test_runner.rb', line 4

def initialize(path, schema:, type_name:, prompt:, validate:)

  @test_files = Dir.glob(File.join(File.expand_path(path), '*.target.json'))
  @lines = []
  @schema = schema
  @type_name = type_name
  @prompt = prompt || Prompt.new(@schema, @type_name)
  @validate = validate || Validate.new(schema: @schema, type_name: @type_name)

  @type_right_total = 0
  @score_total = 0

end

Class Method Details

.report_template_path ⇒ Object 



105
106
107
 
# File 'lib/kaba/test_runner.rb', line 105

def report_template_path
  @report_template_path || File.join(__dir__, 'report.html.erb')
end

.report_template_path=(path) ⇒ Object 



109
110
111
 
# File 'lib/kaba/test_runner.rb', line 109

def report_template_path=(path)
  @report_template_path = path
end

Instance Method Details

#_each(limit: nil) ⇒ Object 



18
19
20
21
22
 
# File 'lib/kaba/test_runner.rb', line 18

def _each(limit: nil)
  @test_files.first(limit || @test_files.size).each do |file|
    yield(Row.new(file), self)
  end
end

#save(file_path) ⇒ Object 



97
98
99
100
101
 
# File 'lib/kaba/test_runner.rb', line 97

def save(file_path)
  File.open(File.expand_path(file_path), 'w') do |file|
    file.puts ERB.new(File.read(self.class.report_template_path)).result(binding)
  end
end

#scan(limit: nil, model: 'spark-general-4.0', judge_model: 'spark-general-4.0', judge_temperature: 0.1, temperature: 0.1, semaphore_limit: 5) ⇒ Object 



24
25
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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
 
# File 'lib/kaba/test_runner.rb', line 24

def scan(
  limit: nil, 
  model: 'spark-general-4.0', 
  judge_model: 'spark-general-4.0',  
  judge_temperature: 0.1, 
  temperature: 0.1,
  semaphore_limit: 5
  )

  progressbar = TTY::ProgressBar.new(
    "Test: [:bar] :percent :current/:total",
     total: @test_files.first(limit || @test_files.size).size
  )

  progressbar.start

  Async do
    semaphore = Async::Semaphore.new(semaphore_limit)
    _each(limit: limit) do |row|
      semaphore.async do |task|
        input = @prompt.render(File.read row.input_file)

        target = "        ```json\n        \#{JSON.pretty_generate(JSON.parse(File.read(row.target_path)))}\n        ```\n        Markdown\n        output = \"\"\n        Application.llm_client.chat(\n          parameters: {\n            model: model,\n            messages: [ { role: 'user', content: input } ],\n            temperature: temperature,\n            stream: proc do |chunk, _bytesize|\n              output += chunk.dig(\"choices\", 0, \"delta\", \"content\")\n            end\n          }\n        )\n\n        output_json = JSON.parse_llm_response output\n\n        type_check_response = JSON.parse @validate.run(output_json).body\n        @type_right_total += 1 if type_check_response[\"success\"]\n        \n        judge_input = Judge.new(input: input, output: output, target: target).render\n        judge_response = Application.judge_llm_client.chat(\n          parameters: {\n            model: judge_model,\n            messages: [ { role: 'user', content: judge_input } ],\n            temperature: judge_temperature,\n          }\n        ).dig(\"choices\", 0, \"message\", \"content\")\n\n        judge_json = JSON.parse_llm_response judge_response\n        @score_total += judge_json[\"score\"].to_i\n\n        @lines << {\n          row: row,\n          input: input,\n          output: output,\n          output_json: output_json,\n          type_check_response: type_check_response,\n          target: target,\n          judge_response: judge_response,\n          judge_json: judge_json,\n        }\n\n        progressbar.advance\n      end\n    end\n  end.wait\nend\n"