Class: CompletionKit::Run

Inherits:
ApplicationRecord show all
Includes:
Turbo::Broadcastable
Defined in:
app/models/completion_kit/run.rb

Constant Summary collapse

STATUSES = 
%w[pending generating judging completed failed].freeze

Instance Method Summary collapse

Instance Method Details

#as_json(options = {}) ⇒ Object 



156
157
158
159
160
161
162
163
164
165
 
# File 'app/models/completion_kit/run.rb', line 156

def as_json(options = {})
  {
    id: id, name: name, status: status, prompt_id: prompt_id,
    dataset_id: dataset_id, judge_model: judge_model, temperature: temperature,
    created_at: created_at, updated_at: updated_at,
    responses_count: responses.count, avg_score: avg_score,
    progress_current: progress_current, progress_total: progress_total,
    error_message: error_message, metric_ids: metric_ids
  }
end

#avg_score ⇒ Object 



24
25
26
27
28
29
30
 
# File 'app/models/completion_kit/run.rb', line 24

def avg_score
  all_reviews = responses.flat_map(&:reviews)
  scores = all_reviews.map(&:ai_score).compact.map(&:to_f)
  return nil if scores.empty?

  (scores.sum / scores.length).round(2)
end

#generate_responses! ⇒ Object 



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
96
97
98
99
100
 
# File 'app/models/completion_kit/run.rb', line 40

def generate_responses!
  rows = if dataset
           CsvProcessor.process_self(self)
         else
           [{}]
         end

  if rows.empty?
    errors.add(:base, "Dataset has no rows")
    return false
  end

  client = LlmClient.for_model(prompt.llm_model, ApiConfig.for_model(prompt.llm_model))

  unless client.configured?
    msg = "LLM API not configured: #{client.configuration_errors.join(', ')}"
    errors.add(:base, msg)
    update_columns(status: "failed", error_message: msg) if persisted?
    return false
  end

  update!(status: "generating", progress_current: 0, progress_total: rows.length, error_message: nil)
  responses.destroy_all
  broadcast_ui
  broadcast_clear_responses

  rows.each_with_index do |row, index|
    input = row.empty? ? nil : row.to_json
    rendered = CsvProcessor.apply_variables(prompt, row)
    response_text = client.generate_completion(rendered, model: prompt.llm_model, temperature: temperature)

    resp = responses.create!(
      input_data: input,
      response_text: response_text,
      expected_output: row["expected_output"]
    )

    update_columns(progress_current: index + 1)
    broadcast_progress
    broadcast_response(resp)
  end

  if judge_configured?
    judge_responses!
  else
    update!(status: "completed")
    broadcast_ui
  end

  true
rescue Faraday::Error => e
  update_columns(status: "failed", error_message: e.message)
  errors.add(:base, e.message)
  broadcast_ui
  false
rescue StandardError => e
  update_columns(status: "failed", error_message: e.message) if persisted?
  errors.add(:base, e.message)
  broadcast_ui if persisted?
  false
end

#judge_configured? ⇒ Boolean

Returns:

  • (Boolean) 


20
21
22
 
# File 'app/models/completion_kit/run.rb', line 20

def judge_configured?
  judge_model.present? && metrics.any? && ApiConfig.valid_for_model?(judge_model)
end

#judge_responses! ⇒ Object 



102
103
104
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
147
148
149
150
151
152
153
154
 
# File 'app/models/completion_kit/run.rb', line 102

def judge_responses!
  total_evaluations = responses.count * metrics.count
  update!(status: "judging", progress_current: 0, progress_total: total_evaluations, error_message: nil)
  broadcast_ui

  judge = JudgeService.new(ApiConfig.for_model(judge_model).merge(judge_model: judge_model))
  evaluation_count = 0

  responses.find_each do |response|
    metrics.each do |metric|
      evaluation = judge.evaluate(
        response.response_text,
        response.expected_output,
        prompt.template,
        criteria: metric.respond_to?(:instruction) ? metric.instruction.to_s : "",
        evaluation_steps: metric.respond_to?(:evaluation_steps) ? metric.evaluation_steps : nil,
        rubric_text: metric.respond_to?(:display_rubric_text) ? metric.display_rubric_text : nil,
        input_data: response.input_data
      )

      response.reviews.find_or_initialize_by(metric_id: metric.id).tap do |review|
        review.assign_attributes(
          metric_name: metric.name,
          instruction: metric.respond_to?(:instruction) ? metric.instruction.to_s : "",
          status: "evaluated",
          ai_score: evaluation[:score],
          ai_feedback: evaluation[:feedback]
        )
        review.save!
      end

      evaluation_count += 1
      update_columns(progress_current: evaluation_count)
      broadcast_progress
    end

    broadcast_response_update(response)
  end

  update!(status: "completed")
  broadcast_ui
  true
rescue Faraday::Error => e
  update_columns(status: "failed", error_message: e.message)
  errors.add(:base, e.message)
  broadcast_ui
  false
rescue StandardError => e
  update_columns(status: "failed", error_message: e.message) if persisted?
  errors.add(:base, e.message)
  broadcast_ui if persisted?
  false
end

#metric_averages ⇒ Object 



32
33
34
35
36
37
38
 
# File 'app/models/completion_kit/run.rb', line 32

def metric_averages
  all_reviews = responses.flat_map(&:reviews).select { |r| r.ai_score.present? }
  all_reviews.group_by(&:metric_name).map do |name, reviews|
    scores = reviews.map { |r| r.ai_score.to_f }
    { name: name, avg: (scores.sum / scores.length).round(1) }
  end
end