Class: CompletionKit::Run

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

Constant Summary collapse

STATUSES = 
%w[pending running completed failed].freeze

Constants inherited from ApplicationRecord

ApplicationRecord::TenantScopedUniquenessValidator

Class Method Summary collapse

Instance Method Summary collapse

Methods included from Taggable

#tag_names, #tag_names=

Class Method Details

.display_scoped ⇒ Object 



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

def self.display_scoped
  filter = CompletionKit.config.runs_display_scope
  filter ? all.instance_exec(&filter) : all
end

.visible_run_ids ⇒ Object 



29
30
31
 
# File 'app/models/completion_kit/run.rb', line 29

def self.visible_run_ids
  display_scoped.select(:id)
end

Instance Method Details

#as_json(options = {}) ⇒ Object 



321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
 
# File 'app/models/completion_kit/run.rb', line 321

def as_json(options = {})
  snap = progress_snapshot
  {
    id: id, name: name, status: status, prompt_id: prompt_id,
    dataset_id: dataset_id, judge_model: judge_model, temperature: temperature,
    output_column: output_column,
    created_at: created_at, updated_at: updated_at,
    responses_count: responses.count, avg_score: avg_score,
    check_pass_rate: check_pass_rate,
    progress_current: snap[:generated_done],
    progress_total: snap[:generated_total],
    progress: {
      generated: { done: snap[:generated_done], total: snap[:generated_total], failed: snap[:generated_failed] },
      judged:    { done: snap[:judged_done],    total: snap[:judged_total],    failed: snap[:judged_failed] }
    },
    failed_response_ids: responses.where(status: "failed").pluck(:id),
    failure_summary: failure_summary,
    error_message: error_message,
    metric_ids: metric_ids,
    tags: tags.as_json
  }
end

#avg_score ⇒ Object 



116
117
118
119
120
121
122
 
# File 'app/models/completion_kit/run.rb', line 116

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

#broadcast_actions ⇒ Object 



368
369
370
371
372
373
374
 
# File 'app/models/completion_kit/run.rb', line 368

def broadcast_actions
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "run_actions",
    html: render_engine_partial("completion_kit/runs/actions", run: self)
  )
end

#broadcast_clear_responses ⇒ Object 



384
385
386
387
388
389
390
 
# File 'app/models/completion_kit/run.rb', line 384

def broadcast_clear_responses
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "run_responses",
    html: '<tbody id="run_responses"></tbody>'
  )
end

#broadcast_progress ⇒ Object 



351
352
353
354
355
356
357
358
 
# File 'app/models/completion_kit/run.rb', line 351

def broadcast_progress
  reload
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "run_status_panel",
    html: render_engine_partial("completion_kit/runs/status_panel", run: self)
  )
end

#broadcast_response(response) ⇒ Object 



392
393
394
395
396
397
398
 
# File 'app/models/completion_kit/run.rb', line 392

def broadcast_response(response)
  broadcast_append_to(
    "completion_kit_run_#{id}",
    target: "run_responses",
    html: render_engine_partial("completion_kit/runs/response_row", run: self, response: response, index: responses.where("id <= ?", response.id).count)
  )
end

#broadcast_response_update(response) ⇒ Object 



400
401
402
403
404
405
406
 
# File 'app/models/completion_kit/run.rb', line 400

def broadcast_response_update(response)
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "response_#{response.id}",
    html: render_engine_partial("completion_kit/runs/response_row", run: self, response: response, index: responses.where("id <= ?", response.id).count)
  )
end

#broadcast_sort_toolbar ⇒ Object 



376
377
378
379
380
381
382
 
# File 'app/models/completion_kit/run.rb', line 376

def broadcast_sort_toolbar
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "run_sort_toolbar",
    html: render_engine_partial("completion_kit/runs/sort_toolbar", run: self)
  )
end

#broadcast_status_header ⇒ Object 



360
361
362
363
364
365
366
 
# File 'app/models/completion_kit/run.rb', line 360

def broadcast_status_header
  broadcast_replace_to(
    "completion_kit_run_#{id}",
    target: "run_status_header",
    html: render_engine_partial("completion_kit/runs/status_header", run: self)
  )
end

#broadcast_ui ⇒ Object 



344
345
346
347
348
349
 
# File 'app/models/completion_kit/run.rb', line 344

def broadcast_ui
  broadcast_progress
  broadcast_status_header
  broadcast_actions
  broadcast_sort_toolbar
end

#check_metrics ⇒ Object 



87
88
89
 
# File 'app/models/completion_kit/run.rb', line 87

def check_metrics
  metrics.where(metric_type: "check")
end

#check_pass_rate ⇒ Object 



140
141
142
143
144
145
146
 
# File 'app/models/completion_kit/run.rb', line 140

def check_pass_rate
  resolved = responses.flat_map(&:reviews).reject { |r| r.passed.nil? }
  return nil if resolved.empty?

  passed = resolved.count { |r| r.passed == true }
  (passed.to_f / resolved.length).round(2)
end

#generate_responses! ⇒ Object 



244
245
246
 
# File 'app/models/completion_kit/run.rb', line 244

def generate_responses!
  start!
end

#gradable? ⇒ Boolean

Returns:

  • (Boolean) 


95
96
97
 
# File 'app/models/completion_kit/run.rb', line 95

def gradable?
  llm_judge_configured? || check_metrics.any?
end

#gradable_metric_ids ⇒ Object 



54
55
56
57
58
 
# File 'app/models/completion_kit/run.rb', line 54

def gradable_metric_ids
  ids = check_metrics.pluck(:id)
  ids += llm_metrics.pluck(:id) if judge_model.present?
  ids
end

#judge_configured? ⇒ Boolean

Returns:

  • (Boolean) 


79
80
81
 
# File 'app/models/completion_kit/run.rb', line 79

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

#judge_only? ⇒ Boolean

A judge-only run grades a pre-existing column on the dataset instead of generating new outputs. No prompt is attached; the response text is read from row; no LLM generation happens.

Returns:

  • (Boolean) 


36
37
38
 
# File 'app/models/completion_kit/run.rb', line 36

def judge_only?
  prompt.nil?
end

#judge_only_input_data_checks? ⇒ Boolean

Returns:

  • (Boolean) 


99
100
101
102
103
104
105
106
 
# File 'app/models/completion_kit/run.rb', line 99

def judge_only_input_data_checks?
  return false unless judge_only?

  attached = run_metrics.filter_map(&:metric)
  return false if attached.empty?

  attached.all?(&:check?) && attached.all? { |m| m.check_config.to_h["target"] == "input_data" }
end

#llm_judge_configured? ⇒ Boolean

Returns:

  • (Boolean) 


91
92
93
 
# File 'app/models/completion_kit/run.rb', line 91

def llm_judge_configured?
  judge_model.present? && llm_metrics.any? && ApiConfig.valid_for_model?(judge_model)
end

#llm_metrics ⇒ Object 



83
84
85
 
# File 'app/models/completion_kit/run.rb', line 83

def llm_metrics
  metrics.where(metric_type: "llm_judge")
end

#mark_completed! ⇒ Object 



49
50
51
52
 
# File 'app/models/completion_kit/run.rb', line 49

def mark_completed!
  update!(status: "completed")
  broadcast_ui
end

#metric_averages ⇒ Object 



124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
 
# File 'app/models/completion_kit/run.rb', line 124

def metric_averages
  responses.flat_map(&:reviews).group_by(&:metric_name).filter_map do |name, reviews|
    scored = reviews.select { |r| r.ai_score.present? }
    if scored.any?
      scores = scored.map { |r| r.ai_score.to_f }
      { name: name, avg: (scores.sum / scores.length).round(1) }
    else
      resolved = reviews.reject { |r| r.passed.nil? }
      next if resolved.empty?

      passed = resolved.count { |r| r.passed == true }
      { name: name, kind: "check", pass_rate: (passed.to_f / resolved.length).round(2) }
    end
  end
end

#missing_dataset_variables ⇒ Object 



40
41
42
43
44
45
46
47
 
# File 'app/models/completion_kit/run.rb', line 40

def missing_dataset_variables
  return [] unless prompt
  vars = prompt.variables
  return [] if vars.empty?
  return vars if dataset.nil?

  vars - dataset.headers
end

#outstanding_work_zero? ⇒ Boolean

Returns:

  • (Boolean) 


60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
 
# File 'app/models/completion_kit/run.rb', line 60

def outstanding_work_zero?
  return false if responses.where.not(status: HasJobStatus::TERMINAL_STATUSES).exists?

  metric_ids = gradable_metric_ids
  return true if metric_ids.empty?

  succeeded_response_ids = responses.where(status: "succeeded").pluck(:id)
  expected_reviews = succeeded_response_ids.size * metric_ids.size
  return true if expected_reviews.zero?

  terminal_review_count = Review.where(
    response_id: succeeded_response_ids,
    metric_id: metric_ids,
    status: HasJobStatus::TERMINAL_STATUSES
  ).count

  terminal_review_count >= expected_reviews
end

#progress_snapshot ⇒ Object 



282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
 
# File 'app/models/completion_kit/run.rb', line 282

def progress_snapshot
  generated_done = responses.where(status: "succeeded").count
  generated_failed = responses.where(status: "failed").count
  generated_total = progress_total

  metric_ids = gradable_metric_ids
  metric_count = metric_ids.size
  judged_total = metric_count > 0 ? generated_done : 0
  judged_done = 0
  judged_failed = 0

  if metric_count > 0 && judged_total > 0
    succeeded_response_ids = responses.where(status: "succeeded").pluck(:id)
    review_counts = Review
      .where(response_id: succeeded_response_ids, metric_id: metric_ids)
      .group(:response_id, :status)
      .count
    succeeded_response_ids.each do |rid|
      ok = review_counts[[rid, "succeeded"]] || 0
      bad = review_counts[[rid, "failed"]] || 0
      next unless ok + bad == metric_count
      if bad > 0
        judged_failed += 1
      else
        judged_done += 1
      end
    end
  end

  {
    generated_done: generated_done,
    generated_total: generated_total,
    generated_failed: generated_failed,
    judged_done: judged_done,
    judged_total: judged_total,
    judged_failed: judged_failed
  }
end

#regrade! ⇒ Object 



248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
 
# File 'app/models/completion_kit/run.rb', line 248

def regrade!
  return false if metrics.empty? || !gradable?

  eligible_responses = responses.where(status: "succeeded")
  eligible_responses = eligible_responses.where.not(response_text: nil) unless judge_only_input_data_checks?
  response_ids = eligible_responses.pluck(:id)
  return false if response_ids.empty?

  transaction do
    Review.where(response_id: response_ids).update_all(
      status: "pending",
      attempts: 0,
      metric_version_id: nil,
      ai_score: nil,
      passed: nil,
      ai_feedback: nil,
      error_provider: nil,
      error_class: nil,
      error_status: nil,
      error_message: nil
    )
    update!(status: "running", failure_summary: nil, error_message: nil)

    response_ids.each do |rid|
      llm_metrics.each { |m| JudgeReviewJob.perform_later(rid, m.id, id) } if llm_judge_configured?
      check_metrics.each { |m| CheckReviewJob.perform_later(rid, m.id, id) }
    end
    RunCompletionCheckJob.perform_later(id)
  end

  broadcast_ui
  true
end

#replace_metrics!(metric_ids) ⇒ Object 



108
109
110
111
112
113
114
 
# File 'app/models/completion_kit/run.rb', line 108

def replace_metrics!(metric_ids)
  return unless metric_ids
  run_metrics.delete_all
  Array(metric_ids).reject(&:blank?).each_with_index do |metric_id, index|
    run_metrics.create!(metric_id: metric_id, position: index + 1)
  end
end

#stale_review_summary ⇒ Object 



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
174
 
# File 'app/models/completion_kit/run.rb', line 148

def stale_review_summary
  review_pairs = Review.where(response_id: response_ids)
                      .where.not(metric_id: nil)
                      .where.not(metric_version_id: nil)
                      .pluck(:metric_id, :metric_version_id, :metric_name)
  return {} if review_pairs.empty?

  metric_ids = review_pairs.map(&:first).uniq
  version_ids = review_pairs.map { |_, vid, _| vid }.uniq
  current_by_metric = MetricVersion.current.where(metric_id: metric_ids).pluck(:metric_id, :id, :version_number).each_with_object({}) do |(mid, vid, vnum), h|
    h[mid] = { id: vid, label: "v#{vnum}" }
  end
  label_by_version = MetricVersion.where(id: version_ids).pluck(:id, :version_number).each_with_object({}) { |(vid, vnum), h| h[vid] = "v#{vnum}" }

  summary = {}
  review_pairs.each do |metric_id, version_id, metric_name|
    current = current_by_metric[metric_id]
    next if current.nil?
    next if version_id == current[:id]
    label = label_by_version[version_id]
    next if label.nil?
    summary[metric_id] ||= { metric_name: metric_name, current_label: current[:label], stale_count: 0, scored_labels: [] }
    summary[metric_id][:stale_count] += 1
    summary[metric_id][:scored_labels] |= [label]
  end
  summary
end

#start! ⇒ Object 



176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
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
 
# File 'app/models/completion_kit/run.rb', line 176

def start!
  unless %w[pending failed].include?(status)
    return fail_with_summary!("Cannot start a run in state \"#{status}\". Use rerun to create a fresh copy, or retry_failures / regrade to work with the existing responses.")
  end

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

  return fail_with_summary!("Dataset has no rows") if rows.empty?

  if judge_only?
    column = output_column.presence || "actual_output"
    unless judge_only_input_data_checks? || (dataset && dataset.headers.include?(column))
      return fail_with_summary!("Dataset has no \"#{column}\" column")
    end
  else
    client = LlmClient.for_model(prompt.llm_model, ApiConfig.for_model(prompt.llm_model))
    unless client.configured?
      return fail_with_summary!("LLM API not configured: #{client.configuration_errors.join(', ')}")
    end
  end

  transaction do
    responses.destroy_all
    update!(
      status: "running",
      progress_current: 0,
      progress_total: rows.length,
      failure_summary: nil,
      error_message: nil
    )
    rows.each_with_index do |row, index|
      input = row.empty? ? nil : row.to_json
      attrs = {
        status: "pending",
        row_index: index,
        input_data: input,
        expected_output: row["expected_output"]
      }
      if judge_only?
        attrs[:status] = "succeeded"
        column = output_column.presence || "actual_output"
        attrs[:response_text] = row[column].to_s if dataset && dataset.headers.include?(column)
      end

      response = responses.create!(attrs)

      if judge_only?
        llm_metrics.each { |m| JudgeReviewJob.perform_later(response.id, m.id, id) } if llm_judge_configured?
        check_metrics.each { |m| CheckReviewJob.perform_later(response.id, m.id, id) }
      else
        GenerateRowJob.perform_later(id, response.id)
      end
    end

    RunCompletionCheckJob.perform_later(id) if judge_only?
  end

  safely_broadcast do
    broadcast_ui
    broadcast_clear_responses
  end
  true
end