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 



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

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

.visible_run_ids ⇒ Object 



32
33
34
 
# File 'app/models/completion_kit/run.rb', line 32

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

Instance Method Details

#as_json(options = {}) ⇒ Object 



340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
 
# File 'app/models/completion_kit/run.rb', line 340

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,
    expected_column: expected_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 



119
120
121
122
123
124
125
 
# File 'app/models/completion_kit/run.rb', line 119

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 



388
389
390
391
392
393
394
 
# File 'app/models/completion_kit/run.rb', line 388

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 



404
405
406
407
408
409
410
 
# File 'app/models/completion_kit/run.rb', line 404

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

#broadcast_progress ⇒ Object 



371
372
373
374
375
376
377
378
 
# File 'app/models/completion_kit/run.rb', line 371

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 



412
413
414
415
416
417
418
 
# File 'app/models/completion_kit/run.rb', line 412

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 



420
421
422
423
424
425
426
 
# File 'app/models/completion_kit/run.rb', line 420

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 



396
397
398
399
400
401
402
 
# File 'app/models/completion_kit/run.rb', line 396

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 



380
381
382
383
384
385
386
 
# File 'app/models/completion_kit/run.rb', line 380

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 



364
365
366
367
368
369
 
# File 'app/models/completion_kit/run.rb', line 364

def broadcast_ui
  broadcast_progress
  broadcast_status_header
  broadcast_actions
  broadcast_sort_toolbar
end

#check_metrics ⇒ Object 



90
91
92
 
# File 'app/models/completion_kit/run.rb', line 90

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

#check_pass_rate ⇒ Object 



143
144
145
146
147
148
149
 
# File 'app/models/completion_kit/run.rb', line 143

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 



263
264
265
 
# File 'app/models/completion_kit/run.rb', line 263

def generate_responses!
  start!
end

#gradable? ⇒ Boolean

Returns:

  • (Boolean) 


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

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

#gradable_metric_ids ⇒ Object 



57
58
59
60
61
 
# File 'app/models/completion_kit/run.rb', line 57

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) 


82
83
84
 
# File 'app/models/completion_kit/run.rb', line 82

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

#judge_only? ⇒ Boolean

A scoring-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) 


39
40
41
 
# File 'app/models/completion_kit/run.rb', line 39

def judge_only?
  prompt.nil?
end

#judge_only_input_data_checks? ⇒ Boolean

Returns:

  • (Boolean) 


102
103
104
105
106
107
108
109
 
# File 'app/models/completion_kit/run.rb', line 102

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) 


94
95
96
 
# File 'app/models/completion_kit/run.rb', line 94

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

#llm_metrics ⇒ Object 



86
87
88
 
# File 'app/models/completion_kit/run.rb', line 86

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

#mark_completed! ⇒ Object 



52
53
54
55
 
# File 'app/models/completion_kit/run.rb', line 52

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

#metric_averages ⇒ Object 



127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
 
# File 'app/models/completion_kit/run.rb', line 127

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 



43
44
45
46
47
48
49
50
 
# File 'app/models/completion_kit/run.rb', line 43

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) 


63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
 
# File 'app/models/completion_kit/run.rb', line 63

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 



301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
 
# File 'app/models/completion_kit/run.rb', line 301

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 



267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
 
# File 'app/models/completion_kit/run.rb', line 267

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 



111
112
113
114
115
116
117
 
# File 'app/models/completion_kit/run.rb', line 111

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 



151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
 
# File 'app/models/completion_kit/run.rb', line 151

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?
    label = label_by_version[version_id]
    next if label.nil?
    next if label == current[:label]
    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 



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
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
 
# File 'app/models/completion_kit/run.rb', line 179

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

  begin
    transaction do
      responses.destroy_all
      update!(
        status: "running",
        progress_current: 0,
        progress_total: rows.length,
        failure_summary: nil,
        error_message: nil
      )

      now = Time.current
      out_col = output_column.presence || "actual_output"
      exp_col = expected_column.presence || "expected_output"
      judge = judge_only?
      has_output = judge && dataset && dataset.headers.include?(out_col)
      scope_defaults = Response.all.where_values_hash.symbolize_keys

      response_attrs = rows.each_with_index.map do |row, index|
        {
          run_id: id,
          status: judge ? "succeeded" : "pending",
          row_index: index,
          input_data: row.empty? ? nil : row.to_json,
          expected_output: row[exp_col],
          response_text: (judge && has_output ? row[out_col].to_s : nil),
          attempts: 0,
          created_at: now,
          updated_at: now
        }.merge(scope_defaults)
      end
      Response.insert_all(response_attrs)
      responses.reset

      response_ids = responses.order(:row_index).pluck(:id)
      if judge
        judge_metrics = llm_judge_configured? ? llm_metrics.to_a : []
        chk_metrics = check_metrics.to_a
        response_ids.each do |rid|
          judge_metrics.each { |m| JudgeReviewJob.perform_later(rid, m.id, id) }
          chk_metrics.each { |m| CheckReviewJob.perform_later(rid, m.id, id) }
        end
        RunCompletionCheckJob.perform_later(id)
      else
        response_ids.each { |rid| GenerateRowJob.perform_later(id, rid) }
      end
    end
  rescue ActiveRecord::RecordInvalid => e
    reload
    return fail_with_summary!(e.record.errors.full_messages.to_sentence)
  end

  safely_broadcast do
    broadcast_ui
    broadcast_clear_responses
  end
  true
end