Class: CompletionKit::Run
- Inherits:
-
ApplicationRecord
- Object
- ActiveRecord::Base
- ApplicationRecord
- CompletionKit::Run
- Includes:
- 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
Instance Method Summary collapse
- #as_json(options = {}) ⇒ Object
- #avg_score ⇒ Object
- #generate_responses! ⇒ Object
- #judge_configured? ⇒ Boolean
- #mark_completed! ⇒ Object
- #metric_averages ⇒ Object
- #missing_dataset_variables ⇒ Object
- #outstanding_work_zero? ⇒ Boolean
- #progress_snapshot ⇒ Object
- #replace_metrics!(metric_ids) ⇒ Object
- #start! ⇒ Object
Instance Method Details
#as_json(options = {}) ⇒ Object
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 |
# File 'app/models/completion_kit/run.rb', line 165 def as_json( = {}) snap = progress_snapshot { 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: 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: , metric_ids: metric_ids } end |
#avg_score ⇒ Object
66 67 68 69 70 71 72 |
# File 'app/models/completion_kit/run.rb', line 66 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
122 123 124 |
# File 'app/models/completion_kit/run.rb', line 122 def generate_responses! start! end |
#judge_configured? ⇒ Boolean
54 55 56 |
# File 'app/models/completion_kit/run.rb', line 54 def judge_configured? judge_model.present? && metrics.any? && ApiConfig.valid_for_model?(judge_model) end |
#mark_completed! ⇒ Object
30 31 32 33 |
# File 'app/models/completion_kit/run.rb', line 30 def mark_completed! update!(status: "completed") broadcast_ui end |
#metric_averages ⇒ Object
74 75 76 77 78 79 80 |
# File 'app/models/completion_kit/run.rb', line 74 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 |
#missing_dataset_variables ⇒ Object
21 22 23 24 25 26 27 28 |
# File 'app/models/completion_kit/run.rb', line 21 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
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
# File 'app/models/completion_kit/run.rb', line 35 def outstanding_work_zero? return false if responses.where.not(status: Response::TERMINAL_STATUSES).exists? metric_ids = metrics.pluck(:id) 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: Review::TERMINAL_STATUSES ).count terminal_review_count >= expected_reviews end |
#progress_snapshot ⇒ Object
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 |
# File 'app/models/completion_kit/run.rb', line 126 def progress_snapshot generated_done = responses.where(status: "succeeded").count generated_failed = responses.where(status: "failed").count generated_total = progress_total metric_count = metrics.count 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) metric_ids = metrics.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 |
#replace_metrics!(metric_ids) ⇒ Object
58 59 60 61 62 63 64 |
# File 'app/models/completion_kit/run.rb', line 58 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 |
#start! ⇒ Object
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 |
# File 'app/models/completion_kit/run.rb', line 82 def start! rows = if dataset CsvProcessor.process_self(self) else [{}] end return fail_with_summary!("Dataset has no rows") if rows.empty? 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 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 response = responses.create!( status: "pending", row_index: index, input_data: input, expected_output: row["expected_output"] ) GenerateRowJob.perform_later(id, response.id) end end broadcast_ui broadcast_clear_responses true end |