|
| 1 | +# frozen_string_literal: true |
| 2 | + |
| 3 | +require 'ldclient-rb' |
| 4 | + |
| 5 | +module LaunchDarkly |
| 6 | + module Server |
| 7 | + module AI |
| 8 | + # |
| 9 | + # Tracks token usage for AI operations. |
| 10 | + # |
| 11 | + class TokenUsage |
| 12 | + attr_reader :total, :input, :output |
| 13 | + |
| 14 | + # |
| 15 | + # @param total [Integer] Total number of tokens used. |
| 16 | + # @param input [Integer] Number of tokens in the prompt. |
| 17 | + # @param output [Integer] Number of tokens in the completion. |
| 18 | + # |
| 19 | + def initialize(total: nil, input: nil, output: nil) |
| 20 | + @total = total |
| 21 | + @input = input |
| 22 | + @output = output |
| 23 | + end |
| 24 | + end |
| 25 | + |
| 26 | + # |
| 27 | + # Summary of metrics which have been tracked. |
| 28 | + # |
| 29 | + class MetricSummary |
| 30 | + attr_accessor :duration, :success, :feedback, :usage, :time_to_first_token |
| 31 | + |
| 32 | + def initialize |
| 33 | + @duration = nil |
| 34 | + @success = nil |
| 35 | + @feedback = nil |
| 36 | + @usage = nil |
| 37 | + @time_to_first_token = nil |
| 38 | + end |
| 39 | + end |
| 40 | + |
| 41 | + # |
| 42 | + # The AIConfigTracker class is used to track AI configuration usage. |
| 43 | + # |
| 44 | + class AIConfigTracker |
| 45 | + attr_reader :ld_client, :config_key, :context, :variation_key, :version, :summary |
| 46 | + |
| 47 | + def initialize(ld_client:, variation_key:, config_key:, version:, context:) |
| 48 | + @ld_client = ld_client |
| 49 | + @variation_key = variation_key |
| 50 | + @config_key = config_key |
| 51 | + @version = version |
| 52 | + @context = context |
| 53 | + @summary = MetricSummary.new |
| 54 | + end |
| 55 | + |
| 56 | + # |
| 57 | + # Track the duration of an AI operation |
| 58 | + # |
| 59 | + # @param duration [Integer] The duration in milliseconds |
| 60 | + # |
| 61 | + def track_duration(duration) |
| 62 | + @summary.duration = duration |
| 63 | + @ld_client.track( |
| 64 | + '$ld:ai:duration:total', |
| 65 | + @context, |
| 66 | + flag_data, |
| 67 | + duration |
| 68 | + ) |
| 69 | + end |
| 70 | + |
| 71 | + # |
| 72 | + # Track the duration of a block of code |
| 73 | + # |
| 74 | + # @yield The block to measure |
| 75 | + # @return The result of the block |
| 76 | + # |
| 77 | + def track_duration_of(&block) |
| 78 | + start_time = Time.now |
| 79 | + yield |
| 80 | + ensure |
| 81 | + duration = ((Time.now - start_time) * 1000).to_i |
| 82 | + track_duration(duration) |
| 83 | + end |
| 84 | + |
| 85 | + # |
| 86 | + # Track time to first token |
| 87 | + # |
| 88 | + # @param duration [Integer] The duration in milliseconds |
| 89 | + # |
| 90 | + def track_time_to_first_token(time_to_first_token) |
| 91 | + @summary.time_to_first_token = time_to_first_token |
| 92 | + @ld_client.track( |
| 93 | + '$ld:ai:tokens:ttf', |
| 94 | + @context, |
| 95 | + flag_data, |
| 96 | + time_to_first_token |
| 97 | + ) |
| 98 | + end |
| 99 | + |
| 100 | + # |
| 101 | + # Track user feedback |
| 102 | + # |
| 103 | + # @param kind [Symbol] The kind of feedback (:positive or :negative) |
| 104 | + # |
| 105 | + def track_feedback(kind:) |
| 106 | + @summary.feedback = kind |
| 107 | + event_name = kind == :positive ? '$ld:ai:feedback:user:positive' : '$ld:ai:feedback:user:negative' |
| 108 | + @ld_client.track( |
| 109 | + event_name, |
| 110 | + @context, |
| 111 | + flag_data, |
| 112 | + 1 |
| 113 | + ) |
| 114 | + end |
| 115 | + |
| 116 | + # |
| 117 | + # Track a successful AI generation |
| 118 | + # |
| 119 | + def track_success |
| 120 | + @summary.success = true |
| 121 | + @ld_client.track( |
| 122 | + '$ld:ai:generation', |
| 123 | + @context, |
| 124 | + flag_data, |
| 125 | + 1 |
| 126 | + ) |
| 127 | + @ld_client.track( |
| 128 | + '$ld:ai:generation:success', |
| 129 | + @context, |
| 130 | + flag_data, |
| 131 | + 1 |
| 132 | + ) |
| 133 | + end |
| 134 | + |
| 135 | + # |
| 136 | + # Track an error in AI generation |
| 137 | + # |
| 138 | + def track_error |
| 139 | + @summary.success = false |
| 140 | + @ld_client.track( |
| 141 | + '$ld:ai:generation', |
| 142 | + @context, |
| 143 | + flag_data, |
| 144 | + 1 |
| 145 | + ) |
| 146 | + @ld_client.track( |
| 147 | + '$ld:ai:generation:error', |
| 148 | + @context, |
| 149 | + flag_data, |
| 150 | + 1 |
| 151 | + ) |
| 152 | + end |
| 153 | + |
| 154 | + # |
| 155 | + # Track token usage |
| 156 | + # |
| 157 | + # @param token_usage [TokenUsage] An object containing token usage details |
| 158 | + # |
| 159 | + def track_tokens(token_usage) |
| 160 | + @summary.usage = token_usage |
| 161 | + if token_usage.total.positive? |
| 162 | + @ld_client.track( |
| 163 | + '$ld:ai:tokens:total', |
| 164 | + @context, |
| 165 | + flag_data, |
| 166 | + token_usage.total |
| 167 | + ) |
| 168 | + end |
| 169 | + if token_usage.input.positive? |
| 170 | + @ld_client.track( |
| 171 | + '$ld:ai:tokens:input', |
| 172 | + @context, |
| 173 | + flag_data, |
| 174 | + token_usage.input |
| 175 | + ) |
| 176 | + end |
| 177 | + return unless token_usage.output.positive? |
| 178 | + |
| 179 | + @ld_client.track( |
| 180 | + '$ld:ai:tokens:output', |
| 181 | + @context, |
| 182 | + flag_data, |
| 183 | + token_usage.output |
| 184 | + ) |
| 185 | + end |
| 186 | + |
| 187 | + # |
| 188 | + # Track OpenAI-specific operations. |
| 189 | + # This method tracks the duration, token usage, and success/error status. |
| 190 | + # If the provided block raises, this method will also raise. |
| 191 | + # A failed operation will not have any token usage data. |
| 192 | + # |
| 193 | + # @yield The block to track. |
| 194 | + # @return The result of the tracked block. |
| 195 | + # |
| 196 | + def track_openai_metrics(&block) |
| 197 | + result = track_duration_of(&block) |
| 198 | + track_success |
| 199 | + track_tokens(openai_to_token_usage(result[:usage])) if result[:usage] |
| 200 | + result |
| 201 | + rescue StandardError |
| 202 | + track_error |
| 203 | + raise |
| 204 | + end |
| 205 | + |
| 206 | + # |
| 207 | + # Track AWS Bedrock conversation operations. |
| 208 | + # This method tracks the duration, token usage, and success/error status. |
| 209 | + # |
| 210 | + # @yield The block to track. |
| 211 | + # @return [Hash] The original response hash. |
| 212 | + # |
| 213 | + def track_bedrock_converse_metrics(&block) |
| 214 | + result = track_duration_of(&block) |
| 215 | + track_success |
| 216 | + track_tokens(bedrock_to_token_usage(result[:usage])) if result[:usage] |
| 217 | + result |
| 218 | + rescue StandardError |
| 219 | + track_error |
| 220 | + raise |
| 221 | + end |
| 222 | + |
| 223 | + private def flag_data |
| 224 | + { variationKey: @variation_key, configKey: @config_key, version: @version } |
| 225 | + end |
| 226 | + |
| 227 | + private def openai_to_token_usage(usage) |
| 228 | + TokenUsage.new( |
| 229 | + total: usage[:total_tokens] || usage['total_tokens'], |
| 230 | + input: usage[:prompt_tokens] || usage['prompt_tokens'], |
| 231 | + output: usage[:completion_tokens] || usage['completion_tokens'] |
| 232 | + ) |
| 233 | + end |
| 234 | + |
| 235 | + private def bedrock_to_token_usage(usage) |
| 236 | + TokenUsage.new( |
| 237 | + total: usage[:total_tokens] || usage['total_tokens'], |
| 238 | + input: usage[:input_tokens] || usage['input_tokens'], |
| 239 | + output: usage[:output_tokens] || usage['output_tokens'] |
| 240 | + ) |
| 241 | + end |
| 242 | + end |
| 243 | + end |
| 244 | + end |
| 245 | +end |
0 commit comments