Module: LlmCostTracker::Integrations::Openai

Extended by:
Base
Defined in:
lib/llm_cost_tracker/integrations/openai.rb

Overview

rubocop:disable Metrics/ModuleLength

Defined Under Namespace

Modules: ChatCompletionsPatch, PatchBuilder, ResponsesPatch

Constant Summary collapse

INPUT_DETAIL_KEYS = 
%i[input_tokens_details input_token_details prompt_tokens_details].freeze
OUTPUT_DETAIL_KEYS = 
%i[output_tokens_details output_token_details completion_tokens_details].freeze
EmbeddingsPatch = 
PatchBuilder.build(record_method: :record_response, methods: %i[create])
ImagesPatch = 
PatchBuilder.build(record_method: :record_image, methods: %i[generate edit create_variation])
TranscriptionsPatch = 
PatchBuilder.build(record_method: :record_transcription, methods: %i[create])
TranslationsPatch = 
PatchBuilder.build(record_method: :record_transcription, methods: %i[create])
SpeechPatch = 
PatchBuilder.build(record_method: :record_speech, methods: %i[create])
ModerationsPatch = 
PatchBuilder.build(record_method: :record_moderation, methods: %i[create])
StreamingImagesPatch = 
PatchBuilder.build_stream(methods: %i[generate_stream_raw edit_stream_raw])
StreamingTranscriptionsPatch = 
PatchBuilder.build_stream(methods: %i[create_streaming])

Constants included from Base

Base::Result

Class Method Summary collapse

Methods included from Base

active?, enforce_budget!, install, minimum_version, normalize_sdk_args, object_dig, object_value, patch_target, patch_targets, record_safely, request_params, status, stream_collector, stream_pricing_mode, track_stream, version_constant

Class Method Details

.audio_input_tokens(usage) ⇒ Object 



327
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 327

def audio_input_tokens(usage)      = detail(usage, INPUT_DETAIL_KEYS, :audio_tokens)

.audio_output_tokens(usage) ⇒ Object 



328
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 328

def audio_output_tokens(usage)     = detail(usage, OUTPUT_DETAIL_KEYS, :audio_tokens)

.auxiliary_patch_targets ⇒ Object 



76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 76

def auxiliary_patch_targets
  [
    patch_target("OpenAI::Resources::Embeddings",
                 with: EmbeddingsPatch, methods: %i[create], optional: true),
    patch_target("OpenAI::Resources::Images",
                 with: ImagesPatch, methods: %i[generate edit create_variation], optional: true),
    patch_target("OpenAI::Resources::Images",
                 with: StreamingImagesPatch,
                 methods: %i[generate_stream_raw edit_stream_raw],
                 optional: true, skip_when_methods_missing: true),
    patch_target("OpenAI::Resources::Audio::Transcriptions",
                 with: TranscriptionsPatch, methods: %i[create], optional: true),
    patch_target("OpenAI::Resources::Audio::Transcriptions",
                 with: StreamingTranscriptionsPatch,
                 methods: %i[create_streaming],
                 optional: true, skip_when_methods_missing: true),
    patch_target("OpenAI::Resources::Audio::Translations",
                 with: TranslationsPatch, methods: %i[create], optional: true),
    patch_target("OpenAI::Resources::Audio::Speech",
                 with: SpeechPatch, methods: %i[create], optional: true),
    patch_target("OpenAI::Resources::Moderations",
                 with: ModerationsPatch, methods: %i[create], optional: true)
  ]
end

.cache_read_input_tokens(usage) ⇒ Object 



325
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 325

def cache_read_input_tokens(usage) = detail(usage, INPUT_DETAIL_KEYS, :cached_tokens)

.chat_completions_search_item(response, model: nil) ⇒ Object 



253
254
255
256
257
258
259
260
261
262
263
264
265
266
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 253

def chat_completions_search_item(response, model: nil)
  choices = object_value(response, :choices)
  return nil unless choices.respond_to?(:any?)

  provider_field = if choices.any? { |choice| choice_used_url_citation?(choice) }
                     LlmCostTracker::Providers::Openai::ServiceCharges::CHAT_COMPLETIONS_ANNOTATION_PROVIDER_FIELD
                   elsif LlmCostTracker::Providers::Openai::ModelFamilies.chat_completions_search?(model)
                     LlmCostTracker::Providers::Openai::ServiceCharges::CHAT_COMPLETIONS_SEARCH_MODEL_PROVIDER_FIELD
                   end
  return nil unless provider_field

  { "type" => "web_search_call", "id" => object_value(response, :id),
    "action" => { "type" => "search" }, "provider_field" => provider_field }
end

.choice_used_url_citation?(choice) ⇒ Boolean

Returns:

  • (Boolean) 


268
269
270
271
272
273
274
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 268

def choice_used_url_citation?(choice)
  message = object_value(choice, :message)
  annotations = message && object_value(message, :annotations)
  return false unless annotations.respond_to?(:any?)

  annotations.any? { |annotation| object_value(annotation, :type).to_s == "url_citation" }
end

.client_host_for(resource) ⇒ Object 



45
46
47
48
49
50
51
52
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 45

def client_host_for(resource)
  client = resource.instance_variable_get(:@client)
  return nil unless client

  URI.parse(client.base_url.to_s).host
rescue URI::InvalidURIError
  nil
end

.detail(usage, containers, key) ⇒ Object 



333
334
335
336
337
338
339
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 333

def detail(usage, containers, key)
  containers.each do |container|
    value = object_dig(usage, container, key)
    return value.to_i if value
  end
  0
end

.hidden_output_tokens(usage) ⇒ Object 



326
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 326

def hidden_output_tokens(usage)    = detail(usage, OUTPUT_DETAIL_KEYS, :reasoning_tokens)

.image_input_tokens(usage) ⇒ Object 



329
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 329

def image_input_tokens(usage)      = detail(usage, INPUT_DETAIL_KEYS, :image_tokens)

.image_output_tokens(usage) ⇒ Object 



330
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 330

def image_output_tokens(usage)     = detail(usage, OUTPUT_DETAIL_KEYS, :image_tokens)

.integration_name ⇒ Object 



15
16
17
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 15

def integration_name
  :openai
end

.minimum_version ⇒ Object 



58
59
60
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 58

def minimum_version
  "0.59.0"
end

.normalize_output_action(action) ⇒ Object 



289
290
291
292
293
294
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 289

def normalize_output_action(action)
  return nil if action.nil?
  return action if action.is_a?(Hash)

  { "type" => object_value(action, :type)&.to_s }
end

.normalize_output_item(item) ⇒ Object 



276
277
278
279
280
281
282
283
284
285
286
287
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 276

def normalize_output_item(item)
  return item if item.is_a?(Hash)
  return nil if item.nil?

  {
    "type" => object_value(item, :type)&.to_s,
    "id" => object_value(item, :id),
    "status" => object_value(item, :status),
    "container_id" => object_value(item, :container_id),
    "action" => normalize_output_action(object_value(item, :action))
  }
end

.patch_targets ⇒ Object 



66
67
68
69
70
71
72
73
74
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 66

def patch_targets
  [
    patch_target("OpenAI::Resources::Responses",
                 with: ResponsesPatch, methods: %i[create stream stream_raw retrieve_streaming]),
    patch_target("OpenAI::Resources::Chat::Completions",
                 with: ChatCompletionsPatch, methods: %i[create stream stream_raw]),
    *auxiliary_patch_targets
  ]
end

.provider_for_host(host) ⇒ Object 



54
55
56
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 54

def provider_for_host(host)
  LlmCostTracker::Providers::Azure::Hosts.openai?(host) ? "azure_openai" : "openai"
end

.record_image(response, request:, latency_ms:, host: nil) ⇒ Object 



133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 133

def record_image(response, request:, latency_ms:, host: nil)
  usage = object_value(response, :usage)
  raw_input = usage ? object_value(usage, :input_tokens).to_i : 0
  raw_output = usage ? object_value(usage, :output_tokens).to_i : 0
  image_input = image_input_tokens(usage).to_i
  cache_read = cache_read_input_tokens(usage).to_i
  text_input = [raw_input - image_input - cache_read, 0].max
  image_output, text_output = split_image_output(usage, raw_output)
  record_passthrough(
    model: request[:model],
    response: response,
    latency_ms: latency_ms,
    host: host,
    input_tokens: text_input,
    image_input_tokens: image_input,
    output_tokens: text_output,
    image_output_tokens: image_output,
    cache_read_input_tokens: cache_read
  )
end

.record_moderation(response, request:, latency_ms:, host: nil) ⇒ Object 



211
212
213
214
215
216
217
218
219
220
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 211

def record_moderation(response, request:, latency_ms:, host: nil)
  record_passthrough(
    model: object_value(response, :model) || request[:model],
    response: response,
    latency_ms: latency_ms,
    host: host,
    input_tokens: 0,
    output_tokens: 0
  )
end

.record_passthrough(model:, response:, latency_ms:, host: nil, service_line_items: [], **token_attributes) ⇒ Object 



222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 222

def record_passthrough(model:, response:, latency_ms:, host: nil, service_line_items: [], **token_attributes)
  return unless active?

  record_safely do
    LlmCostTracker::Tracker.record(
      event: Event.build(
        provider: provider_for_host(host),
        model: model,
        token_usage: TokenUsage.build(**token_attributes),
        usage_source: :sdk_response,
        provider_response_id: response && object_value(response, :id),
        service_line_items: service_line_items
      ),
      latency_ms: latency_ms
    )
  end
end

.record_response(response, request:, latency_ms:, host: nil) ⇒ Object 



101
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
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 101

def record_response(response, request:, latency_ms:, host: nil)
  return unless active?

  record_safely do
    usage = object_value(response, :usage)
    next unless usage

    input_tokens = object_value(usage, :input_tokens, :prompt_tokens)
    output_tokens = object_value(usage, :output_tokens, :completion_tokens)
    next if input_tokens.nil? && output_tokens.nil?

    cache_read = cache_read_input_tokens(usage)
    model = object_value(response, :model) || request[:model]
    LlmCostTracker::Tracker.record(
      event: Event.build(
        provider: provider_for_host(host),
        model: model,
        pricing_mode: LlmCostTracker::Parsers::OpenaiUsage.combined_pricing_mode(
          host: host,
          model: model,
          service_tier: object_value(response, :service_tier) || request[:service_tier]
        ),
        token_usage: token_usage(usage:, input_tokens:, output_tokens:, cache_read:, model: model),
        usage_source: :sdk_response,
        provider_response_id: object_value(response, :id),
        service_line_items: service_line_items_from(response, request: request)
      ),
      latency_ms: latency_ms
    )
  end
end

.record_speech(_response, request:, latency_ms:, host: nil) ⇒ Object 



185
186
187
188
189
190
191
192
193
194
195
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 185

def record_speech(_response, request:, latency_ms:, host: nil)
  record_passthrough(
    model: request[:model],
    response: nil,
    latency_ms: latency_ms,
    host: host,
    input_tokens: 0,
    output_tokens: 0,
    service_line_items: speech_line_items(request)
  )
end

.record_transcription(response, request:, latency_ms:, host: nil) ⇒ Object 



163
164
165
166
167
168
169
170
171
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 163

def record_transcription(response, request:, latency_ms:, host: nil)
  record_passthrough(
    model: request[:model],
    response: response,
    latency_ms: latency_ms,
    host: host,
    **transcription_token_attributes(object_value(response, :usage))
  )
end

.regular_input_tokens(input_tokens, cache_read, audio_input, image_input) ⇒ Object 



341
342
343
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 341

def regular_input_tokens(input_tokens, cache_read, audio_input, image_input)
  [input_tokens.to_i - cache_read - audio_input - image_input, 0].max
end

.service_line_items_from(response, request: nil) ⇒ Object 



240
241
242
243
244
245
246
247
248
249
250
251
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 240

def service_line_items_from(response, request: nil)
  model = object_value(response, :model) || request&.dig(:model)
  output = object_value(response, :output)
  output_items = output.respond_to?(:each) ? output.map { |item| normalize_output_item(item) }.compact : []
  chat_search = output_items.empty? ? chat_completions_search_item(response, model: model) : nil
  output_items << chat_search if chat_search
  return [] if output_items.empty?

  LlmCostTracker::Providers::Openai::ServiceCharges.line_items_from_output(
    output_items, request: request, model: model
  )
end

.speech_line_items(request) ⇒ Object 



197
198
199
200
201
202
203
204
205
206
207
208
209
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 197

def speech_line_items(request)
  input = request[:input]
  return [] unless input.is_a?(String)
  return [] unless LlmCostTracker::Providers::Openai::ModelFamilies.character_billed_tts?(request[:model])

  [LlmCostTracker::Billing::LineItem.build(
    component_key: :text_to_speech_character,
    quantity: input.length,
    cost_status: LlmCostTracker::Billing::CostStatus::UNKNOWN,
    pricing_basis: :provider_usage,
    provider_field: "request.input"
  )]
end

.split_image_output(usage, raw_output) ⇒ Object 



154
155
156
157
158
159
160
161
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 154

def split_image_output(usage, raw_output)
  image_tokens = image_output_tokens(usage).to_i
  text_tokens = text_output_tokens(usage).to_i
  return [raw_output, 0] if image_tokens.zero? && text_tokens.zero?

  text_tokens = [raw_output - image_tokens, 0].max if text_tokens.zero?
  [image_tokens, text_tokens]
end

.split_responses_image_output(output_tokens:, image_output_details:, text_output_details:, audio_output:, default_to_image: false) ⇒ Object 



345
346
347
348
349
350
351
352
353
354
355
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 345

def split_responses_image_output(output_tokens:, image_output_details:, text_output_details:, audio_output:,
                                 default_to_image: false)
  if image_output_details.zero? && text_output_details.zero?
    remainder = [output_tokens - audio_output, 0].max
    return default_to_image ? [remainder, 0] : [0, remainder]
  end

  text_output = text_output_details
  text_output = [output_tokens - image_output_details - audio_output, 0].max if text_output.zero?
  [image_output_details, text_output]
end

.stream_collector(request, host: nil) ⇒ Object 



27
28
29
30
31
32
33
34
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 27

def stream_collector(request, host: nil)
  LlmCostTracker::Capture::StreamCollector.new(
    provider: provider_for_host(host),
    model: request[:model],
    pricing_mode: stream_pricing_mode(request, host: host),
    request: request
  )
end

.stream_pricing_mode(request, host: nil) ⇒ Object 



19
20
21
22
23
24
25
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 19

def stream_pricing_mode(request, host: nil)
  LlmCostTracker::Parsers::OpenaiUsage.combined_pricing_mode(
    host: host,
    model: (request || {})[:model],
    service_tier: (request || {})[:service_tier]
  )
end

.text_output_tokens(usage) ⇒ Object 



331
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 331

def text_output_tokens(usage)      = detail(usage, OUTPUT_DETAIL_KEYS, :text_tokens)

.token_usage(usage:, input_tokens:, output_tokens:, cache_read:, model: nil) ⇒ Object 



296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 296

def token_usage(usage:, input_tokens:, output_tokens:, cache_read:, model: nil)
  audio_input = audio_input_tokens(usage)
  audio_output = audio_output_tokens(usage)
  image_input = image_input_tokens(usage)
  image_output_details = image_output_tokens(usage)
  text_output_details = text_output_tokens(usage)
  image_output, regular_output = split_responses_image_output(
    output_tokens: output_tokens.to_i,
    image_output_details: image_output_details,
    text_output_details: text_output_details,
    audio_output: audio_output,
    default_to_image: LlmCostTracker::Providers::Openai::ModelFamilies.image_output?(model)
  )

  TokenUsage.build(
    input_tokens: regular_input_tokens(input_tokens, cache_read, audio_input, image_input),
    output_tokens: regular_output,
    cache_read_input_tokens: cache_read,
    audio_input_tokens: audio_input,
    audio_output_tokens: audio_output,
    image_input_tokens: image_input,
    image_output_tokens: image_output,
    hidden_output_tokens: hidden_output_tokens(usage)
  )
end

.transcription_token_attributes(usage) ⇒ Object 



173
174
175
176
177
178
179
180
181
182
183
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 173

def transcription_token_attributes(usage)
  return { input_tokens: 0, output_tokens: 0 } unless usage && object_value(usage, :type).to_s == "tokens"

  raw_input = object_value(usage, :input_tokens).to_i
  audio_input = object_dig(usage, :input_token_details, :audio_tokens).to_i
  {
    input_tokens: [raw_input - audio_input, 0].max,
    audio_input_tokens: audio_input,
    output_tokens: object_value(usage, :output_tokens).to_i
  }
end

.version_constant ⇒ Object 



62
63
64
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 62

def version_constant
  "OpenAI::VERSION"
end

.wrap_stream_call(args, kwargs, resource) ⇒ Object 



36
37
38
39
40
41
42
43
 
# File 'lib/llm_cost_tracker/integrations/openai.rb', line 36

def wrap_stream_call(args, kwargs, resource)
  request = request_params(args, kwargs)
  enforce_budget!(request: request)
  host = client_host_for(resource)
  collector = stream_collector(request, host: host)
  stream = yield(normalize_sdk_args(args, kwargs), collector)
  track_stream(stream, collector: collector)
end