Function calculateUsage

app/src/utils/recordRequest.ts:106–167

({
  inputPayload,
  completion,
  fineTune,
}: {
  inputPayload: z.infer<typeof chatCompletionInput>;
  completion: ChatCompletion;
  fineTune?: FineTune;
})

};

export const calculateUsage = ({
  inputPayload,
  completion,
  fineTune,
}: {
  inputPayload: z.infer<typeof chatCompletionInput>;
  completion: ChatCompletion;
  fineTune?: FineTune;
}): CalculatedUsage | undefined => {
  let provider: BaseModel["provider"] | undefined = undefined;
  let baseModel: BaseModel["baseModel"] | undefined = undefined;

  if (fineTune) {
    const typedFT = typedFineTune(fineTune);
    provider = typedFT.provider;
    baseModel = typedFT.baseModel;
  } else if (inputPayload.model.startsWith("ft:")) {
    provider = "openai";
    baseModel = "gpt-3.5-turbo-1106";
  }

  // Return usage for non-fine-tuned OpenAI models
  if (!provider || !baseModel) {
    return openAIModelProvider.getUsage(inputPayload, completion) ?? undefined;
  }

  let inputTokens;
  let outputTokens;

  if (completion.usage) {
    inputTokens = completion.usage.prompt_tokens;
    outputTokens = completion.usage.completion_tokens;
  } else if (baseModel === "gpt-3.5-turbo-0613" || baseModel === "gpt-3.5-turbo-1106") {
    inputTokens = countOpenAIChatTokens(baseModel, inputPayload.messages);
    outputTokens = completion.choices[0]?.message
      ? countOpenAIChatTokens(baseModel, [completion.choices[0].message])
      : 0;
  } else {
    inputTokens = countLlamaInputTokens(inputPayload);
    outputTokens = completion.choices[0]?.message
      ? countLlamaOutputTokens(completion.choices[0]?.message)
      : 0;
  }

  const { cost, inputCost, outputCost } = calculateCost(
    {
      provider,
      baseModel,
    } as BaseModel,
    0,
    inputTokens,
    outputTokens,
  );

  return {
    inputTokens,
    outputTokens,
    cost,