Function getLLMResponse

lib/queryLLM.ts:31–84

(
  prompt: string | ChatGPTMessage[],
  params: ChatGPTParams = {},
  model: string,
)

    };

export async function getLLMResponse(
  prompt: string | ChatGPTMessage[],
  params: ChatGPTParams = {},
  model: string,
): Promise<string> {
  if (typeof prompt === "string" && model !== "gpt-3.5-turbo-instruct")
    throw new Error(
      `String prompts only supported with model gpt-3.5-turbo-instruct. You have selected model: ${model}`,
    );

  const { url, options } =
    typeof prompt === "string"
      ? getOAIRequestCompletion(prompt, params, model)
      : getLLMRequestChat(prompt, params, model);

  const response = await Promise.race([
    fetch(url, options),
    (async () => {
      // Time out after 90s
      await new Promise((resolve) => setTimeout(resolve, 90000));
      return new Response(
        JSON.stringify({ error: { message: "Timed out!" } }),
        {
          status: 500,
        },
      );
    })(),
  ]);
  const responseJson: ChatGPTResponse | { error: OpenAIError } =
    await response.json();
  if (response.status >= 300) {
    console.log(
      "Response from LLM: ",
      JSON.stringify(responseJson, undefined, 2),
    );
  }

  if (response.status === 429) {
    // Throwing an error triggers exponential backoff retry
    throw Error(
      // TODO: Check whether retry_after exists
      `OpenAI API rate limit exceeded. Full error: ${JSON.stringify(
        responseJson,
      )}`,
    );
  }
  if ("error" in responseJson) {
    throw Error(
      `Error from LLM provider: ${JSON.stringify(responseJson.error)}`,
    );
  }

  return removeEmptyCharacters(textFromResponse(responseJson)).trim();
}

export function textFromResponse(
  response: