runFunction throws "'content' is a required property - 'messages.1'" error when using Azure openAI integration

[darkneo14]

Confirm this is a Node library issue and not an underlying OpenAI API issue

• This is an issue with the Node library

Describe the bug

When you use runFunctions function with Azure open AI, it throws the "'content' is a required property - 'messages.1'" error with 400 error code.

const runner = openai.beta.chat.completions .runFunctions({ messages: [ { role: "user", content: what is sqauare root of 4? } ], functions: [ { "name": "calculateSquareRoot", "function": calculateSquareRoot, "description": "Calculate sqaure root of a number", "parameters": { type: 'object', properties: { number: { type: 'integer' }, } }, }, { "name": "calculateLogBase2", "function": calculateLogBase2, description: "Calculate log base 2 of a number", parameters: { type: 'object', properties: { number: { type: 'integer' }, } }, } ] })

After debugging in the library, I found the issue. The second gpt call that happens after the function call is missing the content field in the messages object -

{ model: 'gpt-3.5-turbo', messages: [ { role: 'user', content: 'what is sqauare root of 4?' }, { role: 'assistant', function_call: [Object] }, { role: 'function', name: 'calculateSquareRoot', content: '2' } ], functions: [ { name: 'calculateSquareRoot', parameters: [Object], description: 'Calculate sqaure root of a number' }, { name: 'calculateLogBase2', parameters: [Object], description: 'Calculate log base 2 of a number' } ], function_call: 'auto' }

Here you can see that messages[1].content is not present. In case of openAI call, this is null where it runs smoothly.
A simple solve is to add - params.messages = params.messages.map((message) => { message.content = message.content || null; return message; }).

To Reproduce

Code to reproduce -

use azure configuration for openAI initialisation.

const runner = openai.beta.chat.completions .runFunctions({ model: "gpt-3.5-turbo", messages: [ { role: "user", content: what is sqauare root of 4? } ], functions: [ { "name": "calculateSquareRoot", "function": calculateSquareRoot, "description": "Calculate sqaure root of a number", "parameters": { type: 'object', properties: { number: { type: 'integer' }, } }, }, { "name": "calculateLogBase2", "function": calculateLogBase2, description: "Calculate log base 2 of a number", parameters: { type: 'object', properties: { number: { type: 'integer' }, } }, } ] }) console.log(runner) const finalContent = await runner.finalContent();

Code snippets

No response

OS

macOs

Node version

node v16.14.2

Library version

^4.0.0

[rattrayalex]

Thanks for reporting! Can you share the traceback / error message you get?

[darkneo14]

error stack -

/node_modules/openai/error.js:43
            return new BadRequestError(status, error, message, headers);
                   ^

BadRequestError: 400 'content' is a required property - 'messages.1'
    at APIError.generate (/Users/prashant.gupta/Desktop/Ixigo/ixigo-bot-new/node_modules/openai/error.js:43:20)
    at OpenAI.makeStatusError (/Users/prashant.gupta/Desktop/Ixigo/ixigo-bot-new/node_modules/openai/core.js:251:33)
    at OpenAI.makeRequest (/Users/prashant.gupta/Desktop/Ixigo/ixigo-bot-new/node_modules/openai/core.js:290:30)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async ChatCompletionRunner._createChatCompletion (/Users/prashant.gupta/Desktop/Ixigo/ixigo-bot-new/node_modules/openai/lib/AbstractChatCompletionRunner.js:305:32)
    at async ChatCompletionRunner._runFunctions (/Users/prashant.gupta/Desktop/Ixigo/ixigo-bot-new/node_modules/openai/lib/AbstractChatCompletionRunner.js:333:36)

[rattrayalex]

Thanks! Could you share the code that integrates with this, including where you define the messages you send?

[darkneo14]

Sure... Here is the function I am testing this with -

async function testFunctionAzure() {
  const openai = new OpenAI({
    'apiKey': apiKey,
    baseURL: `https://oai.hconeai.com/openai/deployments/${model}`,
    defaultHeaders: {
      "Helicone-OpenAI-API-Base": `https://${resource}.openai.azure.com`,
      'api-key': `${apiKey}`,
    },
    defaultQuery: { "api-version": `${apiVersion}` },
  });

    const runner = openai.beta.chat.completions
    .runFunctions({
          messages: [
            {
              role: "user",
              content: `what is sqauare root of 4?`
            }
          ],
          functions: [
            {
              "name": "calculateSquareRoot",
              "function": calculateSquareRoot,
              "description": "Calculate sqaure root of a number",
              "parameters": { 
                type: 'object', properties: 
                {
                  number: { type: 'integer' },
                } 
              },
            }, 
            {
              "name": "calculateLogBase2",
              "function": calculateLogBase2,
              description: "Calculate log base 2 of a number",
              parameters: { 
                type: 'object', 
                properties: {
                    number: { type: 'integer' },
                } 
              },
            }
          ]
        })
        console.log(runner)
        const finalContent = await runner.finalContent();
        console.log(finalContent)

        function calculateSquareRoot(params) {
            params = JSON.parse(params)
            console.log(params)
            let response = Math.sqrt(params.number);
            console.log("Response ------",response)
            return response
        }

        function calculateLogBase2(params) {
            return Math.log2(params.number);
        }
}

[rattrayalex]

The fix for this will be released in v4.20.2, and should also be fixed in the next Azure OpenAI API version (cc @johanste).

Thank you @darkneo14 for the report & details!