Is 'functions' parameter not supported?

[chadananda]

Describe the bug

I'm trying to use the functions parameter to help force JSON response. Is this not supported?

To Reproduce

Call createChatCompletion with 'functions' array.

Code snippets

try {
  response = await openai.createChatCompletion({
    model, messages, functions 
  });
} catch (error) {
  if (error.response) {
    console.log(error.response.status);
    console.log(error.response.data);
  } else {
    console.log(error.message);
  }
}

Results in the error:

    {
      error: {
        message: 'Unrecognized request argument supplied: functions',
        type: 'invalid_request_error',
        param: null,
        code: null
      }
    }

### OS

macOS

### Node version

18.7.0

### Library version

3.3.0

[felipebutcher]

I am also having this error.

[killthebuddh4]

@felipebutcher @chadananda I'm not affiliated with this project, so take this with a grain of salt, but you need to pass the correct "model" argument. For example, "gpt-3.5-turbo" doesn't support functions, but "gpt-3.5-turbo-0613" does.

[rob-gordon]

I was able to get this to work using gpt-3.5-turbo-0613. But I found another requirement that may be mentioned in the OpenAI docs but I didn't see it. The JSON schema passed must be an object at the root.

This won't work:

const schema = {
  type: "array",
  items: {
    type: "object",
    properties: {
      title: { type: "string" },
      description: { type: "string" },
    },
    required: ["title", "description"],
  },
};

But this will work:

const schema = {
  type: "object",
  properties: {
    nodes: {
      type: "array",
      items: {
        type: "object",
        properties: {
          title: { type: "string" },
          description: { type: "string" },
        },
        required: ["title", "description"],
      },
    },
  },
  required: ["nodes"],
};

[chadananda]

Ooops. This is a server response due to the fact that function call capability is presently supported by only two models. When I switched to gpt-3.5-turbo-0613 it worked fine.

[etiennea]

Could we have a complete example?

[jcharlesworthuk]

@etiennea Here you go, this is the example from their announcement page in Typescript...

import { Configuration, OpenAIApi } from "openai";

const openai = new OpenAIApi(new Configuration({
        apiKey: env.OPENAI_KEY // <-- YOUR KEY
}));

const gptResponse = await openai.createChatCompletion({
    model: "gpt-3.5-turbo-0613",
    messages: [
        {
            role: "user",
            content: "What is the weather like in Boston?"
        }
    ],
    functions: [
        {
          name: "get_current_weather",
          description: "Get the current weather in a given location",
          parameters: {
            type: "object",
            properties: {
              location: {
                type: "string",
                description: "The city and state, e.g. San Francisco, CA"
              },
              unit: {
                type: "string",
                enum: ["celsius", "fahrenheit"]
              }
            },
            required: ["location"]
          }
        }
      ]
});

// finish_reason should be "function_call"
console.log(gptResponse.data.choices[0].finish_reason)

Note the model version as others have said

[felipebutcher]

I just tested and looks like it's already available on model gpt-4-0613 as well.

[andresgutgon]

It's working for me when the function has parameters. But I'm trying to send the function without parameters and is not detecting it

user_prompt: "Call Santa please"

Function

{
  name: 'call_santa_claus',
  description: "This function calls Santa Claus, Don't worry we have the number"
}

Also tried with parameters: null

{
  name: 'call_santa_claus',
  description: "This function calls Santa Claus, Don't worry we have the number",
  parameters: null
}

The function call_santa_claus is not called. OpenAI does not tell me to call it. Any idea?

[SinanGabel]

You do not need to have an actual function i.e. "function name() { return ... }".

However, you do need to provide all the other parameters, thus "pretend" that there is an actual function: thus still reference a name of a function even though it is just a function name.

On Wed, 28 Jun 2023 at 12:19, Andrés @.***> wrote:

It's working for me when the function has parameters. But I'm trying to send the function without parameters and is not detecting it

user_prompt: "Call Santa please"

Function

{ name: 'call_santa_claus', description: "This function calls Santa Claus, Don't worry we have the number" }

Also tried with parameters: null

{ name: 'call_santa_claus', description: "This function calls Santa Claus, Don't worry we have the number", parameters: null }

The function call_santa_claus is not called. OpenAI does not tell me to call it. Any idea?

— Reply to this email directly, view it on GitHub https://github.com/openai/openai-node/issues/181#issuecomment-1611146721, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABKTZUX44TFQG23L2R2LO63XNQAJRANCNFSM6AAAAAAZJJZUYQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[andresgutgon]

I solved my case like this

{
  name: 'call_santa_claus',
  description: "This function calls Santa Claus, Don't worry we have the number",
  parameters: { 'type': 'object', properties: {} }
}

You have to send parameters even when you don't have any. That worked for me

[remixtedi]

Do you know when this will be fixed on Azure OpenAI services/models? Looks like all 0613 models have this issue and return the same error message.

[rattrayalex]

@remixtedi can you provide sample code for repro? Are you hitting /engines/gpt-3.5-turbo-0613 or equivalent?

[remixtedi]

@rattrayalex
Using OpenAI Azure service and Betalgo.OpenAI package.

Here is a sample code that can be used to test:

using OpenAI;
using OpenAI.Managers;
using OpenAI.ObjectModels.RequestModels;

var openAIClient = new OpenAIService(new OpenAiOptions()
{
    ApiKey = "YOUR API KEY",
    DeploymentId = "YOUR DEPLOYMENT ID",
    ResourceName = "YOUR RESOURCE NAME",
    ProviderType = ProviderType.Azure,
    ApiVersion = "2023-06-01-preview",
    DefaultModelId = "gpt-35-turbo-0613"
});

var func = new FunctionDefinitionBuilder("generate_brand_color", "Generates a brand color based on the brand name");
func.AddParameter("HexCode", "string", "Hex code of the brand color");

var chatCompletionCreateRequest = new ChatCompletionCreateRequest
{
    Messages = new ChatMessage[]
    {
        ChatMessage.FromSystem("You are helping a customer to generate brand color"),
        ChatMessage.FromUser("I need a brand color for my brand, brand name is Shambala and it is a clothing brand")
    },
    Model = "gpt-35-turbo-0613",
    Temperature = 0.9f,
    MaxTokens = 150,
    TopP = 1,
    FrequencyPenalty = 0,
    PresencePenalty = 0,
    FunctionCall = new Dictionary<string, string> { { "name", "generate_brand_color" } },
    Functions = new List<FunctionDefinition> { func.Build() }
};

var chatCompletion = await openAIClient.ChatCompletion.CreateCompletion(chatCompletionCreateRequest);
Console.WriteLine(chatCompletion.Choices.First().Message);

[athyuttamre]

@remixtidi Azure OpenAI does not yet support functions, see the blog post here: https://techcommunity.microsoft.com/t5/ai-cognitive-services-blog/introducing-new-and-updated-models-to-azure-openai-service/ba-p/3860351

Function calling will be coming soon and will only be available for the new model versions announced today.