Support Union Types in Definition Struct for JSON Schema

[vish0l]

Is your feature request related to a problem? Please describe.
The Definition struct currently does not support multiple data types, limiting the flexibility needed for certain JSON schema definitions. According to OpenAI’s documentation, it is possible to emulate optional parameters by using a union type with multiple data types, which is not currently supported by the existing struct.

    "name": "get_weather",
    "description": "Fetches the weather in the given location",
    "strict": true,
    "parameters": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The location to get the weather for"
            },
            "unit": {
                // highlight-start
                "type": ["string", "null"],
                // highlight-end
                "description": "The unit to return the temperature in",
                "enum": ["F", "C"]
            }
        },
        "additionalProperties": false,
        "required": [
            "location", "unit"
        ]
    }
}

In this example, we can only support a single string type with the existing model.

**Describe the solution you'd like**
Update the Type field in the Definition struct to  allow for multiple data types. This enhancement will enable developers to define union types, aligning the library with OpenAI’s guidelines for structured outputs.

**Describe alternatives you've considered**
Create a new type which support single and multiple type both and write custom marshal and unmarshal to handle that.

**Additional context**
https://platform.openai.com/docs/guides/structured-outputs/supported-schemas

[samarthpatil00]

Hi @vish0l #831 will solve this as well. Since it will allow you to pass the schema directly.
Alternate solution for your problem is using structured output via function calls, there it can accept raw json schema in the parameter. So you can do something like :

schema := `{
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The location to get the weather for"
            },
            "unit": {
                // highlight-start
                "type": ["string", "null"],
                // highlight-end
                "description": "The unit to return the temperature in",
                "enum": ["F", "C"]
            }
        },
        "additionalProperties": false,
        "required": [
            "location", "unit"
        ]
}`
tool := openai.Tool{
    Type: openai.ToolTypeFunction,
    Function: &openai.FunctionDefinition{
  	Name:       "get_weather",
	Strict:     true,
	Parameters: json.RawMessage(schema),
    },
}

Hope this helps

[vish0l]

Thank you for your response, @samarthpatil00. In my use case, I need a structured output, but I won't be using any external functions or tools.

[samarthpatil00]

@vish0l you can still use tools to get the argument it passes to function and use them as the structured output

resp, err := c.Client.CreateChatCompletion(
  context.Background(),
  openai.ChatCompletionRequest{
	  Model:            string(c.Model),
	  Messages:         messages,
	  Temperature:      temperature,
	  Tools:            []openai.Tool{tool},
  },
)
op := resp.Choices[0].Message.ToolCalls[0].Function.Arguments

[vish0l]

what's the point of doing that when straight forward structured output functionality is available 🙂