OpenAI.instance.chat.create() accepts the wrong type for responseFormat

[seung-cha]

current definiiton (5.1.0) accepts Map<String, String> for responseFormat. This is incompatible when you are using json_schema.

MRE:

final responseFormatDef = {
      "type": "json_schema",
      "json_schema": jsonEncode({
        "name": "story_generation",
        "schema": {
          "type": "object",
          "properties": {
            "title": {
              "type": "string",
              "description": "The title of the story."
            },
            "genre": {
              "type": "string",
              "description":
                  "The genre of the story, as provided by the user. If user asks for any, make one yourself."
            },
            "language": {
              "type": "string",
              "description":
                  "The language in which the story is written, as requested by the user."
            },
            "content": {
              "type": "string",
              "description": "The main body of the story."
            }
          },
          "required": ["title", "genre", "language", "content"],
          "additionalProperties": false
        },
        "strict": true
      })
    };

    OpenAIChatCompletionModel completion = await OpenAI.instance.chat.create(
        model: 'gpt-4o',
        responseFormat: responseFormatDef,
        messages: [sysMsg, usrMsg]);

Here, I used jsonEncode to convert the json_schema field into a string. This consequently causes .create() to fail.
Once I change the type definition to Map<String, dynamic> and get rid of jsonEncode, it works fine (changing value type to dynamic makes it more versatile too).

[seung-cha]

PR opened: #205

[seung-cha]

dup: #193