Modify response_format to take in JSON object instead of string

[frankaloia]

Description

• We already handle response_format input for openai. It currently assumed that the response_format would be a type, which is incorrect https://platform.openai.com/docs/guides/structured-outputs?api-mode=chat&example=structured-data#how-to-use
• Changing the response_format config element to pass in the json object as opposed to extracting it as a json value
• Created a dedicated type in case we wanted to more specifically type the expected content of the object

Tests

Ran against local endpoint:

Request:

{
      "specification_hash": "973a5fa849b584b384cdd87b70bd7e03ebf12227edf0ff50e9e7f08fa7a27c86",
      "config": {
        "MODEL": {
          "provider_id": "openai",
          "model_id": "gpt-4.5-preview-2025-02-27",
          "function_call": null,
          "use_cache": false,
          "response_format": {
            "type": "json_schema",
            "json_schema": {
              "name": "ABeautifulTest",
              "strict": true,
              "schema": {
              "type": "object",
              "properties": {
                "Name": {
                  "type": "string"
                },
                "Description": {
                    "type": "string"
                }
              },
              "required": ["Name", "Description"],
              "additionalProperties": false
              }
            }
          }
        }
      },
      "blocking": true,
      "inputs": [{"hello":"world"}]
    }

Output:

 {"run":{"run_id":"730bf651dfc1b78f54223084c300dc54339490bd1ed4fbb816b70b06b4d68560","created":1742578696878,"run_type":"deploy","config":{"blocks":{"MODEL":{"provider_id":"openai","model_id":"gpt-4.5-preview-2025-02-27","function_call":null,"use_cache":false,"response_format":{"type":"json_schema","json_schema":{"name":"ABeautifulTest","strict":true,"schema":{"type":"object","properties":{"Name":{"type":"string"},"Description":{"type":"string"}},"required":["Name","Description"],"additionalProperties":false}}}}}},"status":{"run":"succeeded","blocks":[{"block_type":"input","name":"INPUT","status":"succeeded","success_count":1,"error_count":0},{"block_type":"chat","name":"MODEL","status":"succeeded","success_count":1,"error_count":0}]},"traces":[[["input","INPUT"],[[{"value":{"hello":"world"},"error":null,"meta":null}]]],[["chat","MODEL"],[[{"value":{"message":{"content":"{\"Name\":\"Hello\",\"Description\":\"Hi! How can I assist you today?\"}","role":"assistant"}},"error":null,"meta":{"logs":[],"token_usage":{"prompt_tokens":52,"completion_tokens":18},"provider_request_id":"req_b75b23937184f73151094d25df79c036"}}]]]],"specification_hash":"973a5fa849b584b384cdd87b70bd7e03ebf12227edf0ff50e9e7f08fa7a27c86","results":[[{"value":{"message":{"content":"{\"Name\":\"Hello\",\"Description\":\"Hi! How can I assist you today?\"}","role":"assistant"}},"error":null,"meta":{"logs":[],"token_usage":{"prompt_tokens":52,"completion_tokens":18},"provider_request_id":"req_b75b23937184f73151094d25df79c036"}}]]}}

Risk

N/A - I don't believe this is directly exposed to customers yet via the console. It should fix an existing issue.

Deploy Plan

Deploy via Github

[flvndvd]

If this used to work, does it mean that response_format was a string? Or was it simply never meeting this condition?