Can not run with OpenAI-Compatible Provider

[BsoBird]

duckdb 1.4.1

for example
i use minimax endpoint,

Below is an executable demo:

CREATE SECRET IF NOT EXISTS (
    TYPE OPENAI,
    BASE_URL 'https://api.minimax.io/v1',
    API_KEY 'eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJHcm91cE5hbWUiOiJMaXNvZGEiLCJVc2VyTmFtZSI6Ikxpc29kYSIsIkFjY291bnQiOiIiLCJTdWJqZWN0SUQiOiIxOTgyNDU3ODkwOTgzNjUwMDk3IiwiUGhvbmUiOiIxODgyOTM0NTYyOCIsIkdyb3VwSUQiOiIxOTgyNDU3ODkwOTc5NDU1NzkzIiwiUGFnZU5hbWUiOiIiLCJNYWlsIjoiIiwiQ3JlYXRlVGltZSI6IjIwMjUtMTAtMjggMTA6Mjk6NDciLCJUb2tlblR5cGUiOjEsImlzcyI6Im1pbmltYXgifQ.tDdV-bWBRZnxzHpSlgnRuWWCgXn0WcFwAFh9_eoorM7WhHr8lWoNK4WEkvn0X7nQm434UQad-ROORC4Jje10JuotbNQkzqEMuDwnMayvwFdkK45Mtefn-USyvbzLXOP2zvo-_OTzgHtAPf-hWEtwy5_22Y1UYLqi-0fEdC3ZbEdOHjG6xb8xOC9Qqt-QaHGqa8SmpmGC2k3UkZRSM2Mq9L2N3qjbjSJw8hjEuOAd0miBP9xvDS545Dbvu56kHpUpo6BpiVvr88reD6F2MTHnLr3ptZcUHXNTlk4WVMsRiZpLkfMB2PuaTBEd09nf4Vt9yPEjnEmAg9hXYaIoargi5A'
);


CREATE MODEL(
   'Test01Model12',
   'gpt-4o', 
   'openai', 
   {"tuple_format": "json", "batch_size": 32, "model_parameters": {"temperature": 0.7}}
);


SELECT llm_complete(
    {'model_name': 'Test01Model12'},
    {'prompt': 'Talk like a duck 🦆 and explain what a database is 📚'}
);   --- return null

Currently, I have found that flock's compatibility with the APIs of most vendors that provide OpenAI-compatible interfaces is not very good.
doc:https://platform.minimax.io/docs/guides/text-generation

[queryproc]

I recommend you edit the description to remove your API Key.

[BsoBird]

@queryproc Thank you for the reminder, sir. Don't worry, this API key is free and has about a week left of the free period(until 11-07). When the time is up, I will delete this key from the backend myself.

[anasdorbani]

@BsoBird , The issue you’re experiencing is not caused by Flock. With minimax, when using a model that is not supported by the provider, or if there are other API issues, such as insufficient quota, the response may appear as null, so it's returning an error in a different format than the OpenAI API. Flock simply forwards the provider’s response, which is why you may see null completions in DuckDB queries.

Here are two examples illustrating the behavior:

1. Minimax

curl https://api.minimax.io/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <YOUR_MINIMAX_API_KEY>" \
  -d '{
    "model": "gpt-4o",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hi."}
    ],
    "max_tokens": 200
  }'

Response:

{
  "choices": null,
  "usage": null,
  "base_resp": {
    "status_code": 2013,
    "status_msg": "invalid params, fail to get model info"
  }
}

2. OpenAI

curl https://api.openai.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <YOUR_OPENAI_API_KEY>" \
  -d '{
    "model": "BAD_MODEL",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hi."}
    ],
    "max_tokens": 200
  }'

Response:

{
  "error": {
    "message": "The model `BAD_MODEL` does not exist or you do not have access to it.",
    "type": "invalid_request_error",
    "param": null,
    "code": "model_not_found"
  }
}

Notes:

• Minimax wraps errors differently than OpenAI, which can lead to choices: null.
• Insufficient quota, invalid models, or other API issues will also produce similar null results because the error format is not OpenAI-compatible.
• Make sure your Minimax settings and API key are correctly configured in your curl command first.
• To avoid unsupported model errors, refer to the supported models list: [Minimax Supported Models](https://platform.minimax.io/docs/guides/models-intro)

Once a supported model is used and your account has sufficient quota, Flock will return completions as expected.

[BsoBird]

@anasdorbani Sir, I’m sorry—I left out some important details.

When creating the model we are supposed to specify MiniMax-M2, but no matter what I query the response is always null.

To see what would happen I tried inserting a different value—gpt-4o (which Minimax doesn’t actually support)—and still got null without any error.

That’s what’s puzzling: I have no way to tell whether the value I entered is valid or not.

[anasdorbani]

@BsoBird Thanks for the update. Before digging further into the integration, could you please try calling MiniMax directly from your terminal to confirm everything is set up correctly?

Run the following command:

curl https://api.minimax.io/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <YOUR_MINIMAX_API_KEY>" \
  -d '{
    "model": "MiniMax-M2",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hi."}
    ],
    "max_tokens": 200
  }'

Just replace <YOUR_MINIMAX_API_KEY> with your actual key, run it, and let me know what the response is.

If this still returns null, then it’s very likely related to your MiniMax configuration or account permissions rather than the model field itself. Once we confirm this call works properly, we can go back and fix the code implementation.