Fined tuned GPT 4o min model not working in Chat Playground

Manish Kumar 0

Value for cogsvc-openai-model-version is invalid. | Apim-request-id: 5299a6db-9eea-41e8-9cb1-70d6f9ee07f3

^^ Getting above error when I send a picture with a text to the chat in the playground (using chat).

Found that on Aug 1st the same error was there on base GPT mini 4o and was resolved on Aug 6th.

(https://video2.skills-academy.com/en-us/answers/questions/1851591/the-image-cannot-be-recognized-gpt-4o-mini-by-and)

It was fixed by some configuration change by Microsoft. Can this be the same issue but for each fine tuned model?

So far I have tested on both Sweden and North Central US. Same error.

A quick turn-around will be helpful.

Edit: Its happening in both OAI and AI studio. Attaching HAR debughar.txt

YutongTie-MSFT 50,856 Reputation points

2024-08-07T23:03:10.8933333+00:00

Hello @Manish Kumar

Thanks for reaching out to us, is this happening in your Azure OpenAI Studio or you Azure AI Studio, could you please check on both?

Could you please collect the Browser HAR trace at the time of the issue and share it here in .txt format ?

Regards,

Yutong
Manish Kumar 0 Reputation points

2024-08-07T23:51:15.68+00:00

@YutongTie-MSFT Added the har as an edit to the question. Yes, happening on both, just checked.
Manish Kumar 0 Reputation points

2024-08-23T09:10:35.0333333+00:00

@YutongTie-MSFT Did the team revert on why fine tuned models are not working?
YutongTie-MSFT 50,856 Reputation points

2024-09-02T01:03:03.4833333+00:00

Hello Manish

Sorry for all inconveniences, this issue has been escalated to product owner.

However, we received a response from the product owner that Fine-tuned models do not currently support images during inferencing as well as training.

We see there is no clear description in the document, so we raise a document ticket to make this point clear to customers.

Also, we are looking for an ETA from product team but I don't see there is a response.

Thanks.

Regards,

Yutong
Manish Kumar 0 Reputation points

2024-09-02T17:03:53.9666667+00:00

Well, then there is no point in having fine tuning facility for gpt 4o. Earlier models support text fine tuning anyway.

It has been a total waste of time preparing data and training and then waiting for a month to find out "oh, well, we dont do that here. thanks"

Well done, Microsoft, a 3 trillion dollar company.

Share via

Fined tuned GPT 4o min model not working in Chat Playground

Your answer