Azure OpenAI and custom models

Question

Azure OpenAI and custom models

Roberto Guardamagna 0

Hi,

I'm new with Azure OpenAI and I'm trying to "build" a custom model based on "Chat" logic.
When I create a new model, there are only 3 base models available: ada(1), babbage(1), curie(1).
I selected the "ada(1)", selected the JSONL with a few lines just to test and deployed it.
Then I developed a small C# program using "Betalgo OpenAI" nuget package.
Everything works if I use a deploy based on "gpt-35-turbo" (for example), if I select the deployed custom model I get an error

The chatCompletion operation does not work with the specified model, ada.ft-7cef33ac464e49ba87be059053ce3ac1-custom-fields-test. Please choose different model and try again. You can learn more about which models can be used with each operation here: https://go.microsoft.com/fwlink/?linkid=2197993.

so...which are the models that support "ChatCompletion"?
Babbage(1)?
Curie(1)?

Thank you, regards.

Roberto

2 answers

Your answer

Answer 1

Charlie Wei 3,335

Hello Roberto Guardamagna,

The available fine-tuning models and their associated regions can be found in this document.

enter image description here

Best regards,
Charlie

If you find my response helpful, please consider accepting this answer and voting 'yes' to support the community. Thank you!

BobG 41 Reputation points

2024-03-22T09:55:59.4666667+00:00

So it seems the only available model for East Europe (my area) is gpt-35-turbo (0301).
Could I use the same system of training also with that model?
If yes, how?
I can just select "ada, babbage, curie" for a custom model.
Thank you, regards.

Roberto
Charlie Wei 3,335 Reputation points

2024-03-22T14:07:57.1166667+00:00

Hello Roberto Guardamagna,

The model gpt-35-turbo-0301 does not support fine-tuning. For a list of models that do support fine-tuning, please refer to this link.

Additionally, the models you mentioned, "Ada", "Babbage", and "Curie", have all been deprecated, and it is recommended to cease their use.

Although not in your region, I would suggest selecting new fine-tuning models from either North Central US or Sweden Central.

Best regards,
Charlie

If you find my response helpful, please consider accepting this answer and voting yes to support the community. Thank you!
Charlie Wei 3,335 Reputation points

2024-03-25T12:33:04.19+00:00

@Roberto Guardamagna, just checking in to see if above information was helpful. Please let us know if you would like further assistance.
Roberto Guardamagna 0 Reputation points

2024-03-25T13:17:38.8233333+00:00

Hi Charlie,

does the "area" depends on the MS subscription?
Should I change the country setting to use the other version of GPT?
Thank you
Charlie Wei 3,335 Reputation points

2024-03-25T15:19:08.7666667+00:00

@Roberto Guardamagna, you can create multiple Azure OpenAI resources located in different regions under an MS subscription; there should be no issue with this. You may want to try this out first.
Charlie Wei 3,335 Reputation points

2024-03-28T04:16:59.2766667+00:00

@Roberto Guardamagna, just checking in to see if above information was helpful. Please let us know if you would like further assistance.

Answer 2

Hello @Roberto Guardamagna ,Thanks for using Microsoft Q&A Platform.

Regarding your question about the models that support "ChatCompletion"?

The Chat Completion API is a new dedicated API for interacting with the GPT-35-Turbo and GPT-4 models. This API is the preferred method for accessing these models. It is also the only way to access the new GPT-4 models.

So, try with these models. Additionally, as mentioned by @Charlie Wei , please check the finetuning model availability and its region. Try to create resource in these supported regions.

I hope this helps.

Regards,

Vasavi

Share via

Azure OpenAI and custom models

2 answers

Your answer