@azure/openai - Set response_format for GPT-4 Turbo

[joshje]

Now that GPT-4 Turbo is available, it would be great to be able to specify the response_format={"type": "json_object"} when making calls to @azure/openai client.getChatCompletions.

[deyaaeldeen]

Thanks for opening this report! This work is in progress and I will share once we make a release.

[YvesZelros]

@deyaaeldeen Same for the new seed parameter ;-)

[tverdohleb]

Hey @deyaaeldeen! Do you have any ETA on this matter?

[deyaaeldeen]

Tentative release date is 12/7. I'll share here once the release happens.

[SumeetGohil]

Hoping this gets released today.

[joshje]

This is great to see! Right now we're blocked from using GPT-4 Turbo until this feature is available. Any chance it'll be available today @deyaaeldeen?

[SumeetGohil]

Our major application stopped working due GPT-4 is not giving json response as it used to give post 30th, Nov.

[deyaaeldeen]

Thanks for your patience. Please checkout @azure/openai@1.0.0-beta.8 with support for tools, vision, DALL-E v3, and enhanced content filtering!

[deyaaeldeen]

Also, to answer the original question, you can do this with v1.0.0-beta.8, given a capable deployed model:

client.getChatCompletions(
    deploymentName,
    [
      {
        role: "user",
        content:
          "Answer the following question in JSON format: What are the capital cities in Africa?",
      },
    ],
    { responseFormat: { type: "json_object" } }
);

[github-actions]

Hi @joshje. Thank you for opening this issue and giving us the opportunity to assist. We believe that this has been addressed. If you feel that further discussion is needed, please add a comment with the text "/unresolve" to remove the "issue-addressed" label and continue the conversation.

[joshje]

Thanks @deyaaeldeen, I'm very happy to report that this has done the trick, and we can now use gpt-4 turbo, with JSON responses!