Skip to content

AzureOpenAIModel implementation #17

Closed
Closed

Description

@bxyu-nvidia

A lot of enterprise users that are on ChatGPT does not use public server, but the Azure OpenAI private deployment. This endpoint requires an "default_query" and "api_version" parameters on the client init. Since it is almost certain that there will be subtle but critical API differences between the normal OpenAI endpoints and Azure OpenAI endpoints, we should separate this into a separate Responses API model called AzureOpenAIModel.

Example env.yaml

# for azure openai only
policy_api_version: preview

Example responses_api_models/openai_model/configs/openai_model.yaml

openai_model:
  responses_api_models:
    openai_model:
      entrypoint: app.py
      openai_base_url: ${policy_base_url}
      openai_api_key: ${policy_api_key}
      openai_model: ${policy_model_name}
      # for azure openai only
      default_query:
        api-version: ${policy_api_version}

Thank you to @xinyu-dev for the raise!

Activity

  1. xinyu-dev commented on Sep 8, 2025

    @xinyu-dev

    thanks very much Brian for your help

  2. bxyu-nvidia commented on Sep 8, 2025

    @bxyu-nvidia
    ContributorAuthor

    A few additional notes from @xinyu-dev:

    • If you're using chat completion API on Azure OpenAI, you might need to use the AuzureOpenAI object.
    • If you're using response API on Azure OpenAI, you can use the OpenAI object.
    • Not all Azure OpenAi deployment might support response API. But all Azure deployment all support chat completion API at least.
  3. added this to the v0.1.0 milestone on Oct 8, 2025
  4. bxyu-nvidia commented on Oct 19, 2025

    @bxyu-nvidia
    ContributorAuthor

    Just need docs for this

  5. removed this from the v0.1.0 milestone on Oct 23, 2025
  6. added
    modelsinfra for different models and inference providers; middleware to standardize to OAI responses
    and removed
    env-infraInfra related to creating new environments
    on Oct 23, 2025
  7. cwing-nvidia commented on Jan 29, 2026

    @cwing-nvidia
    Contributor

    @bxyu-nvidia can we split docs into a separate ticket and close this if the model server code is complete

  8. cwing-nvidia commented on Feb 20, 2026

    @cwing-nvidia
    Contributor

    tracking docs in this ticket #665

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

modelsinfra for different models and inference providers; middleware to standardize to OAI responses

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions