[Ready To Merge] Adds supporting of the OpenAI and AsyncOpenAI Clients

[mrDzurb]

Description

This PR introduces two new client classes—AquaOpenAI (synchronous) and AsyncAquaOpenAI (asynchronous)—that extend the official OpenAI Python SDK to support OCI-AQUA-based model deployments. The changes are aimed at making our integration with OCI seamless and future‑proof by dynamically patching request headers and normalizing URL paths based on the deployment OCID.

Out Of Scope

OpenAI recently introduced a new API, but VLLM does not support it yet. For more details, refer to this issue. The AquaOpenAIClient can be enhanced to support the new OpenAI specification once VLLM adds compatibility.

Key Features

• Common Request Patching Logic:

  • A new mixin class (AquaAIMixin) encapsulates all common logic for patching request headers and normalizing URL paths.
  • The mixin extracts the OCID from the URL and uses any segments following the OCID (skipping a leading "predict", if present) to build the "route" header dynamically.
  • The URL is normalized so that any trailing endpoint (e.g., /chat/completions, /embeddings, etc.) is replaced with /predict.
  • The mixin also sets the "enable-streaming" header based on whether the request JSON body contains "stream": true.

• Auto-Creation of HTTP Clients:

  • If no HTTP client is provided to the synchronous client, it automatically creates one using ads.aqua.get_httpx_client().
  • Similarly, the asynchronous client auto-creates an httpx.AsyncClient using ads.aqua.get_async_httpx_client().

• Support for Both Sync and Async Workflows:

  • AquaOpenAI and AsyncAquaOpenAI inherit from their respective OpenAI base classes and from the common mixin.

Testing

Sync

import ads
from ads.aqua.client.openai_client import OpenAI, AsyncOpenAI
ads.set_auth(auth="security_token")

client = OpenAI(
    base_url="https://modeldeployment.us-ashburn-1.oci.customer-oci.com/<OCID>/predict/v1",
)

response = client.chat.completions.create(
    model="odsc-llm",
    messages=[
        {
            "role": "user",
            "content": "Write a ten-sentence bedtime story about a unicorn.",
        }
    ],
    stream=True,
)

for event in response:
    print(event)

Async

import asyncio
import nest_asyncio
from ads.aqua.client.openai_client import OpenAI, AsyncOpenAI

async def test_async() -> None:
    client_async = AsyncOpenAI(
        base_url="https://modeldeployment.us-ashburn-1.oci.customer-oci.com/<OCID>/predict/v1",
    )
    response = await client_async.chat.completions.create(
        model="odsc-llm",
        messages=[{"role": "user", "content": "Tell me a long joke"}],
        stream=True
    )
    async for event in response:
        print(event)

asyncio.run(test_async())

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[github-actions]

📌 Cov diff with main:

📌 Overall coverage:

[VipulMascarenhas]

lgtm 👍

[github-actions]

📌 Cov diff with main:

📌 Overall coverage: