Fix: "UserWarning: ... sessions are still open..." when streaming with AsyncInferenceClient

[hanouticelina]

AsyncInferenceClient creates an aiohttp.ClientSession each time it makes a request. when streaming, the session is only closed after the async generator that yields the data finishes iterating:

async for chunk in response.content:  
    yield chunk
await client.close()                   

if the user stops iteration as soon as they see the [DONE] event (see _async_stream_chat_completion_response and _format_chat_completion_stream_output), the await client.close() line is never executed. later, when the client object is garbage collected, we get this warning:

UserWarning: Deleting 'AsyncInferenceClient' client but some sessions are still open. This can happen if you've stopped streaming data from the server before the stream was complete. To close the client properly, you must call `await client.close()` or use an async context (e.g. `async with AsyncInferenceClient(): ...`.
Unclosed client session
client_session: <aiohttp.client.ClientSession object at 0x1026d6120>
Unclosed connector
connections: ['deque([(<aiohttp.client_proto.ResponseHandler object at 0x10a8a6ed0>, 91165.093372916)])']
connector: <aiohttp.connector.TCPConnector object at 0x1026d5e80>

the fix simply wraps the iteration in a try/finally block, regardless of how or when the caller stops consuming the stream, the finally clause guarantees the underlying session s closed.

---

repro:

import asyncio
import os

from huggingface_hub import AsyncInferenceClient

client = AsyncInferenceClient(
    provider="novita",
    api_key=os.environ["HF_TOKEN"],
)


async def main():
    stream = await client.chat.completions.create(
        model="Qwen/Qwen3-Coder-480B-A35B-Instruct",
        messages=[{"role": "user", "content": "What is the capital of France?"}],
        stream=True,
    )

    async for chunk in stream:
        print(chunk.choices[0].delta.content, end="")


if __name__ == "__main__":
    asyncio.run(main())

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Wauplin]

Seems like session is now correctly closed but when running your repro code on your branch, I now get:

UserWarning: Deleting 'AsyncInferenceClient' client but some sessions are still open. This can happen if you've stopped streaming data from the server before the stream was complete. To close the client properly, you must call `await client.close()` or use an async context (e.g. `async with AsyncInferenceClient(): ...`.

This warning comes from the AsyncInferenceClient directly so I guess we need to update the check there

[Wauplin]

All good, the warning I reported is just because we need a await client.close() in the repro script but that's it. So all good to merge!