GitHub discussion is not used anymore, please use the forum for discussion.

[youkaichao]

Please enjoy the discussion forum at https://discuss.vllm.ai . See you there!

[byjlw]

why the change?

[hmellor]

GitHub discussions are not often looked at by users and maintainers.

The hope is that a community will build on the forum, similar to the PyTorch forum, where users can go to ask questions and seek advice about building with vLLM. Then, the GitHub issues will become more focused on building vLLM.

If this goes well, then users will be able to ask and find answers more easily via the forum, and the devs will be able to focus more easily on building vLLM via GitHub. I hope this explanation helps!

[gaby]

How is a forum even better? That requires an extra account.

I'm a maintainer in another project, you get notifications when discussions are created.

[limingde01]

请教下，nv 的orin 平台能vllm支持吗？

[vadimkantorov]

One problem with the forum is that RunLLM responds there without a way to switch it off:
https://discuss.vllm.ai/t/vllm-serve-to-use-all-gpus-of-a-single-node-multi-use/1264/4

It does not even listen when explicitly tell it to not respond. And there is no way to downvote its imperfect responses. At least it would be good to have some way to mention in the post, so that RunLLM would abstain from posting in certain threads.

I'd like to have a place for technical discussion with actual humans (as I'm a human myself).

[raghu-microsoft]

Hi Team,
In my current setup, I'm using the latest version of vLLM (0.10.0) along with aiohttp 3.12.14. However, when running the model using python -m llm.optimized.inference.api_server, it fails with the following error: AttributeError: 'Namespace' object has no attribute 'prompt_adapters'. with the previous version of vllm its working fine