Self-host using official web ui

[NekoNyanners]

Are we able to self-host qwen using the official frontend at https://chat.qwenlm.ai/ ?
I saw in the readme that you can use text-generation-webui or llamafile but there is no mention of the official frontend which apparently uses Open WebUI. Are there plans to release the code for the official chat website so it can be self-hosted too?

[jklj077]

May I know what's the features unique to Qwen Chat that would beneift self-hosting? I think the functionality of Open WebUI is quite complete for self-hosting open-weight models.

also @dxdong1234 on this topic.

[NekoNyanners]

Benefits of self-hosting would be always having full feature parity (esp. future, without having to wait for third-party UI's to also add them) and simply UX preferences (it's nice and clean). Also insights in the workings, etc. It's also kind of weird documenting another third-party UI while the official one uses a completely different one imo.

Counter question: Are there any reasons to not open source that part too?
Out of the entire infrastructure, this part should be the least problematic, right?

[jklj077]

There are many things to consider, and many of them are quite different from what's needed for self-hosting. This makes Qwen Chat's architecture very specific to how we've set it up based on the Alibaba Cloud infrastructure right now. It's not as easy as just open-sourcing the code and expecting it to work smoothly for self-hosting. Making it work for self-hosting would take a lot more effort and resources. Right now, we don't have the capacity to maintain another WebUI. Even if we did have the time and resources, we'd want to make sure it adds real value and benefits the community in the long run.

At a higher level, Qwen works best when we collaborate with the community. We don't open-source first-party tools just to have them labeled as "official". In general, we step in only when there's a critial gap in the ecosystem that needs filling for Qwen. And even then, our preference is to contribute to and collaborate with existing, well-maintained frameworks, rather than starting from scratch. This way, we can work together to build solutions that truly make a difference.

For the benefits you mentioned:

• always having full feature parity with Qwen: I believe Open WebUI is still ahead of Qwen Chat in terms of feature completeness. In terms of new model support, we work with main inference frameworks to ensure that comes in time.
• simply UX preferences (it's nice and clean): the UX should be very similar to Open WebUI. there is a custom theme feat request in Open WebUI. please upvote that!
• also insights in the workings: I failed to see how that's possible with a UI framework.
• also kind of weird documenting another third-party UI while the official one uses a completely different one imo: see above.