Confusion about web UI and existing configuration

[handcraftedbits]

Running server --config ... has a few issues for me, which may just be that I'm missing something:

• All available models are presented. I would've expected it to be limited to just the ones in my configuration (per the documentation, it says that unless you add --ui-management you can't do model switching, so in my mind that would imply that the UI options should be limited to what's in the configuration).
• My server preloads models, and when I try to do TTS with a model I know is loaded, I just get a "Model did not load" error.
• I tried checking out the Models tab and in the console I see:

  {
    "error": {
        "message": "UI path inspection is disabled",
        "type": "forbidden"
    }
  }

  It looks like --ui-management controls that option, fair enough, but it's not really documented. In the UI it's also not really obvious that you would have to turn that option on in order to "unlock" all the functionality.

It seems to me that the UI is geared towards fetching and exploring different models. Which is really nice! But it would be nice if there could be a UI that's restricted to exactly what's in your configuration -- no downloads, no loading (well, apart from lazy loading what's in the config), no distractions. Just a playground for your exact audio.cpp's configuration (including enumerating all the defined voices for a model, that would be awesome).

Thanks again for this great project!

[0xShug0]

@handcraftedbits Thanks for your suggestions! I'll look into those gaps.

[awatuna]

What is the correct file system layout for the models dir?

I started the server with --ui, set the models path in the models tab, but the studio tab did not detect the bf16 gguf?

I have tried putting all the gguf in the models dir directly, as well as mirroring the huggingface layout, neither worked.

models
├───ACE-Step1.5-GGUF
│ └───base
│ ace-step-1.5-base-bf16.gguf
│
├───Fish-Audio-S2-Pro-GGUF
│ fish-audio-s2-pro-bf16.gguf

[0xShug0]

@awatuna The correct layout is https://huggingface.co/audio-cpp/audio.cpp-gguf. I think your layout is correct. This is actually UI display bugs. It always shows Q8 even when the model is 16-bif. If you query the server.

curl -sS http://127.0.0.1:8080/v1/models

It shows something like which shows 16-bit is loaded.

id: ace-step
family: ace_step
task: gen
mode: offline
path: /path/to/ACE-Step1.5-GGUF/turbo/ace-step-1.5-turbo-bf16.gguf

For ACE-Step, there is a separate issue. It currently defaults to the turbo model only and don't allow the selection of the base model.

We will fix the issues.

[0xShug0]

Pushed some changes to fix the UI issues and the UI for the config mode. @handcraftedbits @awatuna Please try the latest commit.

@handcraftedbits Model did not load was a generic fallback error that hid the real load failure. I pushed a fix so the UI now reports the underlying error directly. Could you try again and let me know if it still fails? If it does, please share the exact server command and the relevant model configuration you used.

The config UI is somehow misleading because it can't really change the server config. This and other issues will be addressed in the follow-up PRs.

[handcraftedbits]

@0xShug0 works great, thanks! Only one small thing I noticed and this is probably related to the other person's comment. I have the 1.7B Qwen3 ASR loaded but it displays as 0.6B, which is not a big deal.

[0xShug0]

@handcraftedbits I fixed the issue in #248. But I’ll use this PR to track and fix multiple issues, so it may not be merged immediately.