model name is missing from the request [0.23.0 • llama.cpp b7772]

[mwmcode]

Hello,

I have configured Firefox AI (browser.ml.chat.provider) to use http://localhost:2276 which was working fine until the recent update (v 0.23.0 • llama.cpp b7772). I started getting this error model name is missing from the request

Even when I access the UI directly (localhost:2676), it asks me to pick a model even though there is one loaded already (and is not idle since I had just loaded it).

[erusev]

Hello, and thanks for reaching out.

In https://github.com/ggml-org/LlamaBarn/releases/tag/0.22.0 we migrated to a design where the server stays running in the background and loads models automatically when they are needed. But for this to work your requests need to say which model you want to use (similar to how the OpenAI API works).

Usually when you connect an app to LlamaBarn it makes a request to /models to see what models are available and let you chose what model you want to use.

Can you try setting it up from scratch and see if this helps.

[mwmcode]

Hi @erusev
I set it up on a different machine (from scratch) and I'm getting the same error.

I think Firefox's implementation is very simple.. they just open a chat window and post messages.

I couldn't find where the server part (or more precisely the web interface part is -- not familiar with swift :\ ) to see if we can add an option to the URL i.e. http://localhost:2276?defaultModel=Qwen3 then have the UI component to read/set it (on window load) if no other option is selected

if you're open to such a change

I haven't used chat gpt or claude interfaces much but I think/remember they default to a model when first loaded

[black-snow]

So the actual thing here would be the ability to specify a default/fallback mode, I guess?

[erusev]

This still isn't possible today -- in router mode every request must name a model, and the server has no default-model setting. There's an open llama.cpp PR that adds one (ggml-org/llama.cpp#19855); once it lands we can build on it here.