Issue with Qwen

[6RakaDigiBJ]

Hello Michael,

I tried using your extension with a Qwen model hosted on Scaleway, but I get the following error:

API error (400): {"error":{"message":"System message must be at the beginning.","type":"BadRequestError","param":null,"code":400}}

The same configuration works correctly with a Llama model.

It seems that Qwen may reject requests containing more than one system message, or a system message that is not the first message in the conversation.

Are you aware of this issue? Do you think the extension could merge all system instructions into a single initial system message for better compatibility with Qwen?

Thanks for your work on the extension.

[lxman]

You had it exactly right - Qwen's chat template wants a single system message at position 0, and the extension was inserting per-turn context notes as extra system messages mid-conversation (Llama's template doesn't care, which is why that combination worked). Fixed in #3: all system content is now coalesced into one leading system message at send time, for every backend.

The fix is in v1.0.5, just published - it should show up as an update in Visual Studio shortly, or you can grab the VSIX directly from https://github.com/lxman/LocalModelIntegrator/releases/tag/v1.0.5

Thanks for the clear report. Let me know if Scaleway still gives you trouble on the new version.

[6RakaDigiBJ]

Thanks Michael, I’ve installed v1.0.5 and tested it with Qwen on Scaleway. Everything is working correctly now, including agentic mode.

Thanks again for the very quick fix and the clear explanation!