A bit shame start
Hi everyone,
First, thank you for releasing Magistral—one of the few European LLMs on par with DeepSeek (CN) and the large US models (OpenAI, Microsoft). Impressive work!
Issue
When I began testing, my very first prompt was simply “Hi”. The model responded with:
Summary:
…
Final Response:
Hello! How can I assist you today?
\boxed{Hello!}
The injected “Summary” section and the LaTeX \boxed{} markup feel out of place and let us like we don't put attention to detail.
Thanks again for the great model; looking forward to future updates!
Best regards,
I don't think this has anything to do with Bartowski's quantization of the model, so it's probably better to comment on MistralAI's original model if you have feedback for it.
As for solving your issue, try rewriting the system prompt if you don't like the way it formats its answers. The default system prompt is tailored for solving a task using reasoning, not for casual conversation.