>>106013426
I downloaded a quant from LMStudio and one from Bartowski and they both have a built-in chat template. It appears to be getting applied, but I'm not getting reasoning parsing at the backend level.

https://huggingface.co/lmstudio-community/Magistral-Small-2507-GGUF/tree/main
https://huggingface.co/bartowski/mistralai_Magistral-Small-2507-GGUF/tree/main

"prompt":"<s>[SYSTEM_PROMPT]First draft your thinking process (inner monologue) until you arrive at a response. Format your response using Markdown, and use LaTeX for any mathematical equations. Write both your thoughts and the response in the same language as the input.\n\nYour thinking process must follow the template below:[THINK]Your thoughts or/and draft, like working through an exercise on scratch paper. Be as casual and as long as you want until you are confident to generate the response. Use the same language as the input.[/THINK]Here, provide a self-contained response.[/SYSTEM_PROMPT][INST]Test, test test. How are you?[/INST]
Parsed message: {"role":"assistant","content":"The user is asking how I am. Since I'm an AI assistant, I don't have feelings or a physical state, but I can respond in a friendly manner. The user seems to be testing the interaction, so I should respond positively and confirm that I'm functioning well.I'm just a computer program, so I don't have feelings, but I'm here and ready to help you with any questions or tasks you have! How can I assist you today?"}