Ollama refusing tool role, which always worked fine in llama.cpp with qwen.

What is the issue?

My "tool" role always worked fine in llama.cpp but ollama breaks my client refusing my "tool" answers. It doesnt make sense to send a tool answer as "user" or "assistent", as a tool is neither user nor assistent! And I guess parsing as "system" would also not work as there can only be ONE system prompt at the beginning of the prompt? I found this related bug but wasnt able to reopen it: https://github.com/ollama/ollama/issues/6322

Relevant log output

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

No response