Change default model to Ministral 3 Instruct 8B

[janepie]

to hopefully get some nicer output

needs a llama-cpp-python update or an alternative

[marcelklehr]

This will improve tool calling accuracy. @edward-ly vLLM sadly does not seem to support gguf, which I'd really like to keep supported to avoid a breaking change for our users.

[marcelklehr]

Apparently, there is a maintained fork of llama-cpp-python: https://github.com/JamePeng/llama-cpp-python Let's try this iwth the current models and with Ministral 3 Instruct 8B

[marcelklehr]

abetlen llama-cpp-python is maintained again, it seems. 🎉