Create Distill model based on Mistral Small 3

[elter-tef]

Mistral Small 3 model is optimized for running on local hardware and high output speeds. The model is comparable in quality to the Qwen 32b, but faster due to optimizations, which seems important for using thinking models for local use.