Llm

Mellum 2.1 12B, Flint 40B and Arete 27B: five small coding fixes

I tested three self-hosted models on five small coding tasks from my own projects. Arete finished all five; Mellum was usually faster when it succeeded; Flint sometimes wrote correct code but failed to stop.

ai engineering