Jeseung Lee

Llamion proves you can transplant a model's brain (capabilities like long context and coding) into a new body (Llama architecture) with minimal training, opening the door to easier architecture upgrades and model merging.