Comment by monocasa

Not immediately, but it's not a much larger amount of work for llama than a new foundational model which typically has a tweaked compute graph.