Any future plans you can share to further evolve / train models in this size class?

It seems like a very suitable size for local AI models on reasonably high end consumer devices, given it's low active parameter count and a mixed 8bit/4bit quant would fit easily inside 64GB of memory.