Yeah it will be interesting to see if we ever settle on standard sizes here. My preference would be: - 3B for CPU inference or running on edge devices. - 20-3

Yeah it will be interesting to see if we ever settle on standard sizes here. My preference would be: - 3B for CPU inference or running on edge devices. - 20-30B for maximizing single consumer GPU potential. - 70B+ for those who can afford it. 7-9B never felt like an ideal size.