Comment by yjtpesesu2

How does this differ from anything llama.cpp offers, regarding offloading layers? The repo consistently refers to "DDR4". Is there a reason DDR5 won't work with this?