llama-cpp-python

fix: auto-disable mmap when all layers offloaded to GPU (#1964)

#2147

Open