Loading LoRAs should disable mmap (LoRA + mmap causes segmentation fault)

[horenbergerb]

In the issue that implemented LoRAs in llama.cpp they say "Using --lora disables mmap since the models have to be modified anyway."

It seems like trying to load a LoRA in llama-cpp-python with mmap enabled causes a segmentation fault. Probably this package should copy the default behavior of llama.cpp and disable mmap when loading LoRAs.

[abetlen]

Good catch, will update.