ADD libllama.so target for llama-cpp-python

[bhubbb]

I could get llama-cpp-python working but only when I built libllama.so with make and then replaced the copy of libllama.so that came with llama-cpp-python .

I wanted to add this here to help with the llama-cpp-python project.

[SagsMug]

Did you try mkdir lib && cd lib && cmake -DBUILD_SHARED_LIBS=1 ..?

[bhubbb]

Not manually, but I believe this is what llama-cpp-python and it would just through segmentation faults for me until built by make.

I have not had a lot of luck with cmake builds on my system.

[pjlegato]

Context: Using make sets -march=native in the Makefile, which automatically chooses the best instruction set extensions available for the current CPU.

The CMakeLists supplied with llama.cpp manually specifies certain instruction set extensions, which then fails with "illegal instruction" errors unless the current CPU happens to support those pre-chosen instructions.