Version 2.1.0

- Fix CMake build arguments: CUDA etc. should now be correctly working
- Separate linux arm64 build workflow
- Update implementation to llama.cpp master