Use compiler to determine best optimizations for platform

Author: abetlen

.github/workflows/test.yaml

@@ -73,7 +73,7 @@ jobs:
       - name: Install dependencies
         run: |
           python3 -m pip install --upgrade pip
-          CMAKE_ARGS="-DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_AVX512=OFF -DLLAMA_AVX512_VBMI=OFF -DLLAMA_AVX512_VNNI=OFF -DLLAMA_FMA=OFF -DLLAMA_F16C=OFF -DLLAMA_ACCELERATE=OFF -DLLAMA_METAL=OFF" python3 -m pip install .[all] --verbose --config-settings=cmake.verbose=true --config-settings=logging.level=INFO
+          python3 -m pip install .[all] --verbose
       - name: Test with pytest
         run: |
           python3 -m pytest

CMakeLists.txt

@@ -4,6 +4,14 @@ project(llama_cpp)
 
 option(BUILD_LLAMA_CPP "Build llama.cpp shared library and install alongside python package" ON)
 
+if (APPLE)
+    set(LLAMA_AVX ON CACHE BOOL)
+    set(LLAMA_AVX2 OFF CACHE BOOL)
+    set(LLAMA_FMA OFF CACHE BOOL)
+    set(LLAMA_F16C OFF CACHE BOOL)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=native -mtune=native")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native -mtune=native")
+endif()
 
 if (BUILD_LLAMA_CPP)
     set(BUILD_SHARED_LIBS "On")