Update Dockerfile build mcpu/march

Passing mcpu/march to cmake

Author: tc-wolf

Dockerfile

@@ -12,7 +12,8 @@ COPY . .
 # Install to /usr so that easily findable by cmake
 RUN mv /OpenBLAS /opt/OpenBLAS && cd /opt/OpenBLAS && make install PREFIX=/usr/ && cd /
 
-RUN PKG_CONFIG_PATH="/opt/OpenBLAS/install/lib/pkgconfig" CMAKE_ARGS="-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DGGML_LLAMAFILE=1" pip install -e .[server]
+# Have to disable GGML_LLAMAFILE for Q4_0_4_4 quantization
+RUN PKG_CONFIG_PATH="/opt/OpenBLAS/install/lib/pkgconfig" CMAKE_ARGS="-DGGML_BLAS=ON;-DGGML_BLAS_VENDOR=OpenBLAS;-DGGML_LLAMAFILE=OFF;-DCMAKE_C_FLAGS=-march=armv8.2-a+crypto+fp16+rcpc+dotprod -mcpu=cortex-a78c+crypto+noprofile+nossbs+noflagm+nopauth -mtune=cortex-a78c" pip install -e .[server]
 
 RUN cd /root && pyinstaller -DF /llama_cpp/server/__main__.py \
     --add-data /usr/lib/libopenblas.so:. \