make : fix indentation

Under certain environment, nvcc and gcc is installed under customized path but not standard path

Co-authored-by: Yan Lin <yanlin@baidu.com>

CMakeLists.txt

@@ -186,16 +186,7 @@ if (LLAMA_BLAS)
                 pkg_check_modules(DepBLAS REQUIRED flexiblas_api)
             elseif (${LLAMA_BLAS_VENDOR} MATCHES "Intel")
                 # all Intel* libraries share the same include path
-                pkg_check_modules(DepBLAS mkl-sdl)
-                if (NOT DepBLAS)
-                    if (BUILD_SHARED_LIBS)
-                        set(LINK_METHOD dynamic)
-                    else()
-                        set(LINK_METHOD static)
-                    endif()
-                    string(REGEX REPLACE ".*_" "" DATA_TYPE_MODEL ${LLAMA_BLAS_VENDOR})
-                    pkg_check_modules(DepBLAS REQUIRED mkl-${LINK_METHOD}-${DATA_TYPE_MODEL}-iomp)
-                endif()
+                pkg_check_modules(DepBLAS REQUIRED mkl-sdl)
             elseif (${LLAMA_BLAS_VENDOR} MATCHES "NVHPC")
                 # this doesn't provide pkg-config
                 # suggest to assign BLAS_INCLUDE_DIRS on your own

Makefile

@@ -193,8 +193,12 @@ ifdef LLAMA_CUBLAS
 	CXXFLAGS  += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
 	LDFLAGS   += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib
 	OBJS      += ggml-cuda.o
-	NVCC      = nvcc
 	NVCCFLAGS = --forward-unknown-to-host-compiler
+ifdef LLAMA_CUDA_NVCC
+	NVCC = $(LLAMA_CUDA_NVCC)
+else
+	NVCC = nvcc
+endif #LLAMA_CUDA_NVCC
 ifdef CUDA_DOCKER_ARCH
 	NVCCFLAGS += -Wno-deprecated-gpu-targets -arch=$(CUDA_DOCKER_ARCH)
 else
@@ -223,7 +227,9 @@ ifdef LLAMA_CUDA_KQUANTS_ITER
 else
 	NVCCFLAGS += -DK_QUANTS_PER_ITERATION=2
 endif
-
+ifdef LLAMA_CUDA_CCBIN
+	NVCCFLAGS += -ccbin $(LLAMA_CUDA_CCBIN)
+endif
 ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
 	$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
 endif # LLAMA_CUBLAS

README.md

@@ -360,7 +360,7 @@ Building the program with BLAS support may lead to some performance improvements
   ```bash
   mkdir build
   cd build
-  cmake .. -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=Intel10_lp64 -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+  cmake .. -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=Intel10_64lp -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
   cmake --build . --config Release
   ```
 

ci/run.sh

@@ -243,7 +243,7 @@ function gg_sum_open_llama_3b_v2 {
 if [ -z $GG_BUILD_LOW_PERF ]; then
     rm -rf ${SRC}/models-mnt
 
-    mnt_models=$(realpath ${MNT}/models)
+    mnt_models=${MNT}/models
     mkdir -p ${mnt_models}
     ln -sfn ${mnt_models} ${SRC}/models-mnt
 

flake.nix

@@ -6,7 +6,7 @@
   outputs = { self, nixpkgs, flake-utils }:
     flake-utils.lib.eachDefaultSystem (system:
       let
-        inherit (pkgs.stdenv) isAarch32 isAarch64 isx86_32 isx86_64 isDarwin;
+        inherit (pkgs.stdenv) isAarch32 isAarch64 isDarwin;
         osSpecific = with pkgs; [ openmpi ] ++
         (
           if isAarch64 && isDarwin then
@@ -22,14 +22,13 @@
               CoreGraphics
               CoreVideo
             ]
-          else if isx86_32 || isx86_64 then
-            with pkgs; [ mkl ]
           else
             with pkgs; [ openblas ]
         );
         pkgs = import nixpkgs { inherit system; };
+        nativeBuildInputs = with pkgs; [ cmake pkgconfig ];
         llama-python =
-          pkgs.python310.withPackages (ps: with ps; [ numpy sentencepiece ]);
+          pkgs.python3.withPackages (ps: with ps; [ numpy sentencepiece ]);
       in {
         packages.default = pkgs.stdenv.mkDerivation {
           name = "llama.cpp";
@@ -37,33 +36,21 @@
           postPatch = ''
             substituteInPlace ./ggml-metal.m \
               --replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"
+            substituteInPlace ./*.py --replace '/usr/bin/env python' '${llama-python}/bin/python'
           '';
-          nativeBuildInputs = with pkgs; [ cmake pkgconfig ];
+          nativeBuildInputs = nativeBuildInputs;
           buildInputs = osSpecific;
           cmakeFlags = [ "-DLLAMA_BUILD_SERVER=ON" "-DLLAMA_MPI=ON" "-DBUILD_SHARED_LIBS=ON" "-DCMAKE_SKIP_BUILD_RPATH=ON" ]
             ++ (if isAarch64 && isDarwin then [
               "-DCMAKE_C_FLAGS=-D__ARM_FEATURE_DOTPROD=1"
               "-DLLAMA_METAL=ON"
-            ] else if isx86_32 || isx86_64 then [
-              "-DLLAMA_BLAS=ON"
-              "-DLLAMA_BLAS_VENDOR=Intel10_lp64"
             ] else [
               "-DLLAMA_BLAS=ON"
               "-DLLAMA_BLAS_VENDOR=OpenBLAS"
           ]);
-          installPhase = ''
-            runHook preInstall
-
-            install -D bin/* -t $out/bin
-            install -Dm644 lib*.so -t $out/lib
+          postInstall = ''
             mv $out/bin/main $out/bin/llama
             mv $out/bin/server $out/bin/llama-server
-
-            echo "#!${llama-python}/bin/python" > $out/bin/convert.py
-            cat ${./convert.py} >> $out/bin/convert.py
-            chmod +x $out/bin/convert.py
-
-            runHook postInstall
           '';
           meta.mainProgram = "llama";
         };
@@ -81,7 +68,7 @@
         };
         apps.default = self.apps.${system}.llama;
         devShells.default = pkgs.mkShell {
-          packages = with pkgs; [ cmake llama-python ] ++ osSpecific;
+          packages = nativeBuildInputs ++ osSpecific;
         };
       });
 }