vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

*.nix: nixfmt

Browse source 
nix shell github:piegamesde/nixfmt/rfc101-style --command \
    nixfmt flake.nix .devops/nix/*.nix
This commit is contained in:
Someone Serge 2023-12-25 16:23:56 +00:00
parent fea0239e24
commit c8d39878d9
No known key found for this signature in database
GPG key ID: 7B0E3B1390D61DA4
1 changed files with 82 additions and 80 deletions
Download patch file Download diff file Expand all files Collapse all files

162
.devops/nix/package.nix
View file

@ -97,86 +97,88 @@ let
];
in
effectiveStdenv.mkDerivation (finalAttrs: {
name = "llama.cpp";
src = ../../.;
meta = {
description = "Inference of LLaMA model in pure C/C++${descriptionSuffix}";
mainProgram = "llama";
};
postPatch = ''
substituteInPlace ./ggml-metal.m \
--replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"
# TODO: Package up each Python script or service appropriately.
# If we were to migrate to buildPythonPackage and prepare the `pyproject.toml`,
# we could make those *.py into setuptools' entrypoints
substituteInPlace ./*.py --replace "/usr/bin/env python" "${llama-python}/bin/python"
'';
nativeBuildInputs = [
cmake
ninja
pkg-config
git
] ++ optionals useCuda [ cudaPackages.cuda_nvcc ];
buildInputs =
[ mpi ]
++ optionals useOpenCL [ clblast ]
++ optionals useCuda cudaBuildInputs
++ optionals useRocm rocmBuildInputs
++ optionals effectiveStdenv.isDarwin darwinBuildInputs;
cmakeFlags =
[
(cmakeBool "LLAMA_NATIVE" true)
(cmakeBool "LLAMA_BUILD_SERVER" true)
(cmakeBool "BUILD_SHARED_LIBS" true)
(cmakeBool "CMAKE_SKIP_BUILD_RPATH" true)
(cmakeBool "LLAMA_METAL" useMetalKit)
(cmakeBool "LLAMA_BLAS" useBlas)
]
++ optionals useOpenCL [ (cmakeBool "LLAMA_CLBLAST" true) ]
++ optionals useCuda [ (cmakeBool "LLAMA_CUBLAS" true) ]
++ optionals useRocm [
(cmakeBool "LLAMA_HIPBLAS" true)
(cmakeFeature "CMAKE_C_COMPILER" "hipcc")
(cmakeFeature "CMAKE_CXX_COMPILER" "hipcc")
# Build all targets supported by rocBLAS. When updating search for TARGET_LIST_ROCM
# in https://github.com/ROCmSoftwarePlatform/rocBLAS/blob/develop/CMakeLists.txt
# and select the line that matches the current nixpkgs version of rocBLAS.
# Should likely use `rocmPackages.clr.gpuTargets`.
"-DAMDGPU_TARGETS=gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"
]
++ optionals useMetalKit [ (lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1") ]
++ optionals useBlas [ (lib.cmakeFeature "LLAMA_BLAS_VENDOR" "OpenBLAS") ];
# TODO(SomeoneSerge): It's better to add proper install targets at the CMake level,
# if they haven't been added yet.
postInstall = ''
mv $out/bin/main $out/bin/llama
mv $out/bin/server $out/bin/llama-server
mkdir -p $out/include
cp $src/llama.h $out/include/
'';
# Define the shells here, but don't add in the inputsFrom to avoid recursion.
passthru = {
shell = mkShell {
name = "default${descriptionSuffix}";
description = "contains numpy and sentencepiece";
buildInputs = [ llama-python ];
inputsFrom = [ finalAttrs.finalPackage ];
effectiveStdenv.mkDerivation (
finalAttrs: {
name = "llama.cpp";
src = ../../.;
meta = {
description = "Inference of LLaMA model in pure C/C++${descriptionSuffix}";
mainProgram = "llama";
};
shell-extra = mkShell {
name = "extra${descriptionSuffix}";
description = "contains numpy, sentencepiece, torchWithoutCuda, and transformers";
buildInputs = [ llama-python-extra ];
inputsFrom = [ finalAttrs.finalPackage ];
postPatch = ''
substituteInPlace ./ggml-metal.m \
--replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"
# TODO: Package up each Python script or service appropriately.
# If we were to migrate to buildPythonPackage and prepare the `pyproject.toml`,
# we could make those *.py into setuptools' entrypoints
substituteInPlace ./*.py --replace "/usr/bin/env python" "${llama-python}/bin/python"
'';
nativeBuildInputs = [
cmake
ninja
pkg-config
git
] ++ optionals useCuda [ cudaPackages.cuda_nvcc ];
buildInputs =
[ mpi ]
++ optionals useOpenCL [ clblast ]
++ optionals useCuda cudaBuildInputs
++ optionals useRocm rocmBuildInputs
++ optionals effectiveStdenv.isDarwin darwinBuildInputs;
cmakeFlags =
[
(cmakeBool "LLAMA_NATIVE" true)
(cmakeBool "LLAMA_BUILD_SERVER" true)
(cmakeBool "BUILD_SHARED_LIBS" true)
(cmakeBool "CMAKE_SKIP_BUILD_RPATH" true)
(cmakeBool "LLAMA_METAL" useMetalKit)
(cmakeBool "LLAMA_BLAS" useBlas)
]
++ optionals useOpenCL [ (cmakeBool "LLAMA_CLBLAST" true) ]
++ optionals useCuda [ (cmakeBool "LLAMA_CUBLAS" true) ]
++ optionals useRocm [
(cmakeBool "LLAMA_HIPBLAS" true)
(cmakeFeature "CMAKE_C_COMPILER" "hipcc")
(cmakeFeature "CMAKE_CXX_COMPILER" "hipcc")
# Build all targets supported by rocBLAS. When updating search for TARGET_LIST_ROCM
# in https://github.com/ROCmSoftwarePlatform/rocBLAS/blob/develop/CMakeLists.txt
# and select the line that matches the current nixpkgs version of rocBLAS.
# Should likely use `rocmPackages.clr.gpuTargets`.
"-DAMDGPU_TARGETS=gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"
]
++ optionals useMetalKit [ (lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1") ]
++ optionals useBlas [ (lib.cmakeFeature "LLAMA_BLAS_VENDOR" "OpenBLAS") ];
# TODO(SomeoneSerge): It's better to add proper install targets at the CMake level,
# if they haven't been added yet.
postInstall = ''
mv $out/bin/main $out/bin/llama
mv $out/bin/server $out/bin/llama-server
mkdir -p $out/include
cp $src/llama.h $out/include/
'';
# Define the shells here, but don't add in the inputsFrom to avoid recursion.
passthru = {
shell = mkShell {
name = "default${descriptionSuffix}";
description = "contains numpy and sentencepiece";
buildInputs = [ llama-python ];
inputsFrom = [ finalAttrs.finalPackage ];
};
shell-extra = mkShell {
name = "extra${descriptionSuffix}";
description = "contains numpy, sentencepiece, torchWithoutCuda, and transformers";
buildInputs = [ llama-python-extra ];
inputsFrom = [ finalAttrs.finalPackage ];
};
};
};
})
}
)