From 0126788271b4c6cbf57afb251f6fdd0c6dc098f2 Mon Sep 17 00:00:00 2001 From: ditsuke Date: Thu, 22 Feb 2024 23:55:55 +0530 Subject: [PATCH] style: format with nixfmt/rfc101-style --- .devops/nix/devshells.nix | 11 +- .devops/nix/nixpkgs-instances.nix | 18 +- .devops/nix/package.nix | 388 +++++++++++++++--------------- .devops/nix/scope.nix | 16 +- flake.nix | 4 +- 5 files changed, 213 insertions(+), 224 deletions(-) diff --git a/.devops/nix/devshells.nix b/.devops/nix/devshells.nix index 1862f0f08..9895d803f 100644 --- a/.devops/nix/devshells.nix +++ b/.devops/nix/devshells.nix @@ -2,12 +2,9 @@ perSystem = { config, lib, ... }: { - devShells = - lib.concatMapAttrs - (name: package: { - ${name} = package.passthru.shell; - ${name + "-extra"} = package.passthru.shell-extra; - }) - config.packages; + devShells = lib.concatMapAttrs (name: package: { + ${name} = package.passthru.shell; + ${name + "-extra"} = package.passthru.shell-extra; + }) config.packages; }; } diff --git a/.devops/nix/nixpkgs-instances.nix b/.devops/nix/nixpkgs-instances.nix index 4a2f81c4b..90d683a71 100644 --- a/.devops/nix/nixpkgs-instances.nix +++ b/.devops/nix/nixpkgs-instances.nix @@ -26,16 +26,14 @@ config.cudaSupport = true; config.allowUnfreePredicate = p: - builtins.all - ( - license: - license.free - || builtins.elem license.shortName [ - "CUDA EULA" - "cuDNN EULA" - ] - ) - (p.meta.licenses or [ p.meta.license ]); + builtins.all ( + license: + license.free + || builtins.elem license.shortName [ + "CUDA EULA" + "cuDNN EULA" + ] + ) (p.meta.licenses or [ p.meta.license ]); }; # Ensure dependencies use ROCm consistently pkgsRocm = import inputs.nixpkgs { diff --git a/.devops/nix/package.nix b/.devops/nix/package.nix index cfffac257..e065004af 100644 --- a/.devops/nix/package.nix +++ b/.devops/nix/package.nix @@ -20,12 +20,14 @@ vulkan-loader, curl, shaderc, - useBlas ? builtins.all (x: !x) [ - useCuda - useMetalKit - useRocm - useVulkan - ] && blas.meta.available, + useBlas ? + builtins.all (x: !x) [ + useCuda + useMetalKit + useRocm + useVulkan + ] + && blas.meta.available, useCuda ? config.cudaSupport, useMetalKit ? stdenv.isAarch64 && stdenv.isDarwin, useMpi ? false, # Increases the runtime closure size by ~700M @@ -38,7 +40,7 @@ # otherwise we get libstdc++ errors downstream. effectiveStdenv ? if useCuda then cudaPackages.backendStdenv else stdenv, enableStatic ? effectiveStdenv.hostPlatform.isStatic, - precompileMetalShaders ? false + precompileMetalShaders ? false, }@inputs: let @@ -63,9 +65,9 @@ let pnameSuffix = strings.optionalString (suffices != [ ]) "-${strings.concatMapStringsSep "-" strings.toLower suffices}"; - descriptionSuffix = - strings.optionalString (suffices != [ ]) - ", accelerated with ${strings.concatStringsSep ", " suffices}"; + descriptionSuffix = strings.optionalString ( + suffices != [ ] + ) ", accelerated with ${strings.concatStringsSep ", " suffices}"; executableSuffix = effectiveStdenv.hostPlatform.extensions.executable; @@ -76,41 +78,37 @@ let # # TODO: Package up each Python script or service appropriately, by making # them into "entrypoints" - llama-python = python3.withPackages ( - ps: [ - ps.numpy - ps.sentencepiece - ] - ); + llama-python = python3.withPackages (ps: [ + ps.numpy + ps.sentencepiece + ]); # TODO(Green-Sky): find a better way to opt-into the heavy ml python runtime - llama-python-extra = python3.withPackages ( - ps: [ - ps.numpy - ps.sentencepiece - ps.tiktoken - ps.torchWithoutCuda - ps.transformers + llama-python-extra = python3.withPackages (ps: [ + ps.numpy + ps.sentencepiece + ps.tiktoken + ps.torchWithoutCuda + ps.transformers - # server bench - ps.matplotlib + # server bench + ps.matplotlib - # server tests - ps.openai - ps.behave - ps.prometheus-client + # server tests + ps.openai + ps.behave + ps.prometheus-client - # for examples/pydantic-models-to-grammar-examples.py - ps.docstring-parser - ps.pydantic + # for examples/pydantic-models-to-grammar-examples.py + ps.docstring-parser + ps.pydantic - # for scripts/compare-llama-bench.py - ps.gitpython - ps.tabulate - ] - ); + # for scripts/compare-llama-bench.py + ps.gitpython + ps.tabulate + ]); - xcrunHost = runCommand "xcrunHost" {} '' + xcrunHost = runCommand "xcrunHost" { } '' mkdir -p $out/bin ln -s /usr/bin/xcrun $out/bin ''; @@ -145,178 +143,174 @@ let ]; in -effectiveStdenv.mkDerivation ( - finalAttrs: { - pname = "llama-cpp${pnameSuffix}"; - version = llamaVersion; +effectiveStdenv.mkDerivation (finalAttrs: { + pname = "llama-cpp${pnameSuffix}"; + version = llamaVersion; - # Note: none of the files discarded here are visible in the sandbox or - # affect the output hash. This also means they can be modified without - # triggering a rebuild. - src = lib.cleanSourceWith { - filter = - name: type: - let - noneOf = builtins.all (x: !x); - baseName = baseNameOf name; - in - noneOf [ - (lib.hasSuffix ".nix" name) # Ignore *.nix files when computing outPaths - (lib.hasSuffix ".md" name) # Ignore *.md changes whe computing outPaths - (lib.hasPrefix "." baseName) # Skip hidden files and directories - (baseName == "flake.lock") - ]; - src = lib.cleanSource ../../.; - }; - - postPatch = '' - substituteInPlace ./ggml/src/ggml-metal.m \ - --replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";" - substituteInPlace ./ggml/src/ggml-metal.m \ - --replace '[bundle pathForResource:@"default" ofType:@"metallib"];' "@\"$out/bin/default.metallib\";" - ''; - - # With PR#6015 https://github.com/ggerganov/llama.cpp/pull/6015, - # `default.metallib` may be compiled with Metal compiler from XCode - # and we need to escape sandbox on MacOS to access Metal compiler. - # `xcrun` is used find the path of the Metal compiler, which is varible - # and not on $PATH - # see https://github.com/ggerganov/llama.cpp/pull/6118 for discussion - __noChroot = effectiveStdenv.isDarwin && useMetalKit && precompileMetalShaders; - - nativeBuildInputs = - [ - cmake - ninja - pkg-config - git - ] - ++ optionals useCuda [ - cudaPackages.cuda_nvcc - autoAddDriverRunpath - ] - ++ optionals (effectiveStdenv.hostPlatform.isGnu && enableStatic) [ - glibc.static - ] ++ optionals (effectiveStdenv.isDarwin && useMetalKit && precompileMetalShaders) [ - xcrunHost + # Note: none of the files discarded here are visible in the sandbox or + # affect the output hash. This also means they can be modified without + # triggering a rebuild. + src = lib.cleanSourceWith { + filter = + name: type: + let + noneOf = builtins.all (x: !x); + baseName = baseNameOf name; + in + noneOf [ + (lib.hasSuffix ".nix" name) # Ignore *.nix files when computing outPaths + (lib.hasSuffix ".md" name) # Ignore *.md changes whe computing outPaths + (lib.hasPrefix "." baseName) # Skip hidden files and directories + (baseName == "flake.lock") ]; + src = lib.cleanSource ../../.; + }; - buildInputs = - optionals effectiveStdenv.isDarwin darwinBuildInputs - ++ optionals useCuda cudaBuildInputs - ++ optionals useMpi [ mpi ] - ++ optionals useRocm rocmBuildInputs - ++ optionals useBlas [ blas ] - ++ optionals useVulkan vulkanBuildInputs - ++ optionals enableCurl [ curl ]; + postPatch = '' + substituteInPlace ./ggml/src/ggml-metal.m \ + --replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";" + substituteInPlace ./ggml/src/ggml-metal.m \ + --replace '[bundle pathForResource:@"default" ofType:@"metallib"];' "@\"$out/bin/default.metallib\";" + ''; - cmakeFlags = - [ - (cmakeBool "LLAMA_BUILD_SERVER" true) - (cmakeBool "BUILD_SHARED_LIBS" (!enableStatic)) - (cmakeBool "CMAKE_SKIP_BUILD_RPATH" true) - (cmakeBool "LLAMA_CURL" enableCurl) - (cmakeBool "GGML_NATIVE" false) - (cmakeBool "GGML_BLAS" useBlas) - (cmakeBool "GGML_CUDA" useCuda) - (cmakeBool "GGML_HIPBLAS" useRocm) - (cmakeBool "GGML_METAL" useMetalKit) - (cmakeBool "GGML_VULKAN" useVulkan) - (cmakeBool "GGML_STATIC" enableStatic) - ] - ++ optionals useCuda [ - ( - with cudaPackages.flags; - cmakeFeature "CMAKE_CUDA_ARCHITECTURES" ( - builtins.concatStringsSep ";" (map dropDot cudaCapabilities) - ) + # With PR#6015 https://github.com/ggerganov/llama.cpp/pull/6015, + # `default.metallib` may be compiled with Metal compiler from XCode + # and we need to escape sandbox on MacOS to access Metal compiler. + # `xcrun` is used find the path of the Metal compiler, which is varible + # and not on $PATH + # see https://github.com/ggerganov/llama.cpp/pull/6118 for discussion + __noChroot = effectiveStdenv.isDarwin && useMetalKit && precompileMetalShaders; + + nativeBuildInputs = + [ + cmake + ninja + pkg-config + git + ] + ++ optionals useCuda [ + cudaPackages.cuda_nvcc + + autoAddDriverRunpath + ] + ++ optionals (effectiveStdenv.hostPlatform.isGnu && enableStatic) [ glibc.static ] + ++ optionals (effectiveStdenv.isDarwin && useMetalKit && precompileMetalShaders) [ xcrunHost ]; + + buildInputs = + optionals effectiveStdenv.isDarwin darwinBuildInputs + ++ optionals useCuda cudaBuildInputs + ++ optionals useMpi [ mpi ] + ++ optionals useRocm rocmBuildInputs + ++ optionals useBlas [ blas ] + ++ optionals useVulkan vulkanBuildInputs + ++ optionals enableCurl [ curl ]; + + cmakeFlags = + [ + (cmakeBool "LLAMA_BUILD_SERVER" true) + (cmakeBool "BUILD_SHARED_LIBS" (!enableStatic)) + (cmakeBool "CMAKE_SKIP_BUILD_RPATH" true) + (cmakeBool "LLAMA_CURL" enableCurl) + (cmakeBool "GGML_NATIVE" false) + (cmakeBool "GGML_BLAS" useBlas) + (cmakeBool "GGML_CUDA" useCuda) + (cmakeBool "GGML_HIPBLAS" useRocm) + (cmakeBool "GGML_METAL" useMetalKit) + (cmakeBool "GGML_VULKAN" useVulkan) + (cmakeBool "GGML_STATIC" enableStatic) + ] + ++ optionals useCuda [ + ( + with cudaPackages.flags; + cmakeFeature "CMAKE_CUDA_ARCHITECTURES" ( + builtins.concatStringsSep ";" (map dropDot cudaCapabilities) ) - ] - ++ optionals useRocm [ - (cmakeFeature "CMAKE_HIP_COMPILER" "${rocmPackages.llvm.clang}/bin/clang") - (cmakeFeature "CMAKE_HIP_ARCHITECTURES" (builtins.concatStringsSep ";" rocmPackages.clr.gpuTargets)) - ] - ++ optionals useMetalKit [ - (lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1") - (cmakeBool "GGML_METAL_EMBED_LIBRARY" (!precompileMetalShaders)) - ]; + ) + ] + ++ optionals useRocm [ + (cmakeFeature "CMAKE_HIP_COMPILER" "${rocmPackages.llvm.clang}/bin/clang") + (cmakeFeature "CMAKE_HIP_ARCHITECTURES" (builtins.concatStringsSep ";" rocmPackages.clr.gpuTargets)) + ] + ++ optionals useMetalKit [ + (lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1") + (cmakeBool "GGML_METAL_EMBED_LIBRARY" (!precompileMetalShaders)) + ]; - # Environment variables needed for ROCm - env = optionals useRocm { - ROCM_PATH = "${rocmPackages.clr}"; - HIP_DEVICE_LIB_PATH = "${rocmPackages.rocm-device-libs}/amdgcn/bitcode"; + # Environment variables needed for ROCm + env = optionals useRocm { + ROCM_PATH = "${rocmPackages.clr}"; + HIP_DEVICE_LIB_PATH = "${rocmPackages.rocm-device-libs}/amdgcn/bitcode"; + }; + + # TODO(SomeoneSerge): It's better to add proper install targets at the CMake level, + # if they haven't been added yet. + postInstall = '' + mkdir -p $out/include + cp $src/include/llama.h $out/include/ + ''; + + # Define the shells here, but don't add in the inputsFrom to avoid recursion. + passthru = { + inherit + useBlas + useCuda + useMetalKit + useMpi + useRocm + useVulkan + ; + + shell = mkShell { + name = "shell-${finalAttrs.finalPackage.name}"; + description = "contains numpy and sentencepiece"; + buildInputs = [ llama-python ]; + inputsFrom = [ finalAttrs.finalPackage ]; + shellHook = '' + addToSearchPath "LD_LIBRARY_PATH" "${lib.getLib effectiveStdenv.cc.cc}/lib" + ''; }; - # TODO(SomeoneSerge): It's better to add proper install targets at the CMake level, - # if they haven't been added yet. - postInstall = '' - mkdir -p $out/include - cp $src/include/llama.h $out/include/ - ''; - - # Define the shells here, but don't add in the inputsFrom to avoid recursion. - passthru = { - inherit - useBlas - useCuda - useMetalKit - useMpi - useRocm - useVulkan - ; - - shell = mkShell { - name = "shell-${finalAttrs.finalPackage.name}"; - description = "contains numpy and sentencepiece"; - buildInputs = [ llama-python ]; - inputsFrom = [ finalAttrs.finalPackage ]; - shellHook = '' - addToSearchPath "LD_LIBRARY_PATH" "${lib.getLib effectiveStdenv.cc.cc}/lib" - ''; - }; - - shell-extra = mkShell { - name = "shell-extra-${finalAttrs.finalPackage.name}"; - description = "contains numpy, sentencepiece, torchWithoutCuda, and transformers"; - buildInputs = [ llama-python-extra ]; - inputsFrom = [ finalAttrs.finalPackage ]; - }; + shell-extra = mkShell { + name = "shell-extra-${finalAttrs.finalPackage.name}"; + description = "contains numpy, sentencepiece, torchWithoutCuda, and transformers"; + buildInputs = [ llama-python-extra ]; + inputsFrom = [ finalAttrs.finalPackage ]; }; + }; - meta = { - # Configurations we don't want even the CI to evaluate. Results in the - # "unsupported platform" messages. This is mostly a no-op, because - # cudaPackages would've refused to evaluate anyway. - badPlatforms = optionals useCuda lib.platforms.darwin; + meta = { + # Configurations we don't want even the CI to evaluate. Results in the + # "unsupported platform" messages. This is mostly a no-op, because + # cudaPackages would've refused to evaluate anyway. + badPlatforms = optionals useCuda lib.platforms.darwin; - # Configurations that are known to result in build failures. Can be - # overridden by importing Nixpkgs with `allowBroken = true`. - broken = (useMetalKit && !effectiveStdenv.isDarwin); + # Configurations that are known to result in build failures. Can be + # overridden by importing Nixpkgs with `allowBroken = true`. + broken = (useMetalKit && !effectiveStdenv.isDarwin); - description = "Inference of LLaMA model in pure C/C++${descriptionSuffix}"; - homepage = "https://github.com/ggerganov/llama.cpp/"; - license = lib.licenses.mit; + description = "Inference of LLaMA model in pure C/C++${descriptionSuffix}"; + homepage = "https://github.com/ggerganov/llama.cpp/"; + license = lib.licenses.mit; - # Accommodates `nix run` and `lib.getExe` - mainProgram = "llama-cli"; + # Accommodates `nix run` and `lib.getExe` + mainProgram = "llama-cli"; - # These people might respond, on the best effort basis, if you ping them - # in case of Nix-specific regressions or for reviewing Nix-specific PRs. - # Consider adding yourself to this list if you want to ensure this flake - # stays maintained and you're willing to invest your time. Do not add - # other people without their consent. Consider removing people after - # they've been unreachable for long periods of time. + # These people might respond, on the best effort basis, if you ping them + # in case of Nix-specific regressions or for reviewing Nix-specific PRs. + # Consider adding yourself to this list if you want to ensure this flake + # stays maintained and you're willing to invest your time. Do not add + # other people without their consent. Consider removing people after + # they've been unreachable for long periods of time. - # Note that lib.maintainers is defined in Nixpkgs, but you may just add - # an attrset following the same format as in - # https://github.com/NixOS/nixpkgs/blob/f36a80e54da29775c78d7eff0e628c2b4e34d1d7/maintainers/maintainer-list.nix - maintainers = with lib.maintainers; [ - philiptaron - SomeoneSerge - ]; + # Note that lib.maintainers is defined in Nixpkgs, but you may just add + # an attrset following the same format as in + # https://github.com/NixOS/nixpkgs/blob/f36a80e54da29775c78d7eff0e628c2b4e34d1d7/maintainers/maintainer-list.nix + maintainers = with lib.maintainers; [ + philiptaron + SomeoneSerge + ]; - # Extend `badPlatforms` instead - platforms = lib.platforms.all; - }; - } -) + # Extend `badPlatforms` instead + platforms = lib.platforms.all; + }; +}) diff --git a/.devops/nix/scope.nix b/.devops/nix/scope.nix index 78530c9e8..d9fc1fdf0 100644 --- a/.devops/nix/scope.nix +++ b/.devops/nix/scope.nix @@ -8,12 +8,10 @@ # because it allows users to apply overlays later using `overrideScope'`. # Cf. https://noogle.dev/f/lib/makeScope -lib.makeScope newScope ( - self: { - inherit llamaVersion; - llama-cpp = self.callPackage ./package.nix { }; - docker = self.callPackage ./docker.nix { }; - docker-min = self.callPackage ./docker.nix { interactive = false; }; - sif = self.callPackage ./sif.nix { }; - } -) +lib.makeScope newScope (self: { + inherit llamaVersion; + llama-cpp = self.callPackage ./package.nix { }; + docker = self.callPackage ./docker.nix { }; + docker-min = self.callPackage ./docker.nix { interactive = false; }; + sif = self.callPackage ./sif.nix { }; +}) diff --git a/flake.nix b/flake.nix index c69637d11..4c75cbbcc 100644 --- a/flake.nix +++ b/flake.nix @@ -145,7 +145,9 @@ # the same path you would with an overlay. legacyPackages = { llamaPackages = pkgs.callPackage .devops/nix/scope.nix { inherit llamaVersion; }; - llamaPackagesWindows = pkgs.pkgsCross.mingwW64.callPackage .devops/nix/scope.nix { inherit llamaVersion; }; + llamaPackagesWindows = pkgs.pkgsCross.mingwW64.callPackage .devops/nix/scope.nix { + inherit llamaVersion; + }; llamaPackagesCuda = pkgsCuda.callPackage .devops/nix/scope.nix { inherit llamaVersion; }; llamaPackagesRocm = pkgsRocm.callPackage .devops/nix/scope.nix { inherit llamaVersion; }; };