Unverified Commit d8724afc authored by Pol Dellaiera's avatar Pol Dellaiera Committed by GitHub
Browse files

Merge pull request #323331 from xddxdd/llama-cpp 

llama-cpp: 3091 -> 3260
parents 83a657ca 134743c0

pkgs/by-name/ll/llama-cpp/package.nix

+19 −20
Original line number Diff line number Diff line
@@ -71,13 +71,13 @@ let 
in

effectiveStdenv.mkDerivation (finalAttrs: {

  pname = "llama-cpp";

  version = "3091";

  version = "3260";



  src = fetchFromGitHub {

    owner = "ggerganov";

    repo = "llama.cpp";

    rev = "refs/tags/b${finalAttrs.version}";

    hash = "sha256-ppujag6Nrk/M9QMQ4mYe2iADsfKzmfKtOP8Ib7GZBmk=";

    hash = "sha256-0KVwSzxfGinpv5KkDCgF2J+1ijDv87PlDrC+ldscP6s=";

    leaveDotGit = true;

    postFetch = ''

      git -C "$out" rev-parse --short HEAD > $out/COMMIT
@@ -86,12 +86,12 @@ effectiveStdenv.mkDerivation (finalAttrs: { 
  };



  postPatch = ''

    substituteInPlace ./ggml-metal.m \

    substituteInPlace ./ggml/src/ggml-metal.m \

      --replace-fail '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"



    substituteInPlace ./scripts/build-info.cmake \

      --replace-fail 'set(BUILD_NUMBER 0)' 'set(BUILD_NUMBER ${finalAttrs.version})' \

      --replace-fail 'set(BUILD_COMMIT "unknown")' "set(BUILD_COMMIT \"$(cat COMMIT)\")"

    substituteInPlace ./scripts/build-info.sh \

      --replace-fail 'build_number="0"' 'build_number="${finalAttrs.version}"' \

      --replace-fail 'build_commit="unknown"' "build_commit=\"$(cat COMMIT)\""

  '';



  nativeBuildInputs = [ cmake ninja pkg-config git ]
@@ -109,17 +109,16 @@ effectiveStdenv.mkDerivation (finalAttrs: { 


  cmakeFlags = [

    # -march=native is non-deterministic; override with platform-specific flags if needed

    (cmakeBool "LLAMA_NATIVE" false)

    (cmakeBool "BUILD_SHARED_SERVER" true)

    (cmakeBool "GGML_NATIVE" false)

    (cmakeBool "LLAMA_BUILD_SERVER" true)

    (cmakeBool "BUILD_SHARED_LIBS" true)

    (cmakeBool "BUILD_SHARED_LIBS" true)

    (cmakeBool "LLAMA_BLAS" blasSupport)

    (cmakeBool "LLAMA_CLBLAST" openclSupport)

    (cmakeBool "LLAMA_CUDA" cudaSupport)

    (cmakeBool "LLAMA_HIPBLAS" rocmSupport)

    (cmakeBool "LLAMA_METAL" metalSupport)

    (cmakeBool "LLAMA_RPC" rpcSupport)

    (cmakeBool "LLAMA_VULKAN" vulkanSupport)

    (cmakeBool "GGML_BLAS" blasSupport)

    (cmakeBool "GGML_CLBLAST" openclSupport)

    (cmakeBool "GGML_CUDA" cudaSupport)

    (cmakeBool "GGML_HIPBLAS" rocmSupport)

    (cmakeBool "GGML_METAL" metalSupport)

    (cmakeBool "GGML_RPC" rpcSupport)

    (cmakeBool "GGML_VULKAN" vulkanSupport)

  ]

      ++ optionals cudaSupport [

        (cmakeFeature "CMAKE_CUDA_ARCHITECTURES" cudaPackages.flags.cmakeCudaArchitecturesString)
@@ -138,7 +137,6 @@ effectiveStdenv.mkDerivation (finalAttrs: { 
        (cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1")

        (cmakeBool "LLAMA_METAL_EMBED_LIBRARY" true)

      ] ++ optionals rpcSupport [

        "-DLLAMA_RPC=ON"

        # This is done so we can move rpc-server out of bin because llama.cpp doesn't

        # install rpc-server in their install target.

        "-DCMAKE_SKIP_BUILD_RPATH=ON"
@@ -147,10 +145,11 @@ effectiveStdenv.mkDerivation (finalAttrs: { 
  # upstream plans on adding targets at the cmakelevel, remove those

  # additional steps after that

  postInstall = ''

    mv $out/bin/main $out/bin/llama

    mv $out/bin/server $out/bin/llama-server

    # Match previous binary name for this package

    ln -sf $out/bin/llama-cli $out/bin/llama



    mkdir -p $out/include

    cp $src/llama.h $out/include/

    cp $src/include/llama.h $out/include/

  '' + optionalString rpcSupport "cp bin/rpc-server $out/bin/llama-rpc-server";



  passthru.updateScript = nix-update-script {