python312Packages.llama-cpp-python: fix linux build

Their cmake files were attempting to build with -mcpu=native+no<...>
which fails for gcc. Actually, we should never build for native anyway
(which is the raison d'etre for NIX_ENFORCE_NO_NATIVE).

Thankfully, the package provides a variable to disable native
optimizations.

pkgs/development/python-modules/llama-cpp-python/default.nix

@@ -64,7 +64,16 @@ buildPythonPackage rec {
 
   dontUseCmakeConfigure = true;
   SKBUILD_CMAKE_ARGS = lib.strings.concatStringsSep ";" (
-    lib.optionals cudaSupport [
+    # Set GGML_NATIVE=off. Otherwise, cmake attempts to build with
+    # -march=native* which is either a no-op (if cc-wrapper is able to ignore
+    # it), or an attempt to build a non-reproducible binary.
+    #
+    # This issue was spotted when cmake rules appended feature modifiers to
+    # -mcpu, breaking linux build as follows:
+    #
+    # cc1: error: unknown value ‘native+nodotprod+noi8mm+nosve’ for ‘-mcpu’
+    [ "-DGGML_NATIVE=off" ]
+    ++ lib.optionals cudaSupport [
       "-DGGML_CUDA=on"
       "-DCUDAToolkit_ROOT=${lib.getDev cudaPackages.cuda_nvcc}"
       "-DCMAKE_CUDA_COMPILER=${lib.getExe cudaPackages.cuda_nvcc}"
@@ -133,9 +142,5 @@ buildPythonPackage rec {
     changelog = "https://github.com/abetlen/llama-cpp-python/blob/v${version}/CHANGELOG.md";
     license = lib.licenses.mit;
     maintainers = with lib.maintainers; [ kirillrdy ];
-    badPlatforms = [
-      # cc1: error: unknown value ‘native+nodotprod+noi8mm+nosve’ for ‘-mcpu’
-      "aarch64-linux"
-    ];
   };
 }