Arch Linux User Repository

I'm having trouble compiling the package. Here's the error message:

[ 15%] Built target llama-gguf-hash
/tmp/makepkg/llama.cpp-git/src/llama.cpp/src/llama-model.cpp: In constructor ‘llm_build_apertus::llm_build_apertus(const llama_model&, const llm_graph_params&)’:
/tmp/makepkg/llama.cpp-git/src/llama.cpp/src/llama-model.cpp:19266:43: error: ‘ggml_xielu’ was not declared in this scope; did you mean ‘ggml_silu’?
19266 |                 ggml_tensor * activated = ggml_xielu(ctx0, up, alpha_n_val, alpha_p_val, beta_val, eps_val);
      |                                           ^~~~~~~~~~
      |                                           ggml_silu
make[2]: *** [src/CMakeFiles/llama.dir/build.make:359: src/CMakeFiles/llama.dir/llama-model.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1053: src/CMakeFiles/llama.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

I posted about this on the llama.cpp GitHub repository, and a contributor was kind enough to respond in github issue #15852 comment

You probably have an older ggml installed and it's using that instead of the one in repository.

To be honest, I'm not entirely sure I understand this. Does it mean that ggml is being developed in multiple places (ggml, llama.cpp, and whisper.cpp), and that's causing the issue? Would using libggml-git, which is based on an outdated version of ggml compared to whisper.cpp, explain why I can't compile the package? Currently, my workaround is to downgrade to b6661.

@robertfoster, I follow the intent but I don't think we're able to build only one package, and so we are required to build, and install dependencies, for every package provided by the PKGBUILD. It looks like the ability to build only one was removed a number of years ago: [pacman-dev] [PATCH] makepkg: remove ability to build individual packages. If that's the deal then that's the deal, I guess.

@itsdesmond

because of this https://wiki.archlinux.org/title/PKGBUILD#pkgbase

Here a split PKGBUILD is used. As you can see from documentation, makedepends is not considered in packages functions. This means that with split logic, you need to declare only one array of makedepends and only one build function. Hope that is clear.

I'm a bit confused by this package. Is it necessary that I build every package in order to install just one of them? Or is there some sort of targeting command I'm not taking advantage of?

--- PKGBUILD.old        2024-11-11 12:30:54.000000000 +0800
+++ PKGBUILD.new        2024-11-11 12:33:31.451801874 +0800
@@ -120,6 +120,7 @@ build() {
   echo "Build ${pkgbase} with CUBlas (NVIDIA CUDA)"
   cd "${srcdir}/${_name}-cublas"
   export PATH+=":/opt/cuda/bin"
+  export NVCC_CCBIN="gcc-13"
   cmake "${_cmake_cublas_args[@]}"
   cmake --build build
@@ -181,7 +182,7 @@ package_llama.cpp-cublas-git() {
   provides=("${_name}")
   conflicts=("${_name}")

-  cd "${_name}-clblas"
+  cd "${_name}-cublas"
   _package
 }

llama.cpp-hipblas-git should depends hipblas rather than rocm-hip-runtime

Another report with full output of the error https://gitlab.archlinux.org/archlinux/packaging/packages/cuda/-/issues/10

Looks like CMake CUDA error comes from the wrong compiler used https://bbs.archlinux.org/viewtopic.php?id=300160

CMake Error at /usr/share/cmake/Modules/CMakeDetermineCompilerId.cmake:838 (message): Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.

Your packages llama.cpp-vulkan-git and whisper.cpp-clblas do conflict which each other, which is not reflected in the conflicts array.

Please add the corresponding conflicts entry, or, to ensure compatibility, think of stripping out the libggml stuff and depend on libggml:

error: failed to commit transaction (conflicting files)
/usr/include/ggml-alloc.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-backend.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-blas.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-cann.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-cuda.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-kompute.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-metal.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-rpc.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-sycl.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/include/ggml-vulkan.h exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
/usr/lib/libggml.so exists in both 'llama.cpp-vulkan-git' and 'whisper.cpp-clblas'
Errors occurred, no packages were upgraded.

Regards!