Arch Linux User Repository

I couldn't receive notifications from AUR in real-time, so if you have problems that require immediate feedback or communication, please consider submitting an issue in this GitHub repository, at which I maintain all my AUR packages. Thank you for your understanding.

Alternate versions

llama.cpp
llama.cpp-vulkan
llama.cpp-sycl-fp16
llama.cpp-sycl-fp32
llama.cpp-cuda
llama.cpp-cuda-f16
llama.cpp-hip

it should properly source=( from upstream rather than trying to shallow clone in prepare. You can source the release archive if you're trying to avoid cloning a deep history

Encountered an error again when updating the package during the call to pkgver().

git describe --tags --abbrev=0   
fatal: No names found, cannot describe anything.

Error analysis revealed that the issue arises from combining the code retrieval command with frequent commits to the master branch.

git clone --depth 3 --single-branch --branch master "${url}" "${_pkgname}"

Package updates will only work for a few days after a new version is released. If attempting to update later, the required tag won't be present in the fetched master branch history. As solution to this problem, I suggest using:

git clone --branch "$pkgver" --depth 1 "${url}" "${_pkgname}"

This approach directly fetches the required version and works reliably at all times.

As of now this conflicts with stable-diffusion.cpp-vulkan-git because they both provide /usr/lib/cmake/ggml/ggml-config.cmake and /usr/lib/cmake/ggml/ggml-version.cmake.

This changed now with new ggml-vulkan

[ 27%] Building CXX object src/CMakeFiles/llama.dir/llama-quant.cpp.o
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp: In function ‘bool weight_buft_supported(const llama_hparams&, ggml_tensor*, ggml_op, ggml_backend_buffer_type_t, ggml_backend_dev_t)’:
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp:231:42: error: too many arguments to function ‘ggml_tensor* ggml_ssm_scan(ggml_context*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*)’
  231 |                 op_tensor = ggml_ssm_scan(ctx, s, x, dt, w, B, C, ids);
      |                             ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/../include/llama.h:4,
                 from /var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.h:3,
                 from /var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp:1:
/usr/include/ggml.h:2009:35: note: declared here
 2009 |     GGML_API struct ggml_tensor * ggml_ssm_scan(
      |                                   ^~~~~~~~~~~~~
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp: In lambda function:
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp:9922:37: error: too many arguments to function ‘ggml_tensor* ggml_ssm_scan(ggml_context*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*)’
 9922 |                 return ggml_ssm_scan(ctx, ssm, x, dt, A, B, C, ids);
      |                        ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/include/ggml.h:2009:35: note: declared here
 2009 |     GGML_API struct ggml_tensor * ggml_ssm_scan(
      |                                   ^~~~~~~~~~~~~
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp: In lambda function:
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-model.cpp:10046:37: error: too many arguments to function ‘ggml_tensor* ggml_ssm_scan(ggml_context*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*)’
10046 |                 return ggml_ssm_scan(ctx, ssm, x, dt, A, B, C, ids);
      |                        ~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/include/ggml.h:2009:35: note: declared here
 2009 |     GGML_API struct ggml_tensor * ggml_ssm_scan(

Same issue reported here somehow.

https://aur.archlinux.org/packages/llama.cpp-hip

Still error with b5827

/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:618:23: error: ‘ggml_reglu’ was not declared in this scope; did you mean ‘ggml_relu’?
  618 |                 cur = ggml_reglu(ctx0, cur);
      |                       ^~~~~~~~~~
      |                       ggml_relu
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp: In member function ‘ggml_tensor* llm_graph_context::build_moe_ffn(ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, int64_t, int64_t, llm_ffn_op_type, bool, bool, float, llama_expert_gating_func_type, int) const’:
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:751:23: error: ‘ggml_swiglu_split’ was not declared in this scope
  751 |                 cur = ggml_swiglu_split(ctx0, cur, up);
      |                       ^~~~~~~~~~~~~~~~~
/var/cache/private/pamac/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:759:23: error: ‘ggml_geglu_split’ was not declared in this scope
  759 |                 cur = ggml_geglu_split(ctx0, cur, up);
      |                       ^~~~~~~~~~~~~~~~
make[2]: *** [src/CMakeFiles/llama.dir/build.make:191: src/CMakeFiles/llama.dir/llama-graph.cpp.o] Error 1
make[2]: *** Attesa per i processi non terminati....
make[1]: *** [CMakeFiles/Makefile2:1007: src/CMakeFiles/llama.dir/all] Error 2
make: *** [Makefile:136: all] Error 2

@eSPiYa Upstream issue atm

I already installed ggml-vulkan, but the build still fails

/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp: In member function ‘ggml_tensor* llm_graph_context::build_ffn(ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, ggml_tensor*, llm_ffn_op_type, llm_ffn_gate_type, int) const’:
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:564:23: error: ‘ggml_swiglu_split’ was not declared in this scope
564 |                 cur = ggml_swiglu_split(ctx0, cur, tmp);
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:573:23: error: ‘ggml_geglu_split’ was not declared in this scope
573 |                 cur = ggml_geglu_split(ctx0, cur, tmp);
    |                       ^~~~~~~~~~~~~~~~
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:586:23: error: ‘ggml_reglu_split’ was not declared in this scope
586 |                 cur = ggml_reglu_split(ctx0, cur, tmp);
    |                       ^~~~~~~~~~~~~~~~
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:603:23: error: ‘ggml_swiglu’ was not declared in this scope; did you mean ‘ggml_silu’?
603 |                 cur = ggml_swiglu(ctx0, cur);
    |                       ^~~~~~~~~~~
    |                       ggml_silu
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:608:23: error: ‘ggml_geglu’ was not declared in this scope; did you mean ‘ggml_gelu’?
608 |                 cur = ggml_geglu(ctx0, cur);
    |                       ^~~~~~~~~~
    |                       ggml_gelu
/home/<user>/.cache/paru/clone/llama.cpp-vulkan/src/llama.cpp/src/llama-graph.cpp:613:23: error: ‘ggml_reglu’ was not declared in this scope; did you mean ‘ggml_relu’?
613 |                 cur = ggml_reglu(ctx0, cur);
    |                       ^~~~~~~~~~
    |                       ggml_relu

❯ paru -S llama.cpp-vulkan
:: Resolving dependencies...
error: could not find all required packages:
libggml-vulkan (wanted by: llama.cpp-vulkan)

❯ paru -Sii ggml-vulkan-git |rg Provides
Provides        : ggml-vulkan  ggml  libggml

Use ggml-vulkan or libggml instead