Arch Linux User Repository

I had to roll back to f450a5df747d to rebuild 0.22.1 in order to get it working on my 1070 ti.

Looks good, and thank you!

@csjewell: I basically implemented your suggestion, with a few minor modifications. Could you take a look? I'll add you as a co-maintainer, in case you're interested in contributing anything else! Thanks for your help!

'Drive-by-patching' to stop building mlx-cuda-13 version, and to split what is needed for cuda and vulkan out, is at https://paste.c-net.org/b0eb0a98-a412-2ba5-f0fc-6209365fa761 - @Dominiquini, you may wish to take it, or to do something different.

I use Vulkan with ollama myself is how I know that the OLLAMA_VULKAN variable is needed.

0.23.1-1 cannot be built because there is no lib/ollama/mlx_cuda_13 directory in the .tar.zst.

$ tar -tavf ollama-linux-amd64.tar.zst
-rwxr-xr-x root/root  44668608 2026-05-05 12:52 bin/ollama
drwxr-xr-x root/root         0 2026-05-05 13:31 lib/ollama/cuda_v12/
lrwxrwxrwx root/root         0 2026-05-05 13:31 lib/ollama/cuda_v12/libcublas.so.12 -> libcublas.so.12.8.5.5
-rwxr-xr-x root/root 116388640 2015-07-07 18:00 lib/ollama/cuda_v12/libcublas.so.12.8.5.5
lrwxrwxrwx root/root         0 2026-05-05 13:31 lib/ollama/cuda_v12/libcublasLt.so.12 -> libcublasLt.so.12.8.5.5
-rwxr-xr-x root/root    728800 2015-07-07 18:00 lib/ollama/cuda_v12/libcudart.so.12.8.90
lrwxrwxrwx root/root         0 2026-05-05 13:31 lib/ollama/cuda_v12/libcudart.so.12 -> libcudart.so.12.8.90
-rwxr-xr-x root/root 1790154312 2026-05-05 13:31 lib/ollama/cuda_v12/libggml-cuda.so
-rwxr-xr-x root/root  751775824 2015-07-07 18:00 lib/ollama/cuda_v12/libcublasLt.so.12.8.5.5
drwxr-xr-x root/root          0 2026-05-05 13:02 lib/ollama/cuda_v13/
-rwxr-xr-x root/root  541607888 2015-07-07 18:00 lib/ollama/cuda_v13/libcublasLt.so.13.1.1.3
-rwxr-xr-x root/root     704288 2015-07-07 18:00 lib/ollama/cuda_v13/libcudart.so.13.0.96
-rwxr-xr-x root/root  402318952 2026-05-05 13:02 lib/ollama/cuda_v13/libggml-cuda.so
lrwxrwxrwx root/root          0 2026-05-05 13:02 lib/ollama/cuda_v13/libcublasLt.so.13 -> libcublasLt.so.13.1.1.3
lrwxrwxrwx root/root          0 2026-05-05 13:02 lib/ollama/cuda_v13/libcublas.so.13 -> libcublas.so.13.1.1.3
lrwxrwxrwx root/root          0 2026-05-05 13:02 lib/ollama/cuda_v13/libcudart.so.13 -> libcudart.so.13.0.96
-rwxr-xr-x root/root   54177976 2015-07-07 18:00 lib/ollama/cuda_v13/libcublas.so.13.1.1.3
lrwxrwxrwx root/root          0 2026-05-05 12:51 lib/ollama/libggml-base.so -> libggml-base.so.0
lrwxrwxrwx root/root          0 2026-05-05 12:51 lib/ollama/libggml-base.so.0 -> libggml-base.so.0.0.0
-rwxr-xr-x root/root     748152 2026-05-05 12:51 lib/ollama/libggml-base.so.0.0.0
-rwxr-xr-x root/root     873912 2026-05-05 12:51 lib/ollama/libggml-cpu-alderlake.so
-rwxr-xr-x root/root     873912 2026-05-05 12:51 lib/ollama/libggml-cpu-haswell.so
-rwxr-xr-x root/root    1009080 2026-05-05 12:51 lib/ollama/libggml-cpu-icelake.so
-rwxr-xr-x root/root     820728 2026-05-05 12:51 lib/ollama/libggml-cpu-sandybridge.so
-rwxr-xr-x root/root    1009080 2026-05-05 12:51 lib/ollama/libggml-cpu-skylakex.so
-rwxr-xr-x root/root     636536 2026-05-05 12:51 lib/ollama/libggml-cpu-sse42.so
-rwxr-xr-x root/root     632472 2026-05-05 12:51 lib/ollama/libggml-cpu-x64.so
drwxr-xr-x root/root          0 2026-05-05 12:58 lib/ollama/vulkan/
-rwxr-xr-x root/root   52228392 2026-05-05 12:58 lib/ollama/vulkan/libggml-vulkan.so
lrwxrwxrwx root/root          0 2026-05-05 12:58 lib/ollama/vulkan/libvulkan.so.1 -> libvulkan.so.1.4.321
-rwxr-xr-x root/root    4466776 2026-05-05 12:54 lib/ollama/vulkan/libvulkan.so.1.4.321

I'm testing a 'drive-by-patching' and as soon as I verify it works, I'll post it here.

error: unresolvable package conflicts detected error: failed to prepare transaction (conflicting dependencies) :: ollama-cuda12-bin-0.22.1-1 and ollama-cuda13-bin-0.22.1-1 are in conflict

@bazza: Have you installed the package "ollama-mlx-cuda13-bin"? Anyway, I won't be able to help you with this bug. My video card doesn't support CUDA13! If anyone else can test and let me know if the CUDA13 packages are working correctly, I'd be grateful.

I had this problem with flux 2 Klein

time=2026-03-18T11:49:04.112-03:00 level=WARN source=server.go:139 msg=mlx-runner msg="Error: failed to initialize MLX: libmlxc.so not found"

@Dominiquini, all is good here now. After deleting the ollama directories and reinstalling anew, the symlink is working and the NVIDIA GPU is being detected without modifying PKGBUILD. Thank you!

@ovflowd: Fixed! Thanks for the patch. The installation was broken on my machine and I hadn't noticed! I was just testing and the service seemed to be running without problems... I'll test it more thoroughly before the next updates!