@taronaeo Will this also fix the Docker build?
https://github.com/ggml-org/llama.cpp/actions/runs/18962560883/job/54152829325

@taronaeo Will this also fix the Docker build? https://github.com/ggml-org/llama.cpp/actions/runs/18962560883/job/54152829325

Ooh, looks like it broke for some reason. But yes, this PR + the previous PR (#16664) should fix this. Let me double check tomorrow when I have more time :)

@rishiraj20 Can you assist to test this PR on AQLINUX1 and 2?

1. Build this PR using:

$ cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON

$ cmake --build build --config Release -t llama-cli -j$(nproc)

2. Check that there are 2 libggml-cpu-z*.so files built:

$ ls -la build/bin | grep libggml-cpu

-rwxr-xr-x.  1 root root 1167608 Nov  1 19:04 libggml-cpu-z15.so
-rwxr-xr-x.  1 root root 1167608 Nov  1 19:04 libggml-cpu-z16.so

3. Run a test prompt and let me know which library is loaded via:

$ build/bin/llama-cli -m /opt/hf_models/granite-3.3-2b-instruct-be.Q4_K_M.gguf -no-cnv --seed 42 -n 50 -p "Write me a dog walking business idea 1. " 2>&1 | less

Help me paste the first few outputs from the top. It should print something like this at the top and it should run the prompt completely without problems.

load_backend: loaded CPU backend from /opt/llama.cpp/build/bin/libggml-cpu-z16.so
build: 6819 (b62c93efc) with cc (GCC) 15.1.1 20250521 (Red Hat 15.1.1-2) for s390x-redhat-linux
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_loader: loaded meta data with 40 key-value pairs and 362 tensors from /opt/hf_models/granite-3.3-2b-instruct-be.Q4_K_M.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.

...

None of the CI failures look related to this PR. Merging in a few hours unless the CI failures are related.

None of the CI failures look related to this PR. Merging in a few hours unless the CI failures are related.

Indeed unrelated, go ahead.