llama.cpp

bfd1f453 - ggml-webgpu: support non-square subgroup matrix configs for Intel GPUs (#21669)