llama.cpp

e7f2f95c - ggml webgpu: Fix bug in dispatching large matrix-vector multiplication (#19535)