llama.cpp

a02c7f58 - hexagon: handle multi-sequence in concat_2d (#29344)