llama.cpp

5d5306bf - tests : initialize all tensors in test_dsv4_hc to avoid NaNs in sentinel tensors (#25822)