llama.cpp

kv-cache : use ggml_set_rows

#14285

Merged