llama.cpp

fix: prevent segfault in tokenizer on highly repetitive input

#17786

Merged