llama.cpp

d6129011 - perplexity: add proper batching (#19661)