llama.cpp

b931f81b - server : adjust spec tests to generate up to 16 tokens (#19093)