llama.cpp

536bea4c - server : separate the notion of position and KV tokens, remove prompt truncation