>>105682731
my understanding is that they're no longer officially building off of llama.cpp but they still use a significant amount of llama.cpp code and are fundamentally building off of ggml