When loading the model (at least q4_k_m version) llama.cpp says "BOS token = 151643 '<|endoftext|>'", is that correct?
· Sign up or log in to comment