Add proper pre-tokenization fixed BPE version quantized with Q4_K_M

Copy aux files from Llama3 main repo

initial commit