[sp332]

It does say this: Note that all Code Llama models were initialized with Llama 2 weights before they were further trained on code.