CodeLlama was trained by Meta through continued pre-training and instruction fine-tuning of Llama 2 in the code domain, offered in multiple parameter scales including 7B, 13B, and 34B.