Code Llama is a family of large language models for code that provide state-of-the-art performance, infilling capabilities, support for large input contexts, and instruction following ability. The models are released under a permissive license.