Meta has released an open-source code-generating AI model called Code Llama. This machine learning system can generate and explain code in natural language, specifically English. Code Llama can complete and debug code in various programming languages, including Python, C++, Java, and more. Meta believes that open-source code models like Code Llama benefit innovation and safety in the AI coding space. Code Llama is based on the Llama 2 text-generating model but has been trained specifically to understand code. It has been fine-tuned using a large dataset of code-related data. Code Llama models range in size from 7 billion to 34 billion parameters, with the 34 billion-parameter model being the best-performing and largest code generator open sourced to date. However, there are risks associated with code-generating AI tools, such as security vulnerabilities and potential copyright infringement. Code Llama has limitations and can generate inaccurate or objectionable responses to prompts. Developers are advised to perform safety testing and tuning before deploying Code Llama. Despite the risks, Meta has minimal restrictions on how developers can use Code Llama and hopes it will inspire the creation of new innovative tools in the coding space.
