Code Llama, built on Llama 2, is an AI model specialized in code generation and discussion. It’s offered in three sizes: 7B, 13B, and 34B parameters. Trained on 500B tokens of code-related data, it supports popular languages like Python, C++, Java, and more.
Code Llama can generate code and natural language about code, aiding tasks like code completion and debugging. The 7B and 13B models feature fill-in-the-middle (FIM) capability for instant code completion.
Code Llama — Python is fine-tuned on 100B tokens of Python code, providing language-specific assistance. Code Llama — Instruct excels at understanding user prompts by training on natural language instructions and expected outputs. Designed for diverse sectors including research, industry, and open-source projects, Code Llama aims to accelerate coding tasks and inspire further innovation in AI-assisted coding tools.
