Facebook parent company Meta has unveiled its latest innovation in artificial intelligence (AI), Code Llama. Meta’s latest large language model (LLM) is poised to change how code is written, understood, and debugged.
CodeLLama includes three variants: the base model, a fine-tuned version for Python, and an Instagram instruction-following model. Each model has versions with 7 billion, 13 billion, and 34 billion parameters, suitable for both research and commercial applications.
CodeLLama is derived from the LLama2 training and is specialized for code. Meta trained it on specific code datasets, allowing CodeLLama to generate popular code languages such as Python, C++, Java, and PHP based on natural language prompts.
In benchmark testing, Code Llama has demonstrated remarkable performance. Scoring 53.7% on HumanEval and 56.2% on Mostly Basic Python Programming (MBPP), it surpassed other open-source code-specific LLMs and matched the performance of models like ChatGPT.
Code Llama’s introduction could likely impact everyone in coding, from those entering the field to developers with decades of experience.
This new open-source LLM has the potential to streamline coding processes, aid in debugging, and enhance productivity, aligning with the growing demands of the tech industry.