Meta, the parent company of Facebook, has introduced an advanced large language model driven by artificial intelligence. This model can utilise text prompts to generate and engage in discussions about code. The tool, known as Code Llama, is built upon the foundation of the Llama 2 large language model. Its primary purpose is to generate new code and assist in debugging human-written code, as announced by the company.
Code Llama shares the same community license as Llama 2, and it is available for both research and commercial use at no cost.
Empowering coders and developers
Meta explained that Code Llama has the capability to generate code snippets based on prompts and can also debug code when directed to a specific code string. In addition to the core Code Llama model, Meta has unveiled specialised versions: Code Llama-Python and Code Llama-Instrct. The latter can comprehend instructions conveyed in natural language. It’s important to note that Meta advises against using the base Code Llama or Code Llama-Python for processing natural language instructions, as each version serves distinct purposes.
According to Meta, programmers are already integrating large language models into their workflows for tasks such as software development and code debugging. The overarching objective is to streamline developer workflows, allowing them to concentrate on the more human-centric aspects of their roles.
Meta noted that Code Llama outperformed publicly available large language models in benchmark testing, although specific models were not mentioned. The company plans to offer three different sizes of Code Llama, with the smallest size designed to fit on a single GPU for projects requiring low-latency processing.
Code Llama has support for numerous popular programming languages currently in use by developers, including Python, C++, Java, PHP, Typescript (JavaScript), C#, and Bash.
Globally, investments in AI are projected to reach $200 billion by 2025 and have the potential to significantly impact gross domestic product, according to Goldman Sachs Economic Research.
A recent study by McKinsey suggests that generative AI could contribute nearly $4.4 trillion annually to the global economy and revolutionise productivity across various sectors, with continued investment in the technology.
Programmers are increasingly relying on large language models to streamline their work, reducing repetitive tasks and enhancing efficiency.
Meta designed Code Llama to cater to software engineers across a wide range of sectors, including research, industry, open source projects, non-governmental organisations (NGOs), and various businesses.
Earlier this week, Meta also launched an AI-powered translation engine capable of translating languages in both text and speech.
