Meta releases ‘Code Llama 70B’, an open-source behemoth to rival private AI development

Meta AI’s new Code Llama 70B model can generate code in various languages, such as Python, C++, Java, and PHP, faster and more accurately than ever before. It’s backed by 500 billion tokens of code and code-related data, making it a formidable force in the realm of artificial intelligence. But taming the complexity of code isn’t easy, and Code Llama 70B has been specially trained to understand and unleash the power of precise and rigid programming languages.

The variant titled CodeLlama-70B-Instruct scored high on the HumanEval benchmark and can handle an array of tasks, from sorting and searching to implementing algorithms like binary search and Fibonacci. CodeLlama-70B-Python, tailored for the popular programming language, which has been optimized for tasks ranging from web scraping to machine learning and web development. Code Llama 70B is available for free download, championing accessibility and democratizing the field of code generation.

In a statement on Facebook, Mark Zuckerberg, the CEO of Meta AI, announced the open sourcing of the new Code Llama and its 70B parameter model. This move catapults the importance of AI models in coding to the frontline and sets the stage for possibilities like code translation, documentation, and debugging, making it a game-changer in the software development industry.