Meta Platforms Inc. today announced a new and enhanced version of its code-generating artificial intelligence model, Code Llama, which comes with increased processing power, greater accuracy and support for specific programming languages, such as Python, C++, Java and PHP.

Code Llama 70B is a derivative of Meta’s open-source Llama 2 large language model specifically designed to create code based on natural language prompts. Code generation is often held up as one of the most viable use cases of all for generative AI. By helping to mass-produce code, generative AI can make software development more efficient and accessible, while enabling developers to get more creative, proponents say.

The problem with code generation is that the code must be extremely precise and rigid, following strict rules and syntax, and it must also be predictable and behave as the user expects it. But software code can often be extremely long and complicated, and that means AI models must have tons of context and logic to understand and generate it. To achieve that, AI models must have access to lots of computing power, and this is where Code Llama 70B stands out.

As one of the largest-ever open-source code generating models, Code Llama 70B is likely to become one of the most important benchmarks in this particular subset of AI. It has a lot of promise, because many existing code-generating AIs have struggled to automate software development in a way that’s reliable and safe.

Meta said Code Llama 70B was trained on more than 500 billion tokens of code and related data, which means it is far more capable and robust than earlier iterations of the model. It also benefits from having a larger context window of 100,000 tokens, enabling it to generate more complex code.

Holger Mueller of Constellation Research Inc. told SiliconANGLE that code generating models are especially interesting because many enterprises are struggling to recruit enough developers to build all of the software they need. “Meta’s Code Llama 70B promises to make the idea of autonomous software operations or ASO a reality, leading us into a world where software writes software,” the analyst said. “It makes sense for Meta to pursue this idea as it needs to create a lot of code for its own projects, and Code Llama brings us closer to the day where 100% of enterprise-ready software may be written by generative AI.”

Meta Chief Executive Mark Zuckerberg announced the new model personally in a public post on Facebook. “Writing and editing code has emerged as one of the most important uses of AI models today,” he said. “The ability to code has also proven to be important for AI models to process information in other domains more rigorously and logically. I’m proud of the progress here, and looking forward to including these advances in Llama 3 and future models as well.”

The company is making several variants of Code Llama 70B available to the public, including one known as CodeLlama-70B-Instruct, which has been fine tuned to handle code requests in natural language. Meanwhile, CodeLlama-70B-Python is a version that has been optimized to generate code in the Python language only. It has been boosted with an additional 100 billion tokens based on Python code to ensure it’s more accurate and fluent in that language.

The Code Llama 70B models are available to download now via a range of platforms, including Hugging Face, PyTorch, Jupyter Notebook and TensorFlow.

Image: Meta