Meta Releases Code Llama, Powerful Foundation Models for Coding

Meta Releases Code Llama, Powerful Foundation Models for Coding
Image Credit: Meta AI

Meta has open-sourced a new large language model called Code Llama that promises to be a game-changer for software developers. Code Llama is capable of generating code from text prompts, completing code, and even debugging programs.

The company revealed that Code Llama outperformed other publicly available AI coding assistants in benchmark testing. When tested on popular datasets like HumanEval and Mostly Basic Python Programming, Code Llama scored higher than predecessors like GitHub Copilot.

 Benchmark testing showed that Code Llama performed better than open-source, code-specific LLMs and outperformed Llama 2

Code Llama builds on Meta's Llama 2 model, and is being released under the same community license. This code-specific fine-tuning allows Code Llama to not only generate code, but also comprehend and describe code using natural language.

Meta is providing Code Llama in three model sizes - 7B, 13B, and 34B parameters - to accommodate different latency and serving requirements. The company also trained specialized variations focused on Python and understanding natural language instructions:

  1. Code Llama: The fundamental code model.
  2. Code Llama - Python: Tailored for Python, recognizing its pivotal role in the AI landscape.
  3. Code Llama - Instruct: Refined to comprehend natural language directives with precision.

According to Meta, Code Llama's larger model sizes and input lengths enable more advanced applications like code completion across lengthy codebases and debugging complex scenarios. The AI assistant can handle up to 100,000 tokens of context, significantly more than typical large language models.

Meta says it undertook extensive safety testing and red team penetration attempts before releasing Code Llama publicly. The company found Code Llama tended to provide safer code responses compared to models like GPT-3.5 Turbo.

By open-sourcing Code Llama, Meta aims to advance AI innovation in a responsible manner. The company believes publicly accessible models like Code Llama will lead to new beneficial applications in coding and software development.

Code Llama has the potential to boost productivity for professional developers while also making coding more accessible for novices. But Meta acknowledges cutting-edge AI still poses risks, and urges researchers to continue advancing the responsible development of AI systems.

The release of Code Llama represents an exciting new milestone in AI assistance for programming. With its state-of-the-art capabilities, Code Llama may soon become an invaluable tool for coders across software teams, classrooms, and hobbyist communities.

Chris McKay is the founder and chief editor of Maginative. His thought leadership in AI literacy and strategic AI adoption has been recognized by top academic institutions, media, and global brands.

Let’s stay in touch. Get the latest AI news from Maginative in your inbox.