Meta, the parent company of Facebook, has recently introduced a new tool called Code Llama. This tool is built on top of Meta’s Llama 2 large language model and aims to assist developers in generating new code and debugging their human-written work. The company has made Code Llama available under the same community license as Llama 2, which means it can be used for both research and commercial purposes free of charge.
Code Llama utilizes the power of the Llama 2 model to create code strings based on prompts provided by developers. It can also help complete and debug specific code strings when given the appropriate instructions. To cater to different needs, Meta has released multiple versions of Code Llama. One of these versions is Code Llama-Python, which is specialized for Python programming. Another version, called Code Llama-Instrct, is designed to understand instructions given in natural language. It is worth noting that each specific version of Code Llama is tailored for its intended use and is not interchangeable. Meta does not recommend using the base Code Llama or Code Llama-Python for processing natural language instructions.
In a blog post, Meta emphasized that programmers are already leveraging large language models, such as LLMs, to assist them in various tasks, including writing new software and debugging existing code. The goal of Code Llama, according to Meta, is to enhance developer workflows, making them more efficient and allowing developers to focus on the more human-centric aspects of their jobs.
Meta claims that Code Llama outperforms publicly available LLMs based on benchmark testing. However, the company did not specify which models Code Llama was compared against. According to Meta, Code Llama achieved a score of 53.7 percent on the code benchmark HumanEval, demonstrating its ability to accurately write code based on a given text description.
To cater to different computing requirements, Meta has introduced three different sizes of Code Llama. The smallest size is designed to fit on a single GPU, enabling low-latency projects to be executed more efficiently.
Code generators have become increasingly prevalent in the developer community, with various companies introducing tools to streamline the coding process. For instance, GitHub launched Copilot, which is powered by OpenAI’s GPT-4, to assist developers in rapidly generating and checking code. Copilot is even capable of rewriting old code to bring it up to date. Likewise, Amazon’s AWS offers a code-writing tool called CodeWhisperer, which can generate code, perform checks, and update existing code. Google has also been working on its own code-writing tool, known as AlphaCode, developed by its subsidiary DeepMind, although it has not been released to the public yet.
In summary, Meta’s Code Llama tool built on the Llama 2 model aims to revolutionize the coding process by providing developers with an efficient way to generate code and debug their work. With its specialized versions for different programming languages and ability to understand natural language instructions, Code Llama offers a versatile solution for various development tasks. As the field of code generation continues to grow, developers can expect more innovative tools like Code Llama to enhance their productivity and streamline their workflows.