The tech giant has made Code Llama available for research and commercial purposes, but the new AI model faces a competitive sector.
Meta is taking a leap into a submarket of AI, with a new model that is designed to generate and discuss code.
The tech giant claims its latest AI model – Code Llama – able to do this from simple text prompts, functioning as a coding assistant that can make workflows faster for developers and lower the entry barrier for people who are learning to code.
The new model is fine-tuned for coding and is built on top of Llama 2, the large language model that Meta released last month. Llama 2 is the successor to the company’s research-focused model released earlier this year.
The new code-focused model is being made available for research and commercial use, under the same community license as Llama 2.
“Code Llama has the potential to be used as a productivity and educational tool to help programmers write more robust, well-documented software,” Meta said.
The new model comes in three sizes to serve different uses. These models are built on 7bn, 13bn and 34bn parameters respectively. Meta said each of these models is trained on 500bn tokens of code and code-related data.
“The three models address different serving and latency requirements,” Meta said in a blogpost. “The 7bn model, for example, can be served on a single GPU.
“The 34bn model returns the best results and allows for better coding assistance, but the smaller 7bn and 13bn models are faster and more suitable for tasks that require low latency, like real-time code completion.”
Too many AI assistants
Meta is arriving late to this sector, as generative AI is already being used by many developers to assist in coding, according to various competitors in the market.
Other tech giants have had their own AI coding assistants for some time now, such as Microsoft’s GitHub Copilot. By June 2022, GitHub claimed this AI tool was being used to write nearly 40pc of code on the platform.
In March, Alphabet partnered with AI start-up Replit to help developers edit code using generative AI. Meanwhile, IBM recently revealed its own AI model that is designed to translate old COBOL code to Java and enhance developer productivity.
The sector has proven challenging for some companies to thrive in, however. Kite, a start-up that had been developing AI to help developers write code for nearly a decade, shut down its business last year.
At the time, the company’s founder Adam Smith claimed that even state-of-the-art machine learning models don’t understand the structure of code – and too few developers are willing to pay for available services.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.