Meta continues to push its offerings for AI, with the latest AI tool it has to showcase being built for coders. Code Llama is a new AI coding tool built on top of Llama 2, and it’s available right now. The tool, which follows in the footsteps of Meta’s previous release of SeamlessM4T, will also be available for free both commercially and for research purposes.
This large language model works similarly to others we’ve seen make debuts over the past several months, including ChatGPT, which continues to be one of the biggest and most talked about large language models out there. However, that is where the similarities end, as Code Llama isn’t designed for general use. Instead, it was built specifically to cater to coders, hoping it will make their jobs more efficient as the LLM continues to be trained and improved.
Meta says its new AI coding tool is trained for code completion and debugging, and that it supports Python, C++, Java, Typescript (Javascript), PHP, C#, Bash, and even more. Further, Meta says that the new tool is available in three different sizes, known as 7B, 13B, and 34B. Additionally, every model is trained with 500B tokens of codes and code-related data.
The three different sizes are designed to address different latency and serving requirements, so it should be easy for coders to find one that works for them and easily focus on it. Meta also notes that Code Llama has been further fine-tuned to two additional variations: Code Llama – Python and Code Llama – Instruct.
If you are a coder, or if you’re just interested in messing around with the model itself, Meta has made it freely available to download. However, you will need to request access to the specific models that you want to use.