Meta has launched its own code-writing tool ‘Code Llama,’ that
it claims can make the task of code completion easier. Code Llama is developed off
of Meta’s larger language model ‘Llama 2’ and uses the same community license
as the Llama 2.
Free for research and commercial use, Code Llama is capable
of generating new code strings from prompts as well as debugging specific existing
code strings.
Meta has also developed other versions of Code Llama, including Code Llama-Instrct, which understands instructions in natural language, and Code Llama-Python, a specialized version of the tool.
Speaking about the importance of Code Llama in terms of its
usage, Meta says “the goal is to make developer workflows more efficient so
they can focus on the most human-centric aspects of their jobs.”
The tech giant also emphasizes that Code Llama has been
tested to show a better performance as compared to publicly available LLMs that
are already being used for creating and debugging codes. More precisely, Meta
reveals that Code Llama received a 53.7% score on the code benchmark HumanEval which
indicates the tool’s ability to accurately write code based on a text
description.
Meta plans to release Code Llama in three sizes, the smallest
of which can fit on a single GPU for more low-latency projects.
