Based on its Llama 2 big language concept, Meta has produced a tool called Code Llama. The new program will produce fresh code and fix human-written mistakes. Code Llama is open source and free for commercial and academic use under the same community license as Llama 2.
The new tool can complete and debug code when directed to a specific code string or build strings of code from prompts. The business released Code Llama-Python, a version tailored specifically for Python in addition to the main Code Llama paradigm. Another kind, known as Code Llama-Instrct, can comprehend instructions given in normal language. Each distinct variation of Code Llama is incompatible, claims Meta. For natural language instructions, the business does not advise using the base Code Llama or Code Llama-Python.
Meta Launches Own AI Code-Writing Tool: Code Llama
In a blog post, Meta stated that “programmers are already using LLMs to help in a variety of tasks, ranging from writing new software to debugging existing code.” The objective is to improve developer processes so they may concentrate on the duties that are most centered on people.
ad claims Based on benchmark testing, Code Llama outperformed publicly accessible LLMs, however it did not specify which models it tested against. According to the firm, Code Llama achieved a score of 53.7% on the code benchmark HumanEval. Additionally, it was competent at appropriately writing code from a written description.
Code Llama will be offered in three sizes by the company. One GPU can accommodate its smallest size, which enables more low-latency projects.