Meta launches new AI programming assistant Code Llama 70B

Meta has released Code Llama 70B, another AI-powered programming tool. More powerful than the previous version introduced in August, The update supports more prompts and delivers more accurate results.

Code Llama is built on the Llama 2 model and works as an assistant to developers. Tool It helps build projects by interpreting prompts and evaluating code quality in real time.

The new Code Llama 70B is available in three versions; all free for researchers and commercial use: Code Llama 70B, Code Llama 70B Python, and Code Llama 70B Instruction 70B. Each meets a different need.

In the announcement, Meta noted that Code Llama 70B achieved 53% accuracy in the HumanEval benchmark, beating rival GPT-3.5 (48.1%) and coming close to GPT-4 (67%).

“Writing and editing code has emerged as one of the most important uses of AI models today. The ability to code has also proven important for AI models to process information in other domains more rigorously and logically,” said Meta’s CEO. , Mark Zuckerberg is on Facebook.

Comprehensive training

According to Meta, the new model was trained with more than 1TB of code or programming-related data. The tool is currently hosted in the Hugging Face repository.

Now Code Llama 70B competes with Amazon’s CodeWhisperer and Microsoft’s GitHub Copilot powered by OpenAI models, intensifying competition in the AI ​​programming assistant industry.