Meta Unveils "Code Llama 70B" an Open-Source Coding Tool

Meta has unveiled its latest version of its Code Llama AI model named "Code Llama 70B". The model is being described as the largest and best-performing model in Meta’s arsenal. As it was launched in August, Code Llama tools have acted as a significant assistance for developers offering them a free platform for both research and commercial use. While the recent improvements in Code Llama 70B promise to bridge the gap with GPT-4.

It is noteworthy that the advancement of this model is that it has an increased capacity to handle more queries than its predecessors, which means that developers can now input more prompts while programming, leading to enhanced accuracy in code generation.

Code Llama 70B boasts an impressive 53 percent accuracy on the HumanEval benchmark while outperforming GPT-3.5’s 48.1 percent and approaching the 67 percent reported for GPT-4 in an OpenAI paper.

Code Llama facilitates the creation and debugging of code strings based on prompts and offers support for human-written work to developers. Meta had previously introduced two additional Code Llama tools

• Code Llama-Python
• Code Llama-Instruct

which are tailored for specific coding languages.

The availability of Code Llama 70B on three versions of the code generator is what sets it apart from its predecessors. It continues to cater to both research and commercial uses while remaining free. The model’s training on a massive 1TB of code and code-related data contributes to its prowess.

Code Llama 70B is hosted on the Hugging Face code repository that provides access to GPUs and enhances the efficiency of running AI models. And, Meta’s emphasis on larger models such as 34B and 70B underscores their commitment to delivering optimal results and improved coding assistance. The continued evolution of Code Llama reflects Meta’s dedication to stay at the forefront of AI innovation.

As the landscape of AI-driven code generation is evolving rapidly, even Meta is facing competition from other industry giants. In the meantime, Amazon introduced CodeWhisperer in April, while Microsoft collaborated with OpenAI to introduce GitHub Copilot.

These developments signal a broader trend of AI integration in coding practices and promise a transformative impact on software development.