Code LLaMA 2 70B, a ‘free’ source code generating model beating GPT-4

Meta has unveiled Code LLaMA 70B, an advanced AI model designed to generate programming code. Building upon its predecessor, LLaMA 2, Code LLaMA has been enhanced with specialized datasets to improve its coding capabilities. The model offers three versions: a fundamental code generator, a Python-specific version, and an instruction-tuned variant, the latter achieving a notable 67.8 on the HumanEval index, slightly outperforming GPT-4 in code generation.

Introduced in August 2023, Code LLaMA is positioned as one of the top-performing open models currently available, though its ‘open’ designation is debated due to usage restrictions imposed on large companies. Despite this, individual programmers and startups can freely access and integrate the model into their tools, including commercial applications, by submitting a request form.

Meta’s CEO, Mark Zuckerberg, has expressed excitement over the new model, which boasts 70 billion parameters, highlighting the significance of AI in coding and editing tasks. This development represents a significant step forward for AI-assisted programming, offering a powerful tool for developers worldwide.
Read more at Gearrice…