Code Llama expands Llama 2 with code-focused LLMs (7B/13B/34B) and IDE integrations

AI Impact Summary

Code Llama introduces a family of code-specialized LLMs initialized from Llama 2, including 7B, 13B, and 34B variants with a Python specialist and an instruction-tuned flavor. The suite supports code infilling and long-context up to 100k tokens via RoPE scaling, and is exposed through Hugging Face Transformers with production-grade deployment options via Text Generation Inference and Inference Endpoints, plus IDE integration through a VS Code extension. Released under a permissive license on the Hugging Face Hub, Code Llama enables production-grade code generation and completion workflows across code bases, but teams should plan for licensing compliance and governance around generated code and training data provenance.