Code Llama: Release of Llama 2–based code models (7B/13B/34B) with 16k-100k context and HF integration

AI Impact Summary

Code Llama releases a family of open-access, code-specialized models derived from Llama 2 (7B/13B/34B) with a Python specialist and an instruction-tuned variant. It introduces extended context support (16k native, up to 100k with long-context fine-tuning) and code infilling for the 7B/13B variants, enabling more capable code completion and editing workflows in IDEs. The rollout includes integration with Hugging Face Transformers, Text Generation Inference, Inference Endpoints, and a VS Code extension, signaling production-ready deployment paths, but infilling is not available on the 34B or Python-specialist variants. This expands production-ready developer tooling and automation opportunities for code generation, documentation, and unit tests, while requiring alignment with the licensing and deployment tooling in the HF ecosystem.