Introducing Code Gemma: Emma's Coding Companion


Prepare to be amazed as Emma embarks on her coding journey with the groundbreaking release of Code Gemma by Google DeepMind. This collection of specialized open code models is a game-changer for developers worldwide. Available in two sizes, 2B and 7B, with 8192 context, Code Gemma offers unparalleled support for on-device code completion. These models, initialized from Gemma Base and trained on a staggering 500B additional Tokens encompassing web, code, and math domains, are fine-tuned with Instruct using SFT and RLHF techniques. Achieving remarkable results, the 2B model attains 27% on HumanEval, while the 7B model reaches an impressive 52%. Moreover, commercial use is permitted, making Code Gemma an indispensable tool for businesses and individuals alike. Optimized for on-device usage, this revolutionary coding companion is now accessible on Hugging Face, empowering developers with unparalleled capabilities in code completion and innovation.