CodeGemma - an official Google release for code LLMs

Hugging Face · 2 years ago · 7

Google released CodeGemma, a family of open-access code-specialized language models available in three versions: a 2B model for code infilling, a 7B base model for code and natural language tasks, and a 7B instruction-tuned model for code conversations. The CodeGemma 7B model achieved 40.13 on HumanEval for Python code generation and 43.06 on JavaScript, using 500 billion tokens of additional training beyond the base Gemma checkpoints. The models are integrated into Hugging Face transformers, Google Cloud, and Inference Endpoints for deployment and use across various platforms.