This repository provides the official PyTorch implementation of Gemma models, Google’s family of lightweight open LLMs built from research and technology used to create Gemini. It includes model and inference implementations using both PyTorch and PyTorch/XLA, supporting multiple model versions (Gemma v1, v2, v3) and variants including text-only, multimodal, and instruction-tuned checkpoints. The implementation supports running inference on CPU, GPU, and TPU hardware.