Gemma 2 2B IT

Gemma 2 2B IT is Google's instruction-tuned variant of the 2.6 billion parameter Gemma 2 base model, built from the research and technology behind the Gemini family. As a decoder-only transformer using Grouped-Query Attention (GQA), rotary positional embeddings (RoPE), and GeGLU activations, it balances capability with efficiency for deployment on laptops, desktops, and modest cloud infrastructure. The model is fine-tuned through supervised instruction-following and reinforcement learning from human feedback (RLHF) to align outputs with human preferences. It handles question answering, summarization, and reasoning tasks while keeping a minimal resource footprint, within an 8K token context. With open weights freely available, Gemma 2 2B IT lets developers fine-tune and deploy their own specialized versions without relying on a hosted API, which suits edge deployment and privacy-sensitive applications.