google/gemma-3-12b-it-qat-int4-unquantized

Gemma 3 12B IT QAT INT4 Unquantized is a 12 billion parameter instruction-tuned multimodal model from Google DeepMind, built from the same research as Gemini models. This version is optimized with Quantization Aware Training (QAT) to maintain bfloat16 quality while significantly reducing memory requirements for int4 quantization. It supports a 128K context window, handles both text and image inputs, and excels at text generation, image understanding, and reasoning tasks across over 140 languages.