unsloth/gemma-3-4b-it-qat

The unsloth/gemma-3-4b-it-qat model is a 4.3 billion parameter instruction-tuned variant of Google DeepMind's Gemma 3 family, utilizing Quantization Aware Training (QAT). This multimodal model processes text and image inputs (896x896 resolution, 256 tokens each) with a 128K context window and generates text outputs. It excels in diverse text generation and image understanding tasks, including question answering, summarization, and reasoning, while being optimized for deployment in resource-limited environments.