Google Unveils Gemma 3: The "Ultimate Single-GPU Powerhouse"
2025-03-13 / Read about 0 minute
Author:小编   

On March 12th, Google introduced Gemma 3, marking a pioneering integration of multimodal functionality that seamlessly supports visual-language input and text output. This advanced model boasts the ability to process a contextual window of up to 128,000 tokens and comprehend over 140 languages. Offering flexibility, Gemma 3 comes in four parameter sizes: 1B, 4B, 12B, and 27B, including pre-trained models that can be fine-tuned for tailored scenarios and a general instruction-optimized version. Google asserts that Gemma 3 stands as the most potent model capable of running on a single GPU or TPU, setting a new benchmark in single-device performance.