This AI Paper from China Unveils ‘Vary-toy’: A Groundbreaking Compact Large Vision Language Model for Standard GPUs with Advanced Vision Vocabulary