Suite of Qwen 3 models, compressed using sparsification and 4-bit floating point precision for low-resource deployment.