DeepSeek-V3.2-Exp model officially released and open sourced

The model introduces a sparse Attention architecture, which can effectively reduce computing resource consumption and improve model reasoning efficiency.