DeepSeek has recently launched and open-sourced DeepSeek V3 and DeepSeek R1, which have generated significant attention both domestically and internationally. Its outstanding model quality and extremely low training costs have attracted widespread attention and indirectly contributed to a sharp drop in Nvidia's market capitalization. DeepSeek V3 can be considered the most powerful open-source base model to date, and it has achieved performance on par with leading closed-source models (including GPT-4 and