Recently, DeepSeek unveiled the open-source availability of its latest flagship model, V4, which is entirely compatible with Huawei's homegrown Ascend chips. This model enables full-chain native cooperation, spanning from low-level computational operations to communication protocols, memory management, and framework customization. This achievement signifies that domestic large-scale models and domestic computing capabilities have now entered the 'Day 0 Adaptation' stage, where the model is intricately integrated with the foundational computing power right from the pre-training phase. Through joint optimization efforts, it offers ready-to-use, high-performance capabilities, progressively diminishing dependence on NVIDIA.
