July 23, 2026 SLAI T-Rex: Full-Parameter Post-training of the DeepSeek-V4 Family on Ascend SuperPOD 在华为昇腾 NPU SuperPOD 上对万亿参数 MoE 模型（DeepSeek-V4）进行全参数后训练的系统优化与面向运筹学的 CPT-SFT 工作流 llm-training system-optimization ascend-npu sparse-attention kernel-optimization post-training