SLAI introduces T-Rex, an end-to-end optimization framework for the full-parameter post-training of trillion-parameter-scale MoE models on Ascend NPU SuperPOD. Using the DeepSeek-V4 model family as the target workload, the system addresses memory pressure and communication overhead through hierarchical parallelism and kernel execution optimizations.

  • Achieves 34.22% Model FLOPs Utilization (MFU), representing a 2.93x improvement over the open-source baseline while maintaining training stability.
  • Establishes a Continuous Pre-training (CPT) and Supervised Fine-Tuning (SFT) workflow for complex Operations Research tasks using DeepSeek-V4-Flash.
  • Creates a dataset of 10K high-quality SFT samples combining domain resources with solver-verified synthetic optimization documents across four task categories.
  • The specialized model achieves a zero-shot Pass@1 score of 71.81%, outperforming GPT-5.4-Mini and the base DeepSeek-V4-Flash by 3.98 and 11.27 percentage points respectively.

This work demonstrates a full-stack pathway from efficient trillion-parameter model post-training on Ascend infrastructure to domain-specialized Flash models for solver-grounded mathematical modeling.