LLMEmu: Execution-Driven Emulator for High-Fidelity Distributed LLM Training
Siyuan Yang, Pingjing Lu, Enda Yu, Dezun Dong · 2025
Transformer-based large models, with trillions of parameters and massive datasets, have driven breakthroughs in NLP, vision, and multimodal tasks. However, their rapid growth poses substantial challenges for training within limited GPU resources, making distributed training indispensable.