The rapid scaling of large language models (LLMs) has made communication overhead a critical bottleneck in distributed model-parallel training. Although exis…
机构:中科院
来源:arXiv 2611.04323 | AI4Papers 论文推荐平台