1
InfiGFusion: Graph-on-Logits Distillation via Efficient Gromov-Wasserstein for Model Fusion
提出一种基于高效Gromov-Wasserstein的图对数蒸馏方法,解决异构LLM模型融合中的词汇维度独立问题,提升推理效率与性能。
arXiv:2505.13893v2 Announce Type: replace Abstract: Recent advances in large language models (LLMs) have intensified efforts to fuse heterogeneous ope…