Knowledge Fusion of Large Language Models

Multi-Level Optimal Transport for Universal Cross-Tokenizer Knowledge Distillation on Language Models