This repo releasing the code and benchmark datasets for paper "Token Alignment via Character Matching for Subword Completion" in ACL Findings 2024. In our paper, we noticed LLMs usually generate ...