Achieving Tokenizer Flexibility in Language Models through Heuristic Adaptation and Supertoken Learning Paper • 2505.09738 • Published May 14 • 9 • 2