Exact Byte-Level Probabilities from Tokenized Language Models for FIM-Tasks and Model Ensembles
This addresses performance degradation in fill-in-the-middle tasks and integration issues in model ensembles for NLP practitioners, offering a zero-shot solution to tokenization-related shortcomings.
The paper tackles the problem of tokenization bias in language models, where tokenized models produce different predictive distributions than byte-level models even when statistically equivalent, and introduces a method to convert tokenized LMs into token-free ones without training, achieving an 18% improvement in FIM coding benchmarks and up to 3.7% better performance in model ensembles.
Tokenization is associated with many poorly understood shortcomings in language models (LMs), yet remains an important component for long sequence scaling purposes. This work studies how tokenization impacts model performance by analyzing and comparing the stochastic behavior of tokenized models with their byte-level, or token-free, counterparts. We discover that, even when the two models are statistically equivalent, their predictive distributions over the next byte can be substantially different, a phenomenon we term as ``tokenization bias''. To fully characterize this phenomenon, we introduce the Byte-Token Representation Lemma, a framework that establishes a mapping between the learned token distribution and its equivalent byte-level distribution. From this result, we develop a next-byte sampling algorithm that eliminates tokenization bias without requiring further training or optimization. In other words, this enables zero-shot conversion of tokenized LMs into statistically equivalent token-free ones. We demonstrate its broad applicability with two use cases: fill-in-the-middle (FIM) tasks and model ensembles. In FIM tasks where input prompts may terminate mid-token, leading to out-of-distribution tokenization, our method mitigates performance degradation and achieves 18% improvement in FIM coding benchmarks, while consistently outperforming the standard token healing fix. For model ensembles where each model employs a distinct vocabulary, our approach enables seamless integration, resulting in improved performance up to 3.7% over individual models across various standard baselines in reasoning, knowledge, and coding. Code is available at: https://github.com/facebookresearch/Exact-Byte-Level-Probabilities-from-Tokenized-LMs