Mind the Gap: A Closer Look at Tokenization for Multiple-Choice Question Answering with LLMs
This work addresses a subtle but impactful issue in LLM evaluation for researchers, revealing that seemingly trivial tokenization choices can undermine the reliability of model comparisons, making it an incremental but important contribution.
The paper investigates how tokenization of the space after 'Answer:' in multiple-choice question answering prompts affects LLM accuracy, finding differences up to 11% and recommending a specific tokenization strategy that improves performance and calibration.
When evaluating large language models (LLMs) with multiple-choice question answering (MCQA), it is common to end the prompt with the string "Answer:" to facilitate automated answer extraction via next-token probabilities. However, there is no consensus on how to tokenize the space following the colon, often overlooked as a trivial choice. In this paper, we uncover accuracy differences of up to 11% due to this (seemingly irrelevant) tokenization variation as well as reshuffled model rankings, raising concerns about the reliability of LLM comparisons in prior work. Surprisingly, we are able to recommend one specific strategy -- tokenizing the space together with the answer letter -- as we observe consistent and statistically significant performance improvements. Additionally, it improves model calibration, enhancing the reliability of the model's confidence estimates. Our findings underscore the importance of careful evaluation design and highlight the need for standardized, transparent evaluation protocols to ensure reliable and comparable results.