Forcing a language model to output only the allowed answer-letter tokens, with a tuned penalty, recovers accuracy lost to a single extra space in multiple-choice prompts.
PRD etect: Perturbation-robust LLM -generated text detection based on syntax tree
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Token Constraint Decoding Improves Robustness on Question Answering for Large Language Models
Forcing a language model to output only the allowed answer-letter tokens, with a tuned penalty, recovers accuracy lost to a single extra space in multiple-choice prompts.