CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation

CoTranslator turns reverse chain-of-thought into verified forward traces, cutting distillation compute while keeping most math accuracy.

CoTranslator learns to map reverse chain-of-thought responses, which condition on a known answer, toward the distribution of verified forward responses used for reasoning distillation. Trained on a small paired set, it then supports large-scale generation of student training data. Across six student models and six mathematics benchmarks, it improves average accuracy over semantic suppression by up to 5.82 percentage points. For two 4B students it retains 95.0% of rejection sampling's average accuracy while using 35.9% of the estimated compute, and it transfers to new math data, another teacher, and STEM problems.