Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Breaking the Tokenizer Barrier: On-Policy Distillation across Model Families

About

On-Policy Distillation (OPD) has become a core technique in the post-training of Large Language Models (LLMs) for transferring knowledge from domain experts to student models. However, existing OPD distillation methods require teacher and student models to share the same tokenizer, restricting the applicability of OPD within the model series. Current mainstream practice typically employs Supervised Fine-Tuning (SFT) on teacher-generated responses for cross-tokenizer distillation, which fails to capture the rich knowledge embedded in the teacher's probability distribution. In this work, we enable the standard on-policy distillation method to operate across model families, ensuring that high-fidelity token-level signals can propagate across different tokenizers with a precise token-mapping algorithm. Extensive experiments show that cross-tokenizer OPD is significantly more compute-efficient than baselines on various benchmarks. Our results unlock a broader range of teacher-student pairs for OPD, opening up new avenues for adapting and enhancing interactions between LLMs.

Yifan Niu, Han Xiao, Dongyi Liu, Zelong Wang, Dihong Gong, Yasheng Wang, Jia Li• 2026

Related benchmarks

TaskDatasetResultRank
Scientific Question AnsweringGPQA Diamond
Accuracy48.1
131
Mathematical ReasoningMATH 500
Pass@1 Rate89
113
Code GenerationLiveCodeBench
Accuracy25.1
28
General ReasoningAggregate (AIME, MATH-500, GPQA-D, LCB)
Average Score52
8
Mathematical ReasoningAIME 2024
Exact Match Accuracy56.7
8
Mathematical ReasoningAIME 2025
Exact Match Accuracy43.3
8
Mathematical ReasoningAIME 2026
Exact Match Accuracy50.8
8
Showing 7 of 7 rows

Other info

Follow for update