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

Not All Disagreement Is Learnable: Token Teachability in On-Policy Distillation

About

On-policy distillation (OPD) trains a student on its own rollouts with token-level teacher supervision. Recent selective OPD methods exploit the non-uniformity of OPD signals by prioritizing high-entropy or high-disagreement tokens. We revisit this principle and ask: which token-level teacher signals are actually learnable? Using a fixed-context diagnostic that measures same-context teacher-student KL reduction, we show that raw KL disagreement is a coarse proxy for learning value. It conflates learnable disagreement, where the teacher assigns corrective mass to the student's top-K candidates, with incompatible disagreement, where the teacher places mass mostly off the student's current support. We formalize this local compatibility as token teachability and show that it better predicts fixed-context improvement than raw KL alone. Motivated by this finding, we propose Teachability-Aware OPD (TA-OPD), a lightweight token-position selection method that applies OPD loss to high-teachability positions without reward models or verifiers. Across Qwen2.5 and Qwen 3 teacher-student settings, TA-OPD often surpasses full-token OPD with only 5% retained tokens and improves over entropy- and divergence-based baselines. Our results reframe selective OPD as selecting learnable teacher signals rather than merely salient tokens.

Yuanyi Wang, Su Lu, Yanggan Gu, Pengkai Wang, Yifan Yang, Zhaoyi Yan, Congkai Xie, Jianmin Wu, Hongxia Yang• 2026

Related benchmarks

TaskDatasetResultRank
Scientific Question AnsweringGPQA Diamond
Accuracy47.47
123
Mathematical ReasoningMATH-500 1 (test)
Accuracy87.4
38
Code GenerationHumanEval v1 (test)
Accuracy79.88
37
Instruction FollowingIFEval v1 (test)
Accuracy78.93
28
Mathematical ReasoningAIME 2024 1
Accuracy30
24
Mathematical ReasoningAIME 2025 1
Accuracy25
24
Showing 6 of 6 rows

Other info

GitHub

Follow for update