Counterfactual Relevance for On-Policy Distillation is introduced, which operationalizes task relevance through a paraphrase-calibrated counterfactual sensitivity margin and establishes CROP as a model-internal, contrast-specific method for allocating token-level supervision.
Abstract
On-policy distillation (OPD) supervises a student language model on trajectories sampled from its current policy, but assigns equal credit to response tokens with unequal supervision value. Selective OPD addresses this limitation by allocating supervision non-uniformly across response tokens according to their estimated training value. Most existing criteria, however, focus primarily on optimization need, such as uncertainty or teacher-student disagreement, while task relevance, namely whether the supervision is tied to the semantic content of the current input, remains less directly characterized as a complementary dimension. To address this gap, we introduce Counterfactual Relevance for On-Policy Distillation (CROP), which operationalizes task relevance through a paraphrase-calibrated counterfactual sensitivity margin. For each source prompt, CROP constructs a validated original-paraphrase-counterfactual triplet, holds the student rollout fixed, and measures each response position by its sensitivity to a task-relevant condition change calibrated by its sensitivity to a meaning-preserving rewrite. Matched selection controls show that CROP identifies more useful supervision positions than random or lowest-relevance selection, while component comparisons confirm the value of both counterfactual sensitivity and paraphrase calibration. Across two teacher-student settings, CROP improves aggregate performance by 1.92 and 2.96 points over the strongest non-CROP selector. These results support task relevance as a complementary criterion for selective OPD and establish CROP as a model-internal, contrast-specific method for allocating token-level supervision.
A unified view is developed in which the log-probability ratio between a proxy expert and a proxy reference defines an implicit reward for KL-regularized policy optimization, which enables any-to-any transfer across models that differ in scale, training stage or generation.
Rong Wu, Dao-Cheng Fu, Yu Yang et al.· 0 citations
On-policy post-training can improve large language models by learning from their own trajectories, but requires costly rollout generation. We ask whether fixed demonstrations can support competitive off-policy learning through better supervision. Our premise is that their usefulness depends not only on the training tra...
Tian-Le Wang, Jia-Yu Liu, Rui-Zhi Zhao et al.· 0 citations
This work identifies Teacher Uncertainty Contraction (TUC), a systematic phenomenon whereby the teacher's predictive uncertainty decreases as it continues from a student-generated prefix, and proposes Adaptive-Continuations On-Policy Distillation (AC-OPD), which augments informative states along student rollouts with t...
Jin-Gang Zhou, Yu-Yi Zhou, Hai-Yang Guo et al.· 3 citations
Experiments spanning mathematical reasoning, multi-domain STEM, code generation, and multi-turn agentic tasks show that RISE outperforms RLVR-only training and on-policy self-distillation across all settings.
Yang Li, Semih Yavuz, Shafiq Joty· 5 citations· ⚡1
DiffGate is introduced, an outcome-gated objective that combines GRPO with selective, bounded teacher guidance, and improves pass@8 across all four model--domain settings, demonstrating improved solution coverage under the evaluation protocol.
Karn Tiwari, V. Chordia, P. PrathoshA· 0 citations
Group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance, and demonstrates that group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance.
Zhu Zhang, Ji-Xun Wang, Xiao-An Xu et al.· 3 citations