The paper investigates how teacher signals influence parameter updates in Multi‑Teacher On‑Policy Distillation (MOPD) by analyzing Qwen3‑1.7B and SmolLM3‑3B. It shows that loss averaging, Adam’s first‑moment bias, BF16 rounding, and the choice of averaging rule all shape the gradients and ultimately affect task performance. The study quantifies these effects, revealing, for example, that token‑averaging favors longer responses and that BF16 rounding masks most weight changes.
Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63. 0-68.
The paper investigates how the sampled-token reverse-KL loss in on‑policy distillation distributes updates across tokens. By analyzing the gradient of the per‑token K2 estimator, the authors find that tokens with low student probability and large teacher‑student gaps receive disproportionately large gradient norms. They propose Surprise‑aware Reweighting (SuRe), a lightweight weighting rule that further amplifies this allocation, and demonstrate that SuRe improves math metrics on Qwen3 student models without harming out‑of‑domain performance.
arXiv:2608. 09826v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards yields no group-relative signal when rollout groups are uniformly correct or uniformly wrong, which account for 63.
arXiv:2609.39405v1 Announce Type: new Abstract: Task vectors provide a simple mechanism for composing learned capabilities through model merging. However, the composability of task vectors produced b...
arXiv:2606. 22600v2 Announce Type: replace-cross Abstract: On-Policy Distillation (OPD) improves the learning efficiency of standard reinforcement learning through dense, token-level supervision from teachers.
arXiv:2606. 12171v1 Announce Type: cross Abstract: Knowledge Distillation (KD) and mixup have proven effective at inducing smoothness in class boundaries; KD captures inherent class relationships in probability distributions, and mixup enforces them through convex combinations of inputs.
arXiv:2605.10889v2 Announce Type: replace-cross Abstract: On-policy distillation offers dense, per-token supervision for training reasoning models; however, it remains unclear under which conditions...
The paper introduces Group‑Calibrated On‑Policy Distillation (GC‑OPD), a method that aligns token‑level teacher guidance with trajectory‑level verifier rewards for long‑context reasoning tasks. GC‑OPD normalizes rewards within rollout groups, uses the signed teacher‑verifier disagreement as a residual, and distributes this residual across tokens via Relative‑Advantage‑Based Credit Assignment (RACA). Experiments on five long‑context benchmarks show that GC‑OPD improves Qwen3‑4B and Qwen3‑8B checkpoints from 29.08/35.12 to 40.47/44.65, outperforming vanilla OPD and demonstrating the effectiveness of group‑relative residual calibration.
arXiv:2607. 07050v2 Announce Type: replace-cross Abstract: Agentic language models must learn when to call tools, when to consume tool responses, and when to answer directly.
arXiv:2608. 03632v1 Announce Type: new Abstract: On-Policy distillation (OPD) transfers teacher capabilities by supervising student-sampled trajectories with dense token-level teacher signals.
arXiv:2607. 07050v1 Announce Type: cross Abstract: Agentic language models must learn when to call tools, when to consume tool responses, and when to answer directly.
arXiv:2609.38342v1 Announce Type: new Abstract: On-policy self-distillation uses a model as its own teacher to provide dense supervision for reasoning, often through reference-solution conditioning....