Search papers, labs, and topics across Lattice.
This paper introduces OmniOPD, a novel framework for On-Policy Distillation that eliminates the need for direct access to teacher token-level logits by utilizing a logit-free, chunk-level supervision signal. By employing Monte Carlo rollouts to approximate local preferences and a peak-entropy scheduler for focused auditing, OmniOPD significantly reduces the noise and brittleness associated with traditional token-level logit matching. The method outperforms standard OPD by up to +28.64% on math tasks and achieves further improvements when paired with advanced black-box teachers, demonstrating its effectiveness in extracting reliable learning signals.
Chunk-level semantic verification in OmniOPD yields a +28.64% boost in math performance over traditional OPD, challenging the reliance on token-level logit matching.
On-Policy Distillation (OPD) trains a student model on its own generative trajectories under dense token-level feedback from a stronger teacher, mitigating both the off-policy distribution shift of Supervised Fine-Tuning (SFT) and the sparse credit assignment of Reinforcement Learning (RL). However, standard OPD faces two coupled limitations. First, it requires direct access to the teacher's token-level logits, excluding a broad class of capable proprietary models from serving as teachers. Second, the token-level logit signal itself is brittle, depending on a narrow overlap of plausible next tokens between teacher and student, and prone to amplifying degenerate patterns such as repetition loops. In this paper, we introduce OmniOPD, a novel framework that addresses both limitations through a logit-free, chunk-level supervision signal. OmniOPD replaces deterministic logit matching with Monte Carlo rollouts that approximate the teacher's local preferences through a continuous semantic similarity metric over multi-token chunks, and concentrates this supervision via a peak-entropy scheduler that audits the student only at its high-uncertainty reasoning forks. A Dirichlet-Multinomial Bayesian prior and a base-model KL anchor further bound the variance of discrete sampling and prevent policy collapse across unaudited tokens. Across competitive benchmarks, OmniOPD surpasses the standard OPD approach by up to +28.64% on math, confirming that chunk-level semantic verification extracts a more reliable learning signal than token-level logit matching, whose high information density is offset by significant noise and brittleness. Furthermore, when paired with stronger black-box teachers such as Claude-4.5-Haiku and Gemini-2.5-Flash, OmniOPD achieves an additional +9.54% relative on math over its open-weight teacher counterpart, advancing the student past the performance of self-exploratory RL.
Meta AI