The paper introduces a minimal‑edit on‑policy distillation method for vision‑language models to correct misread diagrams in visual mathematical reasoning. By using reference figure descriptions and verified solutions only during training, the teacher evaluates student rollouts and penalizes only the tokens that deviate from correct reasoning, while correct rollouts receive reinforcement learning rewards. The approach eliminates the need for figure descriptions at test time, improves low‑level geometric perception, and transfers well to external math and perception benchmarks.
arXiv:2610.14939v1 Announce Type: new Abstract: GRPO-style reinforcement learning has advanced vision-language reasoning, but its sequence-level rewards provide limited credit assignment for individu...
Privileged on-policy distillation improves multimodal reasoning by allowing a teacher to evaluate student trajectories using rich, training-only visual evidence. Both models score these trajectories w...
OPD‑Aha is a privileged on‑policy distillation method that improves multimodal reasoning by reconstructing the distillation target from the teacher’s isolated visual preference instead of relying on fragile teacher‑student discrepancies. It suppresses continuations that contradict the image, encouraging students to interrupt flawed reasoning with reflection tokens such as "wait" and "actually." This approach leads to consistent improvements across fine‑grained perception and complex multimodal reasoning benchmarks.
arXiv:2609.36838v1 Announce Type: cross Abstract: Visual agents solve problems by interleaving reasoning with image operations, and on-policy distillation (OPD) provides guidance from a strong teache...
arXiv:2608. 01263v1 Announce Type: new Abstract: On-policy distillation (OPD) samples trajectories from the current student policy and minimizes token-level divergence between student and teacher next-token distributions at prefixes along those trajectories.
arXiv:2606. 19120v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) trains a model on its own rollouts and uses a frozen copy to provide dense token-level targets conditioned on a reference target.
arXiv:2609.39120v1 Announce Type: new Abstract: On-policy distillation (OPD) improves reasoning by providing token-level supervision from a teacher on a student's own trajectories. Existing methods p...
arXiv:2606. 05718v1 Announce Type: cross Abstract: On-policy distillation (OPD) improves reasoning by training a student on trajectories sampled from its own policy under supervision from a teacher.
CA-OPD is a confidence‑aware on‑policy distillation framework that improves structured visual prediction by using teacher confidence to selectively correct unreliable student transitions and gradually transfer rollout control to the student. The method aligns supervision with intervention decisions, providing direct cross‑entropy loss for corrected tokens and full predictive distribution for retained tokens. In a multi‑teacher setting for GUI grounding and OCR, CA‑OPD significantly outperforms the Qwen3.5‑0.8B baseline, achieving large gains on benchmarks such as ScreenSpot‑Pro and OCRBench‑v2 English.
arXiv:2605. 18740v4 Announce Type: replace-cross Abstract: Multimodal Large Language Models (MLLMs) still struggle with fine-grained visual understanding, where answers often depend on small but decisive evidence in the full image.
arXiv:2607. 23125v1 Announce Type: new Abstract: Post-training enables vision-language models (VLMs) to understand human instructions and perform various downstream tasks.
arXiv:2609.10321v1 Announce Type: new Abstract: Knowledge distillation offers an efficient route to transfer a task-adapted vision-language teacher to a compact student. The training target in curren...