Negative Self-Distillation (NSD) is a new framework for improving large language models by encouraging them to diverge from their own flawed reasoning rather than imitate privileged solutions. Unlike On-Policy Self-Distillation, which can suppress uncertainty and exploratory behavior, NSD generates a question‑specific negative condition (e.g., a careless reasoner) and uses a dynamic gating mechanism to target only reasoning‑critical tokens for penalization. This approach preserves foundational language capabilities while consistently outperforming OPSD and other label‑free self‑bootstrapping reinforcement learning baselines.
By Rongcan Pei, Zhepei Wei, Shuyao Xu, Xinyu Zhu, Wei-Lin Chen, Yu Meng
arXiv:2606. 19327v1 Announce Type: new Abstract: Post-training of reasoning language models is commonly driven by supervised distillation and reinforcement learning with verifiable rewards.
By Siyi Gu, Jialin Chen, Sophia Zhou, Arman Cohan, Rex Ying
Post-training of reasoning language models is commonly driven by supervised distillation and reinforcement learning with verifiable rewards. Distillation often relies on chain-of-thought annotations that are expensive to obtain and may themselves be noisy, incomplete, or partially incorrect; even when the final solution is correct, an imperfect rationale can interfere with learning.
arXiv:2605. 12400v2 Announce Type: replace-cross Abstract: We study on-policy self-distillation (OPSD), where a language model improves its reasoning ability by distilling privileged teacher distributions along its own on-policy trajectories.
By Yuxiao Yang, Xiaoyun Wang, Weitong Zhang
arXiv:2608. 04794v1 Announce Type: new Abstract: Self-distillation (SD) has emerged as a compute-efficient alternative to reinforcement learning with verifiable rewards: a self-teacher, conditioned on privileged information (PI) about the answer such as a reference solution, supplies dense per-token supervision to a student that never sees it.
By Sarthak Harne, Chinmay Karkar, Yash Pandya, Ahmed Awadallah, Akshay Nambi
arXiv:2604. 27960v2 Announce Type: replace Abstract: Recent large language models (LLMs) have achieved impressive reasoning milestones but continue to struggle with high computational costs, logical inconsistencies, and sharp performance degradation on high-complexity problems.
By Adam Ishay, Joohyung Lee
The paper introduces Echo-GRPO, a method that rewrites privileged reasoning traces into a model’s own idiolect to align off‑policy supervision with the student policy’s vocabulary. By preserving semantics through Dual‑Reference Decoding, Echo‑GRPO mitigates gradient clipping on critical reasoning tokens and improves reasoning distillation. The approach is instantiated as VideoEcho‑R1 for video reasoning, yielding consistent gains across multiple multimodal LLM backbones and benchmarks, and it can be applied as a plug‑in to both RL and supervised fine‑tuning frameworks.
By Ji Soo Lee, Jinyoung Park, Seohyun Lee, Jongha Kim, Joonmyung Choi, Jinsung Yoon, Hyunwoo J. Kim
EOPSA (Efficient On-Policy Self-Distilled Safety Alignment) addresses inefficiencies in On-Policy Self-Distillation (OPSD) for safety alignment by focusing training on safety-critical tokens. It introduces Adaptive Rollout Scheduling, which limits generation length based on a Teacher Rescue Rate metric, and Selective Distillation, which filters out safety-neutral tokens to concentrate gradient updates on safety-pivotal transitions. Experiments on models up to 32B parameters show that EOPSA reduces rollout computation by about 50% and backpropagates through only roughly 2% of tokens, outperforming full-token distillation baselines in safety compliance and reasoning retention.
The paper investigates on‑policy self‑distillation (OPSD) as a method to enhance reasoning in language models, focusing on mathematical reasoning across models from 0.6B to 8B parameters. Through controlled experiments and token‑level analysis, the authors find that OPSD’s effectiveness depends on alignment between the teacher’s reasoning mode and the full teacher prefix, rather than on privileged semantics alone. They observe that OPSD only improves reasoning in limited compatibility regimes, while often causing length growth, degradation, or behavioral collapse, and that the teacher’s signal is unstable and not predictive of downstream performance.
By Yang Li, Gongle Xue, Yuheng Yuan, Yijia Guo, Shizhe Zhang, Liwen Hu, Lei Ma
The paper introduces SOLID, a framework that enables operations research language models to self-improve without relying on verified answers or external evaluators. SOLID uses solver-generated artifacts from the model’s own rollouts to create pseudo-references, clustering objectives and applying group-relative advantages for dense self-supervision. Experiments on multiple OR benchmarks show that SOLID enhances solution accuracy for both general-purpose and OR-tuned models compared to outcome-only training.
By Rui Zhu, Minglong Cao, Chenyu Zhou, Jianghao Lin, Dongdong Ge
Reflective Recovery is a self‑supervised method that turns failed reasoning attempts into training data, enabling large language models to learn how to correct mistakes during inference. By extracting initial segments of erroneous trajectories and using them as prompts, the approach teaches models to recognize and recover from errors without external critics. Experiments show significant accuracy gains on benchmarks such as AIME 2025 and Minerva, and the method overcomes the scaling collapse problem, fostering emergent self‑correction behaviors.
By Qirui Chen, Renjie Pi, Jiahui Gao, Lingpeng Kong
Cliff is a reward‑shaping method for reinforcement learning with verifiable rewards that identifies the first mistake in a language model’s reasoning process using an off‑the‑shelf teacher. It splits each rollout into a correct prefix and an incorrect suffix, assigning positive token‑level advantages to the prefix and negative feedback to the suffix. Across 12 scenarios, Cliff improves reasoning performance, outperforming on‑policy distillation by 15% and standard GRPO by 7%, even with modest teachers.