无监督在线策略自蒸馏：仅靠内部一致性提升大模型推理

AI 摘要

该研究提出无监督在线策略自蒸馏方法（U-OPSD），仅利用模型自身生成和内部一致性（多数投票）构建伪解决方案，无需外部监督即可纠正模型自信的错误。在多个数学推理基准上，U-OPSD 在 Qwen3 非思考模式下相比基础模型提升 8.5% 和 10.7%，并优于或持平于有监督方法 OPSD 和 GRPO。代码已开源。

正文节选

On-Policy Self-Distillation without Any Supervision Abstract Unsupervised on-policy self-distillation improves large language models by using internal consistency and majority-vote pseudo-solutions to correct confident errors without external supervision. On-policy (Self-)Distillation (OPD / OPSD) has shown strong potential for post-training large language models (LLMs). However, existing methods still rely heavily on external supervision, including ground-truth signals, environmental feedback,

发布时间：—

抓取时间：2026-08-12 05:42

来源机构：Hugging Face