研究库 论文知识库

内容库 / 论文

arXiv:2609.38721 · 多模态

UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement

UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement

类型：arxiv
标识：2609.38721
链接：https://arxiv.org/abs/2609.38721
主分类：multimodal
形态：method
TLDR：Modern multimodal models bring generation and understanding into a single unified system, which enables them to provide and learn from their own feedback. Motivated by this unified capacity, we introduce UniEvo-VL, a self-evolving framework for multimodal models to learn from this constructive self-correction feedback during test-time compute. Instead of relying on a separate, often larger, teacher, we leverage their self-critiques as privileged information and ask a single multimodal model to act as both teacher and student with different contexts. The student only sees the vanilla question,
副分类：engineering
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-10-01-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

攻略 nick7nlp/Awesome-LLM-On-Policy-Distillation · 上手攻略

仓库 nick7nlp/Awesome-LLM-On-Policy-Distillation

文档 Stephen 自测 · R67 · 2026-09-06

笔记 SPEAR: 训练-free 符号化过程奖励用于 RL 知识蒸馏 — 精读与批判

笔记 flyP 精读与批判 · 2026-07-28 15:50

文档 Tom 评 flyP 的产出：On-Policy Distillation 三大要点解读（2607.13399）

arXiv

2609.38721

TLDR

Modern multimodal models bring generation and understanding into a single unified system, which enables them to provide and learn from their own feedback. Motivated by this unified capacity, we introduce UniEvo-VL, a self-evolving framework for multimodal models to learn from this constructive self-correction feedback during test-time compute. Instead of relying on a separate, often larger, teacher, we leverage their self-critiques as privileged information and ask a single multimodal model to act as both teacher and student with different contexts. The student only sees the vanilla question,

相关解读

arXiv ↗ 中文全文