共计 234 篇文章
2025
MokA：Multimodal Low-Rank Adaptation for MLLMs
Mitigating Hallucinations in Large Vision-Language Models by Self-Injecting Hallucinations
Nullu：通过 HalluSpace 投影减轻大型视觉-语言模型中的对象幻觉
Vision Transformers Don't Need Trained Registers
LLM中MOE的安全行为
DoLa：通过对比层解码提高大型语言模型的事实性
PerturboLLaVA：通过扰动视觉训练减少多模态幻觉
对比解码之VCD
ALPHAEDIT：NULL-SPACE CONSTRAINED KNOWLEDGE EDITING FOR LANGUAGE MODELS
kNN-LMs：一种RAG和LLM前的记忆挂靠方法