稳定报道时间 2026-10-02 12:00
用于多模态大语言模型能力故障诊断的因果任务分解
一种新的因果分解框架被引入，用于通过隔离内在缺陷与级联错误来诊断多模态大语言模型（MLLMs）的故障，从而更深入地了解这些模型的性能。
01
证据
- AarXiv cs.CL研究机构2026-10-02 12:00
> Abstract: End-to-end accuracy on compositional tasks records how often MLLMs fail, but cannot distinguish whether a failure reflects an intrinsic deficit in…
- AarXiv cs.CL研究机构2026-10-02 12:00
Abstract: End-to-end accuracy on compositional tasks records how often MLLMs fail, but cannot distinguish whether a failure reflects an intrinsic deficit in th…