Encoded but Disconnected: Decomposing Vision-Language Model Failures under a Patching Null
Across three vision-language model architectures (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL3-8B), we report a universal negative finding for mid-layer interpretability.
By Genpei ZhangarXiv - cs.CV (CV)
Read the full story
Originally published by arXiv - cs.CV (CV) on Oct 2, 2026.
Comments
Sign in to join the discussion.