Text-derived steering vectors, especially mean shift, improve spatial relation and counting accuracy in multimodal LLMs by up to 7.3% on CV-Bench and by larger margins on several out-of-distribution datasets.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models
Text-derived steering vectors, especially mean shift, improve spatial relation and counting accuracy in multimodal LLMs by up to 7.3% on CV-Bench and by larger margins on several out-of-distribution datasets.