VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set

Devs

VL-SAE: Interpreting and Enhancing Vision-Language Alignment with a Unified Concept Set | Read Paper on Bytez