Anonymousblind/

svs-subspace-validity-suite

Diagnostic toolkit for validating "visual directions" in Vision-Language Models. Paper: "What PCA-Based Visual Directions in VLMs Actually Capture" (WACV 2027)

git clone https://huggingface.co/datasets/Anonymousblind/svs-subspace-validity-suite cd svs-subspace-validity-suite pip install .

from svs import SubspaceValiditySuite

svs = SubspaceValiditySuite() report = svs.full_report(… See the full description on the dataset page: https://huggingface.co/datasets/Anonymousblind/svs-subspace-validity-suite.