Add hook functions to the ViT

soniajoseph / ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision Transformers (ViTs).

Other

165 stars 18 forks source link

Closed soniajoseph closed 11 months ago

soniajoseph commented 11 months ago

Add hook functions so you can get all intermediate activations of an image with one forward inference pass (probs return activations in a dictionary?)
Add function so you can get intermediate activations of multiple images